Electronic Healthcare

Lecture Notes of the Institute
for Computer Sciences, Social Informatics

and Telecommunications Engineering 91
Editorial Board
Ozgur Akan
Middle East Technical University, Ankara, Turkey
Paolo Bellavista
University of Bologna, Italy
Jiannong Cao
Hong Kong Polytechnic University, Hong Kong
Falko Dressler
University of Erlangen, Germany
Domenico Ferrari
Università Cattolica Piacenza, Italy
Mario Gerla
UCLA, USA
Hisashi Kobayashi
Princeton University, USA
Sergio Palazzo
University of Catania, Italy
Sartaj Sahni
University of Florida, USA
Xuemin (Sherman) Shen
University of Waterloo, Canada
Mircea Stan
University of Virginia, USA
Jia Xiaohua
City University of Hong Kong, Hong Kong
Albert Zomaya
University of Sydney, Australia
Geoffrey Coulson
Lancaster University, UK
Patty Kostkova Martin Szomszor
David Fowler (Eds.)
Electronic
Healthcare
4th International Conference, eHealth 2011
Málaga, Spain, November 21-23, 2011
Revised Selected Papers
13
Volume Editors
Patty Kostkova
Martin Szomszor
David Fowler
City University London, School of Health Sciences
City eHealth Research Centre
Northampton Square, London EC1V 0HB, UK
E-mail: patty@soi.city.ac.uk; {martin.szomszor.1, david.fowler.1}@city.ac.uk
ISSN 1867-8211 e-ISSN 1867-822X

ISBN 978-3-642-29261-3 e-ISBN 978-3-642-29262-0
DOI 10.1007/978-3-642-29262-0
Springer Heidelberg Dordrecht London New York
Library of Congress Control Number: 2012934259
CR Subject Classification (1998): K.4, I.2, J.3-4, H.4, C.2, H.5
© ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2012
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting,
reproduction on microfilms or in any other way, and storage in data banks. Duplication of this publication
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965,
in its current version, and permission for use must always be obtained from Springer. Violations are liable
to prosecution under the German Copyright Law.
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply,
even in the absence of a specific statement, that such names are exempt from the relevant protective laws
and regulations and therefore free for general use.
Typesetting: Camera-ready by author, data conversion by Scientific Publishing Services, Chennai, India
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)
Preface
It is our great pleasure to introduce the special issue of LNICST comprising

the scientific presentations at the 4th ICST International Conference on eHealth
(eHealth 2011) that took place in Málaga, Spain, during November 21–23, 2011.
Following the great successes of eHealth 2008 in London, 2009 in Istanbul,
and 2010 in Casablanca, a specific aim of eHealth 2011 was research into social
networks for epidemic intelligence, a new rapidly emerging field bringing together
data mining, Web intelligence, social networks, and Web 2.0 technologies to assist
public health and field epidemiology.
We selected 20 full and 8 short technical presentations by speakers from five
continents. Revisited papers appear in these proceedings and cover a wide range
of topics, including: social media analysis, personalization and patient support
systems, early warning systems and mobile monitoring, games and learning,
online support for professionals and patients, and agents in eHealth.
As in previous years, we were delighted to welcome two prestigious keynote
speakers. Denis Coulombier, the head of unit for Surveillance and Response
Support at ECDC, discussed epidemic intelligence from a European perspective,
while S. Yunkap Kwankam delivered a talk on realizing the full potential of
information and communication technology on global health.
In addition to our keynote speakers we invited a wide range of speakers from
industry, hospitals, and public health organizations. These were: Johan Muskens
(Philips Research Europe), Caroline Crousillat (Orange Healthcare), Jeremy
Nettle (Oracle Corporation), Corinne Marsolier (Cisco Europe), Frederic Lievens
(Med-e-Tel and ISfTeH), Anouk Berger (WHO), Nuria Oliver (Telefónica Digital),
Josette Najjar-Pellet (Fondation Mérieux), Alberto E. Tozzi (Bambino Gesu Hos-
pital, Rome, Italy), Rossana Alessandrello (Barcelona Digital Technology Centre)
and Ciro Cattuto (ISI Foundation).
In the current economic situation it is essential to understand better the finan-
cial landscape supporting eHealth research, translational research, and health-
care delivery. It has become ever more challenging to secure funding for eHealth
and ICT services to improve national and global health. We created a special
funding session at eHealth 2011 and were delighted to have Jenny Tooth (Angel
Capital Group) as invited speaker to discuss these issues and challenges with the
audience.
In a panel discussion chaired by Connie St. Louis entitled “Panic or Risk?”, Pe-
ter Openshaw from Imperial College and Ben Duncan of ECDC gave their views
and answered audience questions on what health care agencies and the media
can do to communicate risk without causing panic using the H1N1 outbreak in
2009 as an example. As a novel (and successful) experiment, both panellists took
part by video conference, from the UK and Sweden, respectively.
VI Preface
Ed de Quincey and Gawesh Jawaheer chaired a joint posters and demos

session at eHealth 2011, which was very well attended and provoked lively dis-
cussions. Twelve posters and eight demos were accepted for presentation. This
year the session was “gamified” to encourage participation - in addition to awards
for best poster and best demo, we also gave a prize for “best participant.”
Organized and chaired by Tim Meldrum on the first evening of the conference,
the Speed Networking Session was an enjoyable and informal event to bring
together medical practitioners, industry delegates, and eHealth researchers in a
unique forum to explore the cutting edge of eHealth knowledge, technology, and
practice.
This year’s edition of the conference made more use of social media than ever
before. In addition to a conference Twitter account (@ehealthConf, with almost
600 followers at the time of writing) and hashtag (#ehealth2011), we had a Face-
book page (www.facebook.com/eHealthConf, with 79 “fans”) and a Flickr poster
gallery (www.flickr.com/photos/eddequincey/galleries/72157628028815537/, 374
views at the time of writing). We covered the conference using a live blog writ-
ten by Gozde Zorlu, with descriptions of the presentations and other sessions
(http://electronic-health.org/). The blog had over 400 readers, and attracted
375 comments and questions during the conference.
We would like to thank all those who helped make eHealth 2011 such a success
- including the authors of all submitted papers, demos and posters; the speakers;
keynote and invited speakers; the panellists; the members of the Program and Or-
ganizing Committees; the Session Chairs; and the conference participants (both
in person and online). Special thanks go to the Local Chair, Cristina Urdiales,
together with student volunteers and local organizers. Finally, we thank ICST,
CREATE-NET, and IEEE EMB for sponsoring the conference, and Springer for
publishing this LNICST book.
Patty Kostkova
Martin Szomszor
David Fowler
Organization
General Co-chairs
Patty Kostkova City eHealth Research Centre, City University
London, UK
Martin Szomszor City eHealth Research Centre, City University
London, UK
Clinical Chair
Femida Gwadry-Sridhar Lawson Health Research Institute, Canada
Industry Chair
Corinne Marsolier Cisco Systems Inc.
Knowledge Transfer Chair

Tim Meldrum Enterprise Office, City University London, UK
Scientific Media Chair

Connie St Louis Graduate School of Journalism, City University
London, UK
Social Media Chair

Marcel Winandy Ruhr University Bochum, Germany
Publications Chair
David Fowler City eHealth Research Centre, City University
London, UK
Poster Chair
Ed de Quincey University of Greenwich, UK
VIII Organization
Demo Chair
Gawesh Jawaheer City eHealth Research Centre, City University
London, UK
Web Chair
Simon Hammond City eHealth Research Centre, City University
London, UK
Local Chair
Cristina Urdialis University of Malaga, Spain
IEEE EMB Representative

Laura Roa Romero University of Seville, Spain
Conference Coordinator
Justina Senkus European Alliance for Innovation
Senior Program Committee

Patty Kostkova City eHealth Research Centre, City University,
London, UK
Martin Szomszor City eHealth Research Centre, City University,
London, UK
Toni Moreno Universitat Rovira i Virgili, Tarragona, Spain
Floriana Grasso University of Liverpool, UK
Femida Gwadry-Sridhar Lawson Health Research Institute, Canada
Ulises Cortez Technical University of Catalonia, Spain
Rob Procter University of Manchester, UK
Albert Burger Heriot-Watt University, UK
John Powell University of Warwick, UK
Erik van der Goot Joint Research Centre, Ispra, Italy
Cecile Paris CSIRO ICT Centre, New South Wales,
Australia
Robert Stevens University of Manchester, UK
Ciro Cattuto ISI Foundation, Turin, Italy
Laurent Billonnet Université de Limoges, France
Organization IX
Technical Program Committee

Jan Aidemark Linnéuniversitet, Sweden
Pedro Alexandre
Da Costa Sousa UNINOVA, Portugal
Dimitra Alexopoulou The Biotechnology Center, Technische
Universität Dresden, Germany
Margarita Anastassova CREATE-NET, Trento, Italy
Bill Andreopoulos The Biotechnology Center, Technische
Universität Dresden, Germany
Roberta Annicchiarico IRCCS Fondazione Santa Lucia, Italy
Raphael Bahati The University of Western Ontario, Canada
Martin Beer Sheffield Hallam University, Sheffield, UK
Ayse Bener Bogazici University, Istanbul, Turkey
Elizabeth Borycki University of Victoria, Canada
Miroslav Bursa Czech Technical University, Czech Republic
Tito Castillo University College London, UK
Olivier Corby INRIA, France
Joanne Curry University of Western Sydney, Australia
Foad Dabiri University of California Los Angeles, USA
Ed De Quincey University of Greenwich, London, UK
Danilo De Rossi University of Pisa, Italy
Gayo Diallo University of Bordeaux, France
Mark Donnelly University of Ulster, Belfast, UK
Mariusz Duplaga Jagiellonian University Medical College, Poland
Jonathan Elford City University, London, UK
Henrik Eriksson Linköping University, Sweden
Elisabetta Farella University of Bologna, Italy
Anthony Fleury École des Mines de Douai, France
David Fowler City University, London, UK
Monique Frize Carleton University, Ottawa, Canada
Leo Galway University of Ulster, Belfast, UK
Xiaohong Gao Middlesex University, London, UK
George Ghinea Brunel University, Uxbridge, UK
Amir Hajjam-el-hassani University of Technology of
Belfort-Montbéliard, France
Simon Hammond City University, London, UK
Ali Hamou The University of Western Ontario, Canada
David Hansen eHealth Research Centre Australia
Mark Hawker University of Leeds, Leeds, UK
Anja Henner Oulu University of Applied Sciences, Finland
Hermie Hermens University of Twente, The Netherlands
Jesse Hoey University of Dundee, UK
Alexander Hörbst University for Health Sciences, Medical
Informatics and Technology, Austria
Roberto Hornero University of Valladolid, Spain
X Organization
David Isern Universitat Rovira i Virgili, Tarragona, Spain

Gawesh Jawaheer City University, London, UK
Malina Jordanova Solar-Terrestrial Influences Institute, Bulgaria
Simon Jupp University of Manchester, UK
Anastasia Kastania Athens University of Economics and Business,
Greece
Sabine Koch Karolinska Institutet, Stockholm, Sweden
Waldemar W. Koczkodaj Laurentian University, Ontario, Canada
Peter Kokol University of Maribor, Slovenia
Jorma Komulainen Ylilääkäri National Institute for Health and Welfare,
Finland
Patty Kostkova City University, London, UK
Vaclav Kremen Czech Technical University, Czech Republic
Elaine Lawrence University of Technology, Sydney, Australia
Athina Lazakidou University of the Peloponnese, Greece
Lenka Lhotska Czech Technical University, Czech Republic
Panos Liatsis City University, London, UK
Magı́ Lluch-Ariet MicroArt, Barcelona, Spain
Paul Marshall University of Warwick, UK
Maria G. Martini Kingston University, London, UK
Kenneth Mcleod Heriot-Watt University, Edinburgh, UK
Andrew Miller University of Wollongong, NSW, Australia
Maria Mirto University of Lecce, Italy
Fleur Mougin Université Bordeaux, France
Bruce Moulton University of Technology Sydney, Australia
Farid Naı̈t-abdesselam University of Paris Descartes, France
Sonja O’Neill University of Ulster, Belfast, UK
Silvia Olabarriaga Universiteit van Amsterdam, The Netherlands
Venet Osmani CREATE-NET, Ireland
Julian Padget University of Bath, UK
Daniela Paolotti Institute for Scientific Interchange Foundation,
Italy
Dorina Petriu Carleton University, Ottawa, Canada
Joao Pimentao UNINOVA, Portugal
Edwige Pissaloux Institut des Systèmes Intelligents et de
Robotique, Université Pierre et Marie Curie,
Paris, France
George Polyzos Athens University of Economics and Business,
Greece
Alain Pruski Lasc Université de Metz, France
Dietrich Rebholz-Schuhmann European Bioinformatics Institute, Cambridge,
UK
Blaine Reeder University of Washington, Seattle, USA
Organization XI
Pekka Ruotsalainen National Institute of Health and Welfare,

Finland
Sandro Scattareggia Marchese Signo Motus Srl, Messina, Italy
Andreas Schrader University of Lübeck, Germany
Michael Schumacher The Institute of Business Information Systems,
Switzerland
Tacha Serif Yeditepe University, Istanbul, Turkey
Gregor Stiglic University of Maribor, Slovenia
Kåre Synnes Luleå University of Technology, Sweden
Martin Szomszor City University, London, UK
Adel Taweel Kings College, London, UK
Manolis Tsiknakis Centre for eHealth Technologies, Heraklion,
Crete, Greece
Alexey Tsymbal Siemans AG, Germany
Asli Uyar Bogazici University, Turkey
Aı̈da Valls Universitat Rovira i Virgili, Tarragona, Spain
Bert-jan van Beijnum University of Twente, The Netherlands
Erik Van Der Goot Joint Research Centre, Ispra, Italy
Mateja Verlic University of Maribor, Slovenia
Nadine Vigouroux IRIT, France
Vivian Vimarlund Linköping University and Jönköping
International, Sweden
Peter Weller City University, London, UK
Tatjana Welzer University of Maribor, Slovenia
Jana Zvarova EuroMISE Center, UK and AV ČR, Prague,
Czech Republic
Table of Contents
Session 1: Social Media Analysis

Social Media: A Systematic Review to Understand the Evidence and
Application in Infodemiology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Stacey Guy, Alexandria Ratzki-Leewing, Raphael Bahati, and
Femida Gwadry-Sridhar
An Examination of the Behaviour of Young and Older Users

of Facebook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Darren Quinn, Liming Chen, and Maurice Mulvenna
Session 2: Knowledge Integration and EPR

An Agile Methodology for IHE-Enabled Deployments . . . . . . . . . . . . . . . . . 17
Bruno Alves and Michael Schumacher
Obstetric Medical Record Processing and Information Retrieval . . . . . . . . 26

Miroslav Bursa, Lenka Lhotska, Vaclav Chudacek, Michal Huptych,
Jiri Spilka, Petr Janku, and Martin Huser
Session 3: Personalisation and Patient Support

Systems
Health Care Performance Monitoring Using an Inpatient Reference
Process Model and Clinical KPIs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Eva Gattnar, Okan Ekinci, and Vesselin Detschew
A Model for a Motivational System Grounded on Value Based Abstract

Argumentation Frameworks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Eugenio Di Tullio and Floriana Grasso
An Assessment of the Potential for Personalization in Patient Decision

Aids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Øystein Eiring and Laura Slaughter
Session 4: Early Warning Systems and Mobile

Monitoring
An Analysis of Twitter Messages in the 2011 Tohoku Earthquake . . . . . . 58
Son Doan, Bao-Khanh Ho Vo, and Nigel Collier
XIV Table of Contents
Web-Based Stress Management System Goes Mobile:

Considerations of the Design of a Mobile Phone App . . . . . . . . . . . . . . . . . 67
Åsa Smedberg and Hélène Sandmark
Use of SMS for Tsunami Early Warnings at a Table Top Exercise . . . . . . 76

Vincenzo Lagani, Vasilis Kontogiannis, Panagiotis Argyropaidas, and
Catherine Chronaki
Session 5: Games and Learning

Seamless Evaluation of Interactive Digital Storytelling Games:
Edugames4All . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Patty Kostkova
Patient-Centered Care as a Learning Process . . . . . . . . . . . . . . . . . . . . . . . . 85

Jan Aidemark and Linda Askenäs
Wireless Computer Games and Applications in the Medical Education

Curriculum: Adventures in Pedagogy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Juanita Fernando and Nicolette Peters
Session 6: Security, Privacy and Prevention

The Process of Policy Authoring of Patient-Controlled Privacy
Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Thomas Trojer, Basel Katt, Thomas Schabetsberger,
Richard Mair, and Ruth Breu
Modelling a User Authorisation and Data Access Framework

for Multi-specialty Research Systems in Secondary Health Care . . . . . . . . 105
Ire Ogunsina, Sarah N. Lim Choi Keung, Lei Zhao, Gavin Langford,
Edward Tyler, and Theodoros N. Arvanitis
Software Engineering-Inspired Approach to Prevention Healthcare . . . . . . 109

Victor Rentea, Andrei Vasilateanu, Radu Ioanitescu, and
Luca Dan Serbanati
Session 7: Online Support for Professionals and

Patients
Evaluation of a Web-Based Patient Portal for Chronic Disease
Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Stacey Guy, Alexandria Ratzki-Leewing, and Femida Gwadry-Sridhar
Cardio Online Reader/COR: A Web 2.0-Based Tool Aimed at Clinical

Decision-Making Support in Cardiology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Vendula Papı́ková and Miroslav Zvolský
Table of Contents XV
Session 8: Agents in eHealth

An Agent Based Pervasive Healthcare System: A First Scalability
Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
Johannes Krampf, Stefano Bromuri, Michael Schumacher, and
Juan Ruiz
An Agent-Based Approach to Real-Time Patient Identification for
Clinical Trials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Gareth Tyson, Adel Taweel, Simon Miles, Michael Luck,
Tjeerd Van Staa, and Brendan Delaney
Session 9: Online Communities of Practice

The Use of Social Bookmarking by Health Care Students to Create
Communities of Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Ed de Quincey, Avril Hocking, Josephine O’Gorman,
Simon Walker, and Liz Bacon
Engagement in Online Medical Communities of Practice in Healthcare:
Analysis of Messages and Social Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
David Fowler, Martin Szomszor, Simon Hammond,
John Lawrenson, and Patty Kostkova
Towards Delivering Disease Support Processes for Patient
Empowerment Using Mobile Virtual Communities . . . . . . . . . . . . . . . . . . . . 158
Bert-Jan van Beijnum, Pravin Pawar, Lamia Elloumi, and
Hermie Hermens
Session 10: eHealth Solutions

E-Health Readiness Assessment for E-Health Framework for Africa:
A Case Study of Hospitals in South Africa . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Alfred Coleman, Marlien E. Herselman, and Dalenca Potass
The Physical Activity Loyalty Card Scheme: Development and
Application of a Novel System for Incentivizing Behaviour Change . . . . . 170
Ruth F. Hunter, Michael Davis, Mark A. Tully, and Frank Kee
Session 11: Social Media Surveillance

Tracking Media Reports on the Shiga Toxin-Producing Escherichia coli
O104: H4 Outbreak in Germany . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
Jens P. Linge, Jas Mantero, Flavio Fuart, Jenya Belyaeva,
Martin Atkinson, and Erik van der Goot
XVI Table of Contents
Syndromic Classification of Twitter Messages . . . . . . . . . . . . . . . . . . . . . . . . 186

Nigel Collier and Son Doan
Towards Spatial Description-Based Integration of Biomedical Atlases . . . 196

Nurzi Juana Mohd Zaizi and Albert Burger
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

Social Media: A Systematic Review to Understand
the Evidence and Application in Infodemiology
Stacey Guy, Alexandria Ratzki-Leewing, Raphael Bahati,

and Femida Gwadry-Sridhar
Lawson Health Research Institute,

Commissioners Rd E. 801, N6C 5J1 London, Canada
stacey.guy@sjhc.london.on.ca
{alexandria.ratzkileewing,raphael.bahati}@lawsonresearch.com
femida.gwadry-sridhar@lhsc.on.ca
Abstract. Social media represents a new frontier in disease surveillance.

Infoveillance allows for the real-time retrieval of internet data. Our objective
was to systematically review the literature utilizing social media as a source for
disease prediction and surveillance. A review of English-language conference
proceedings and journal articles from 1999 to 2011 using EMBASE and
PubMed was conducted. A total of 12 full-text articles were included. Results
of these studies show the use of open-source micro-blogging sites to inform
influenza-like-illness monitoring. These results inform recommendations for
future research directions.
Keywords: social media, review, population surveillance, data mining.
1 Introduction
Today, the number of social media users continues to skyrocket with rates of
participation on social networking sites already quadrupling from 2005-2009 [1]. As
an easily accessible, highly cost-effective and interoperable system, social media
opens doors to a better understanding of community creation, providing fast access to
information anywhere in the world, 24 hours a day [2,3].
Due to the popularity of online communication, open-source social media
platforms present excellent opportunities in health research [4]. Using a strategy
called infoveillance, real-time online data can be systematically mined, aggregated
and analyzed to inform public health and policy [5,6]. More specifically, social
media can be used as a relevant and real-time source of epidemic intelligence [4].
Mining online information can provide insight to abnormal patterns of disease and
aid in predicting disease outbreaks. Various studies have confirmed the potential of
infoveillance to advance epidemic intelligence.
The aim of this paper is to illustrate how data generated through social media can
be used to inform planning and implementation of strategies to address communicable
disease emergence - in turn, changing the future of health research.
P. Kostkova, M. Szomszor, and D. Fowler (Eds.): eHealth 2011, LNICST 91, pp. 1–8, 2012.
© Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2012
2 S. Guy et al.
This paper is organized as follows. The methodology used in this systematic

review is described in section 2. The results of the review are presented in section 3.
A discussion of these findings and their implications are debated in section 4.
2 Method
2.1 Data Sources

We conducted a systematic review of the literature utilizing the bibliographic
databases EMBASE and PubMed in June 2011. The following keywords were used to
search EMBASE; keywords were divided into three categories: (1) Disease (Early
detect$, Pandemic$, Epidemic$, Communicable disease$, Early diagnosis), (2)
Medium (Information technol$, Internet, Mass medium, Medical computing, Social
Media, Social network$, Geolocation), and (3) Methodology (Disease surveillance,
Monitor$, Disease control, Algorithm$, Data min$, Query process$, Information
retriev$). MeSH terms used to conduct the search in PubMed were also divided into
three categories: (1) Disease (Pandemics, Communicable diseases, Disease
outbreaks, Early diagnosis), (2) Medium (Communications media, Databases, factual,
Internet*, User-computer interface*), and (3) Methodology (Population surveillance,
Information storage and retrieval, Forecasting, Data mining, Sentinel surveillance).
The keyword in which the article was indexed, the title of the article, and the abstract
were searched for these categories described above. Unfortunately, terms such as
‘infoveillance’ have not yet been coded in these databases.
Search terms were chosen to reflect our objective – to review published research on
disease surveillance using open-source social media. Conducting a generalized search
in Google, elicited a few relevant publications to inform our search term selection.
Index terms of these publications informed our formalized search strategy.
EMBASE and PubMed were chosen as they limit our search to published material
in the arena of health care as opposed to using a computer-focused bibliographic
engine that would require more terms to narrow the search to health. By including
both databases we were able to search both North American and European published
literature. The search was limited to both journal articles and conference proceedings
– specifically, the terms ‘conference paper’ and ‘proceeding’ was used for EMBASE,
while ‘clinical conference’ was used for PubMed) written in the English-language and
published between 1999 and 2011 (for EMBASE, ‘current’ was chosen; for PubMed
we used 1999/01/01 to 2011/12/01). The time period chosen reflects the addition of
the term ‘internet’ to the PubMed index – no such MeSH term exists for social media
however the term ‘internet’ is broad and encompassing.
2.2 Data Extraction and Synthesis

Using the search terms described above, a total of 384 journal articles and 1
conference proceeding was retrieved form PubMed and a total of 484 journal articles
from EMBASE (with limits applied). In addition, our hand search (informed by the
Social Media: A Systematic Review to Understand the Evidence and Application 3
references within the retrieved publications and Google searching) revealed 16

possible publications. The combined databases of 885 publications were searched for
duplicates which resulted in a total of 287 publications being eliminated. Three of the
authors (R.B, S.G, A.R-L) reviewed the publication abstracts to further establish
relevance. Publications without abstracts were excluded. Publications that mined
RSS feeds, survey data, physician records, medical records and search engines were
excluded. Twelve publications focusing on data mining social media were reviewed.
3 Results
Fig. 1. This flow diagram illustrates the study selection process. A total of 885 publications
were identified by bibliographic and hand searching. Through application of inclusion and
exclusion criteria, 12 full-text publications were included in this review. This diagram is based
upon The PRISMA Statement [7].
3.1 Study Characteristics
All publications reported focused on mining social media for the purpose of disease
surveillance and prediction. As infodemiology is a relatively novel and emerging
field, most studies were exploratory in design. The diseases of choice include
influenza-like-illness (ILI) and H1N1.
3.2 Results of Individual Studies

Social Network Enabled Flu Trends (SNEFT) [8,9] is used to track and predict ILI
activity. Tweets were retrieved (including metadata) with ILI content over approximately
a 12 month period. SNEFT consists of separate data repositories where aggregated
anonymous data is stored, and uses the ARMA model to predict ILI incidence. The web
crawler was developed using Twitter’s Search API. Regressive models were built and
4 S. Guy et al.
evaluated with CDC (Centres for Disease Control) data. The system retrieved 4.7 million
– retweets were extracted. To validate the observed trends, this data was compared to
CDC data. The dataset containing no retweets and no tweets from the same user, resulted
in the highest correlation (r= 0.9846) with CDC data.
Corley et al., [10] evaluated blog posts containing ILI keywords. Flu-related posts
were extracted from 44 million posts collected over a 3 month period. A seven day
period in posting was identified and verified. Categorized baseline trends were
compared to CDC data to identify anomalies. Results show a significant correlation
between the frequency of ILI posts per week and CDC data. In addition, Corley et al
{{4750 Corley 2010; 4028 Corley 2010;}} collected 97,955,349 weblogs, micro-
blogs and social media items pertaining to ILI data over a 20 week period. English
language items containing relevant keywords were retrieved and grouped by month,
week and day of the week. Flu-related data was compared to CDC data, and found to
be highly correlated (r = 0.626 at 95% confidence).
Culotta et al., [4] have developed regression models using 574, 643 tweets
collected over a 10 week period to predict ILI. To obtain a random sample of tweets
the authors searched for common words. The percentage of tweets that reported ILI
was estimated. Their findings indicated multiple regression out performs simple
regression, keywords selected based on residual sum of squares is more effective than
selecting keywords based on a correlation coefficient, and the best model of
prediction (r=0.78) was one where “a simple bag-of-words classifier trained on
roughly 200 documents can effectively filter erroneous document matches”.
de Quincey & Kostkova [13] conducted a study to identify ILI trends present in
tweets. The Twitter Search API was used to retrieve 100 tweets (including metadata).
A PHP code parsed returned tweets (every minute) which were then saved to a
MYSQL database. The system ran for 1 week in May 2009. A total of 135, 438
tweets containing ‘flu’ terminology were retrieved. The content was analyzed for
trends using ‘flu’ in conjunction with other keywords (‘swine’ + ‘flu’). Future plans
involve the use of collocation analysis to identify trends.
Infovigil is an open-source infoveillance system which mines, analyzes and
visually represents textual health-related data from Twitter [6,14]. Infovigil was used
to plot term prevalence, and provide content analysis of tweets pertaining to H1N1.
Two million tweets were retrieved over an 8 month period. English-language tweets
were selected for, and retweets were excluded. Tweet patterns were influenced by
media with the most commonly tweeted material being news (52.6%) Original tweets,
as opposed to retweets, contained more personal experiences. There was very little
misinformation found in the tweets (4.5%). The majority of automated queries
correlated with manual coding results.
Lampos & Cristianini [15] developed a monitoring tool to track ILI patterns using
UK specific Twitter data. Tweets containing symptom-related keywords were
collected over 6 months during 2009. A daily average of 160,000 tweets were
retrieved. This data, converted to a flu-score, was compared to weekly H1N1 reports
from the Health Protection Agency. This resulting score correlated highly with reports
(>95%). This method works independently of language, can determine self-diagnostic
statements in tweets, and uses time series geolocated data.
Data Collector [16] is a system that uses social media as a source for real-time
data. The backend consists of a web crawler, written in PHP and utilizing the Twitter
Search API. Tweets are stored in a relational database according to UML class with 2
main categories: disease and location. Data Collector supplies a RESTful API divided
into location, disease, and occurrence, which specifies methods and parameters
through which one can access the database. The frontend web interface uses AJAX to
produce real- time graphs and maps. A dataset containing H1N1 tweets was collected
from 6 European countries between May and July 2009. This system collected an
average of 3200 tweets per day; 700 pertained to H1N1.
Signorini et al., [17] tracked sentiment and H1N1 activity using tweets. Keywords
were used to retrieve tweets from the US. Prediction models were trained using CDC
ILI values. Results are divided into 2 sets of data. The first saw 951,697 tweets from
334,840,972 retrieved over 34 days and was used to plot spikes of public interest. The
second set represented over 4 million tweets from 8 million over a 3 month period.
When analyzed, this showed no sustained interest in vaccine-related issues by the
public. ILI estimates were gathered using a model trained on 1 million ILI tweets for
8 months. This model produced estimates that the authors believed were fairly
accurate (average error = 0.28%; SD 0.23%). In order to garner real time estimates of
ILI according to region, Signorini et al., [17] developed a model with region readings
fitted to geolocated tweets. This model was less precise (average error = 0.37%) than
the national weekly model.
The majority of articles included in the systematic review collected and examined
data from Twitter (n=7); only 1 article looked at weblogs, micro-blogs, and social
media. Studies demonstrated that a general correlation exists between ILI content in
Tweets and CDC data. Research also indicated that Tweet patterns are strongly
influenced by media, with news being the most commonly tweeted material. Finally,
additional research is needed to determine the effectiveness of geo-location in
garnering real-time estimates of ILI according to region.
4 Discussion
In realizing the potential of infodemiology in healthcare, it is important to consider

the advantages and disadvantages of mining social media. Additionally, researchers
must acknowledge and identify key target audiences; indeed, social media has
specific target audiences with unique engagement behaviour specific to a platform
that may or may not be representative of the population at large. Assessing the uses
and potential of infodemiology in healthcare can improve user interaction,
information access, evidence-based medicine, and knowledge representation.
4.1 Advantages and Disadvantages of Mining Social Media

Infoveillance can provide real-time, immediate and relevant information [6,14]. This
is particularly useful when seeking timely and reliable data on the spread or severity
of influenza [18]. Analyzing and disseminating real-time information can also
6 S. Guy et al.
improve public access to health surveillance information. As data-mining sources

utilize open-source information, the operating costs of these systems can be extremely
low [6,14]. In addition ‘mashups’ are the new multi-taskers, capable of mining,
categorizing, filtering, and visualizing online, real-time data on epidemics [19].
Certain pitfalls to mining social media exist. First, textual data can be difficult to
classify and interpret since harvested data (e.g., a tweet) may not provide enough
information and meaning to facilitate automatic classification [17]. Second, the
collected data may not be representative of the entire population—this challenge is
especially pertinent as social media users are often younger, more educated, and
urban-dwelling with higher incomes [17]. Furthermore, while coding for geographic
origin may resolve certain limitations, not all profile accounts on networking sites
contain geographic information; even so, visible geographic information cannot be
verified for accuracy.
It is thus worthwhile to explore data mining sources that track IP addresses, or
techniques to monitor social media activity on mobile phones [17]. Alternatively
using GPS monitoring – using GIS systems that are either embedded in smart phones
or attached to independent devices can provide supplemental information. One
question that has not been resolved is whether participants need to provide explicit
consent or whether the fact they are using publically available communication tools
renders their information available and subsequently usable.
4.2 Study Limitations
Our search strategy was not as streamlined as we had planned. Not having the ability
to choose a MeSH term as distinct as “social media”, “infoveillance”, or “Twitter”,
meant that time was wasted sifting through irrelevant publications. While we
extended our search to include conference proceedings as well, the publications
retrieved were not of value. Rather, the references in retrieved publications provided
direction to relevant crucial proceedings. However, some of the publications we
retrieved through hand searching were not retrieved through direct bibliographic
database searching.
4.3 Research Directions

In this section, we outline key recommendations that we believe are essential to
forging new opportunities in data mining and collaborative analysis within user-
driven content sharing paradigms. This will enable the full realization of the
significant potential of patient engagement and information sharing and may help
transform healthcare as a whole.
Target Audiences. The initial task for any text mining solutions is identifying entities
of interest from the relevant textual content. This is often achieved through the use of
natural language processing techniques. Each social media has a specific target
audience with unique engagement behavior specific to that platform. Information
harvested through social networking sites may not be representative of the population
at large. It is therefore important that data mining solutions take into account
demographic characteristics of audiences within individual platforms.
User Interactions: Every social networking platform has a set of rules governing
how its users interact with one another. For example, some social media platforms
such as Twitter and Facebook enable real-time interactions between users while
YouTube tends to be less interactive. As such, the types of interactions will
determine when and how often data must be collected in order to derive any
meaningful information.
Information Access: Social media platforms are already compiling fine-grained
user-generated content based on individuals’ online activities. While the means for
deciphering what is relevant through information mining already exists and have
proven extremely successful considering the amount of money companies are willing
to pay to have such kind of access, such personal and information-rich content is not
often publicly available. More open-source social networking solutions are therefore
needed to facilitate any meaningful data mining solutions beyond the basic alert
systems discussed in this paper.
Evidence-Based Medicine: Research is needed to identify effective ways of
embedding evidence within social media platforms that could support monitoring
positive impact on desired behavior changes and allowing users to share/compare
experiences and provide support. Given the broad range of users, there is also a need
to provide levels of detail regarding the evidence itself so that meaning information
could be mined.
Knowledge Representation: Finally, mining social media content for medical
information can only succeed if we recognize the role of ontologies in knowledge
management and knowledge discovery. Ontology offers significant benefit to
knowledge harvesting in social networking platforms as it facilitates data pruning and
can help accelerate the discovery of meaningful information.
Summary. Evolving the use of infodemiology in healthcare will involve the

examination of: (1) user interactions, which may determine the time and frequency of
data collection; (2) information access, which may require the creation of more open-
source social networking sites to facilitate more meaningful data mining solutions; (3)
evidence-based medicine, which will allow users to share and compare experiences
and provide support from within a social media platform; and (4) knowledge
representation, which can allow the mining of social media for medical information,
knowledge management and knowledge discovery. Health and healthcare might soon
be achieved at the click of a mouse.
References
1. Evans, D.: Social Media Marketing: An Hour a Day. Wiley Publishing, Indiana (2008)
2. McNab, C.: What Social Media Offers to Health Professionals and Citizens. Bulletin of the
WHO 87, 566 (2009)
3. Bacigalupe, G.: Is There a Role for Social Technologies in Collaborative Healthcare?
Families, Systems & Health 29, 1–14 (2011)
8 S. Guy et al.
4. Culotta, A.: Towards Detecting Influenza Epidemics by Analyzing Twitter Messages. In:
1st Workshop on Social Media Analytics (SOMA 2010), Washington DC (2010)
5. Wilson, N., Mason, K., Tobias, M., Peacey, M., Huang, Q.S., Baker, M.: Interpreting
Google Flu Trends Data for Pandemic H1N1 Influenza: The New Zealand Experience.
Euro. Surveill. 14, 19386 (2009)
6. Chew, C., Eysenbach, G.: Pandemics in the Age of Twitter: Content Analysis of Tweets
During the 2009 H1N1 Outbreak. PLoS ONE 5 (2010)
7. Moher, D., Liberati, A., Tetzlaff, J., Altman, D.G., PRISMA Group: Preferred Reporting
Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement. Annals of
Internal Medicine 151, 264–269 (2009)
8. Achrekar, H., Avinash, G., Lazarus, R., Yu, S., Liu, B.: Predicting Flu Trends using
Twitter Data. In: International Workshop on CPNS, Shanghai, China (2011)
9. Chen, L., Achrekar, H., Liu, B., Lazarus, R.: Vision: Towards Real Time Epidemic
Vigilance through Online Social Networks: Introducing SNEFT. In: 1st ACM Workshop
on Mobile Cloud Computing, San Francisco, California (2010)
10. Corley, C.D., Mikler, A.R., Singh, K.P., Cook, D.J.: Monitoring Influenza Trends through
Mining Social Media. In: International Conference on Bioinformatics and Computational
Biology, Las Vegas, Nevada (2009)
11. Corley, C.D., Cook, D.J., Mikler, A.R., Singh, K.P.: Text and Structural Data Mining of
Influenza Mentions in Web and Social Media. International Journal of Environmental
Research and Public Health 7, 596–615 (2010)
12. Corley, C.D., Cook, D.J., Mikler, A.R., Singh, K.P.: Using Web and Social Media for
Influenza Surveillance. Advances in Experimental Medicine and Biology 680, 559–564
(2010)
13. de Quincey, E., Kostkova, P.: Early Warning and Outbreak Detection Using Social
Networking Websites: The Potential of Twitter. In: Kostkova, P. (ed.) eHealth 2009.
LNICST, vol. 27, pp. 21–24. Springer, Heidelberg (2010)
14. Eysenbach, G.: Infodemiology and Infoveillance Tracking Online Health Information and
Cyberbehavior for Public Health. American Journal of Preventive Medicine 40, S154–
S158 (2011)
15. Lampos, V., Cristianini, N.: Tracking the Flu Pandemic by Monitoring the Social Web. In:
Int. Workshop on CIP, Elba Island, Italy (2010)
16. Lopes, L.F., Zamite, J.M., Tavares, B.C., Couto, F.M., Silva, F., Silva, M.J.: Automated
Social Network Epidemic Data Collector. In: INForum (2009)
17. Signorini, A., Segre, A.M., Polgreen, P.M.: The Use of Twitter to Track Levels of Disease
Activity and Public Concern in the U.S. during the Influenza A H1N1 Pandemic. PloS
One 6, e19467 (2011)
18. Wilson, K., Brownstein, J.S.: Early Detection of Disease Outbreaks Using the Internet.
CMAJ: Canadian Medical Association Journal 180, 829–831 (2009)
19. Brownstein, J.S., Freifeld, C.C., Madoff, L.C.: Digital Disease Detection–Harnessing the
Web for Public Health Surveillance. The New England Journal of Medicine 360, 2153–
2155, 2157 (2009)
An Examination of the Behaviour of Young
and Older Users of Facebook
Darren Quinn, Liming Chen, and Maurice Mulvenna
School of Computing and Mathematics, University of Ulster,

Jordanstown. BT37 0QB, Northern Ireland
quinn-d15@email.ulster.ac.uk,
{l.chen,md.mulvenna}@ulster.ac.uk
Abstract. Increased use in recent years of social networking applications such

as Facebook has resulted in a rapid growth of user generated content. As a
communication tool for social engagement, the online patterns and
characteristics of individuals and user types can now be better understood by
monitoring user behaviour, activity frequency and modality of use. The aim of
this paper is to present an analysis of the behaviour and characteristics of young
and older user groups in an online social network. Our findings show that
correlations exist between user behaviour and age, based upon activities as
evidenced within the data. Visualisation of full user activity has further helped
our understanding of user behaviour by identifying interaction patterns, which
may be used for the monitoring of social engagement.
Keywords: Interaction Analysis, Social Networking, Social Computing.
1 Introduction
The growth and adoption of Internet technologies has increased markedly over the
past decade. From 2000 to 2011, growth in Internet usage has witnessed an increase
of 353% and as of March 2011, 58% of Europe's 816 million population were Internet
users [1]. As a subsequent development of the Internet, online social networks
emerged through a desire by users to share data, experiences, events and emotions. As
a communication tool they performed a major role in the increased adoption rates of
Internet technologies. As such a diverse range of social networking sites have been
developed, from generic social networks such as Facebook1, to specialist content
sharing sites such as the photo sharing network of flickr2. The impact of social
networking has been acclaimed as rendering a new global footprint [2], due to its
profound social and economic impact, on both business and individual users. With
increased access to web services, and ever more innovative developments for end
users, high profile social networking applications such as Facebook and Twitter3 are
being routinely applied as a core communication approach for users desiring to share
1
www.facebook.com
2
www.flickr.com
3
www.twitter.com
10 D. Quinn, L. Chen, and M. Mulvenna
information on a range of topics. An increased use of social networking sites has led
to a rise in research interest, particularly within the domain of Social Network
Analysis (SNA), a research area with a traditional role of analysis and visualisation of
networks [3], discovering details such as a networks structure, or the strength and
cohesiveness of community structures [4].
More recently within SNA a trend is emerging which is extending traditional
concepts towards a more contemporary approach of Social Network Interaction
Analysis (SNIA). As an example, Facebook is currently the most popular Social
Networking Site with an estimated 700 million unique monthly visitors [5]. The aim
of SNIA is to understand the "individual" and their use of functions in these vast
networks. Through a variety of techniques such as web crawlers or bespoke programs,
SNIA seeks to discover the behavioural patterns of each individual by analysing their
user generated content. Contemporary studies have been interested in aspects such as
user profiles, communication patterns and interactions in online networks [6], [7], [8].
As an approach it not only allows for the analysis of users and groups etc, but may
also be applied to develop a greater understanding of the behavioural patterns and
related characteristics of each user.
1.1 Aim
The aim of our current research is to harness and mine user generated content,
through data collection approaches applied to Facebook, exploring the behaviour and
characteristics of two distinct user groups, young (15-30) and older (50+) users,
comprised of equal numbers of males and females. This study is made up in two
parts; firstly we investigated user activity, defined as the posting of comments or
replies to comments. As a study, its aim is to define the individual interaction metrics
of any given user’s activity. These measures are applied to compare interaction rates
of a user, and across two age categories. Secondly we explore activity visualisation,
to disclose interaction patterns. Illustrating user interaction data may help to
understand previously undisclosed patterns, detailing the: who, what and when of a
user’s social network engagement, illustrating patterns such as network joining or the
activity evolution of a user, defining their high and low periods of user activity.
The remainder of the paper is organised as follows. Section 2 discusses the
concept, exploring in detail the chosen social network structure, defining interactions
metrics and analysis methods applied in the study. Section 3 describes the user data
and the data collection process, with results of activity frequencies and activity
visualisation presented. Section 4 discusses the patterns that can be observed from the
results. Section 5 summarises current work, discussing future work and limitations.
2 Concept and Methodology
The aim of the research is to develop an understanding into the individual activity of
users. The research is aimed at determining individual metrics, and also to establish if
correlations are identifiable between age groups and interaction frequency. Studies are
An Examination of the Behaviour of Young and Older Users of Facebook 11
designed with the primary objectives of: quantifying user activity, visualising user
engagement, identifying behavioural patterns and characteristics of disparate user
groups, based on real world observations in an online social network. Metrics are
subsequently applied to categorise users based on measures of activity, based upon
frequency of posting comments and replies to comments. The primary motivation for
investigating young and older online users is that it helps to establish if different
behaviours (e.g., high or low activity patterns) are present and identifiable. Secondly
the research may help to determine if these behaviours are attributable to the age of
online social networks users. Future investigations will focus on establishing if
indicators in the behaviour of people in online social networks relate to user well
being (e.g., low activity frequency corresponding to low feelings of well being).
2.1 Methodology
"Facebook is a social utility that connects people with friends and others who work,
study and live around them. People use Facebook to keep up with friends, upload an
unlimited number of photos, post links and videos and learn more about the people
they meet" [9]. Facebook was selected to investigate user's online social network
engagement for the following reasons: prominence as the leading social network, user
volume, data accessibility and a range of interaction features. Essentially Facebook
users engage through their ‘Wall’, a facility which controls user content, such as for
the posting of comments. As users contribute to their wall a chronicle of interactions
is amassed. User's security options are self-imposed and dictate the accessibility of
data by other Facebook network users. Privacy settings are set and controlled by the
user, and information can be disclosed to: 'everyone', 'friends of friends' or 'friends
only'. We class information as publically available if a user's privacy settings are set
to everyone. It is only these profiles with which the study was concerned.
As is dependent on the information provided by the user, the age of a user is not
always explicitly given. However, provided certain information is available it then
becomes possible to determine a user's age (within a close proximity), in one or a
combination of three ways, either by: date of birth, school leaving year or university
leaving year. Interactions were defined within the context of this study as being any
measurable user activity occurring on the user wall. A full range of potential
Facebook interactions were assessed for their inclusion within the scope of the current
study. However, the remit of the current study is to assess only the user's personal
contribution to their network, therefore all non-user elements (e.g., non-user
comments) were omitted. Due to the subjective nature of particular wall features our
study concerned itself solely on user comments and user replies as core application
components to provide a measurement of user contribution. It is viewed that later
studies may provide an opportunity to provide a broader scoped analysis. As a
specific example, the following comment scenario is provided; a user posts a
comment and another user posts a reply in response to this comment, both interactions
are recorded. However, only the first interaction will be of interest in the scope of this
study as this is the only interaction attributable to the user in question. Corresponding
metadata of date and time was also captured for each user element.
2.2 User Frequency

User metrics were derived using the following approach. Activity frequency was
calculated as a measure of each individual's activity count, which was then divided
over the length duration of available user data. In reality this means extraction of the
first and last activity dates of any particular user, resulting in a value for the total
number of active days. This value was then divided by the number of activities to
obtain an overall frequency metric for the entire active period. Activity frequency
calculation:
af = n / (x1 – x2) (1)
Where; activity frequency (af), first date stamp (x1), last date stamp (x2), total
number of user activities (n).
3 Data Collection and Trends

This section describes the data collection process, and provides specific details of the
user data obtained for young and older users. The process was repeated in both
instances altering only for the setting of profile ages. A user profile was created and
using the ‘find classmates’ tool, a seeding point of local universities was applied.
Based on the defined profile attributes (age, gender, etc), a listing of potential friends
were identified and returned by Facebook. User suitability for inclusion in the study
was then determined on the two following issues; firstly a user’s wall must be
publically viewable; secondly age of the user must be identifiable from the stated user
profile. Profiles of each suitable user’s 'friends' were then assessed using the same
approach. Process was repeated until a sufficient amount for each user group and
gender was acquired. Data was acquired over two months from December 2010, and
subjected to a three stages of analysis process. (1) User walls were manually extracted
through generated source code, containing all visible wall data. (2) A bespoke parsing
program was developed for the identification and extraction of required user
interaction data. (3) Comparison evaluations were applied for each user group and
gender set. In total 500 full user profiles were obtained (125 young males, 125 young
females, 125 older males and 125 older females). User data consisted of the entire
wall history for each user, with walls exhausted until posts were no longer available.
User age data was determined on a users 'info' page, and extracted in conjunction with
wall data.
3.1 Activity Frequency

Based on initial analysis, user groups were categorised into the following five activity
groupings: 'G1': Zero activity, 'G2': Greater than 0 and less than 5 days, 'G3': Greater
or equal to 5 days and less than 10 days. 'G4': Greater or equal to 10 days and less
than 20 days. 'G5': Greater than 20 days. These classifications are designed to
represent increasing activity ranges, from G1 to G5. The first group 'G1' is the lowest
possible activity value of zero, applied when no user activity was recorded on the user
wall. As detailed the majority of users can then be subsequently categorised within
An Examination of
o the Behaviour of Young and Older Users of Facebook 13
the remaining groupings off G2 to G5. The user comment frequency for both the ollder
and younger users (Fig.1), and also the user reply frequencies for older and younnger
users (Fig.2) have been illlustrated below. Further detailed is the activity frequeency
comparison table of young and older users (Table 1).
Young User Older Users

60%
50%
40%
30%
20%
10%
0%
G1 G2 G3 G4 G5
Fig. 1. User Comment Frequency
Young User Older Users

70%
60%
50%
40%
30%
20%
10%
0%
G1 G2 G3 G4 G5
Fig. 2. User Reply Frequency
Table 1. Activity Frequency Table

Comment Frequency Reply Frequency (%)
(%)
Group Y
Younger Older Younger Olderr
G1: Zero days 1% 52 % 2% 65 %
G2: 0 - 4 days 52 % 10 % 24 % 2%
G3: 5 - 9 days 23 % 6% 21 % 4%
G4: 10 - 19 days 15 % 10 % 21 % 4%
G5: Over 20 days 10 % 21 % 31 % 24 %
3.2 Activity Visualisattion

In order to visualise user acctivity a 'representative user' was selected (Fig.3). Analyysis
of activities for both groupps indicates that younger users generate greater volumes of
14 D. Quinn, L. Chen, and
a M. Mulvenna
nger cohort was chosen. Of the active groups, statistically

activity, therefore the youn
'G2' (Frequency values 0-5 5) was shown to be the most representative for younnger
users. In order to isolate and visualise a user, an individual was selected whhich
contained the value closest to the average number of days value. The average num mber
of active user days for younger
y users was 368 days. To ensure objectivity the
foremost G2 young user to o be found either side of the average number of days w was
selected. In this instance (w
whilst ensuring user privacy) we can reveal a numberr of
user elements; the user wass male, active for 370 days, retaining a frequency for thheir
user comments of 2.03 and for their replies 3.06. As shown below (Fig.3):
Comments Replies
30
A B C
25
Activity Count
20
15
10
0
Feb-10
Jun-10
Sep-10
Jan-10
Mar-10
Oct-10
Nov-10
Jan-11
Jul-10
Apr-10
May-10
Aug-10
Dec-10
Activity Period (per month scale)
Fig. 3. User Activity Chart
4 Analysis and Discussion
Analysis of comment activitya frequencies discloses the following. Relatting

specifically to older users, the
t most dominant grouping was G1, containing 52% off all
older users activity frequeency values. As a group it identifies users whose w walls
contain no user activity. The
T second largest grouping was 'G5', identifying all uuser
activity frequencies greaterr than 20, with 21%. A decreasing trend further continnued
through groups G4 and G3 with 10% and 6% respectively. 'G2' observed 10%. Witthin
younger users notably only 1% of user values were contained within 'G1'. The larggest
younger user grouping waas 'G2' retaining 52% of all younger frequency valuues.
Younger user values declined sharply throughout the remainder of the groupings w with;
G3 containing 23%, G4 15% % and G5 10%. Comparisons of the two groups show cllear
disparities are identifiable between the activity frequencies of young (Fig.4a) and
older groups (Fig.4b). It canc be shown that the majority of young users enggage
frequently, in contrast to older
o users. Older user engagement can now be said too be
much less frequent with maarkedly lower levels of activity.
An Examination of
o the Behaviour of Young and Older Users of Facebook 15
Comment Reply Comment Reply

120 120
90 90
60 60
30 30
0 0
G2 G3 G4 G5 Nil G2 G3 G4 G5 Nill
Fig. 4a. Younger Frequencies

F Fig. 4b. Older Frequencies
As illustrated (Fig.3) in plotting

p user activity patterns are identifiable providing and
understanding of a users habits. A total of 303 activities were observed made upp of
182 user comments and 121 1 user replies. Within Fig.3 a series of groupings have bbeen
highlighted (A-C). Band A illustrates early social network engagement were a cluuster
of low volume recurring acctivity was recorded. A period of consistent activity cann be
seen over approximately thee first 16 weeks from January 2010 to April 2010, were the
majority of activity focused d on user comments. Band B is identifiable as a periodd of
heightened and concerted activity, were a significant increase in activity is visibble,
with a nucleus of activity between
b May 2010 and June 2010. Band C is identifiedd as
current behaviour in the fiinal quarter of the year, with consistent activity. Spikking
patterns are observed in th his period resulting in increased user volumes. Given the
length of engagement (370 days), it is viewed activity is non random and that patteerns
are evident. Notable howev ver is the observation of low or non engagement perioods,
such as that of late July 201
10 or early November 2010.
5 Conclusion
This study makes a num mber of contributions to the knowledge of online uuser
interactions and user behaaviour. This paper has shown that in the online social
network of facebook.com the activity frequency at which younger and older ussers
engage is varied. By harvessting user generated content, contained within a user's wwall
profile, the frequency at whhich a user engages can be determined. The developmennt of
activity frequency metrics now
n quantifies the frequency at which a user contributees to
a social network. As is sh hown, such approaches can enable users to be classiffied
based on their activity. Su ubsequently it is possible for the interaction rates too be
compared across different users
u and against different user types in age groups. Thhese
results show that younger users
u can be classified as high frequency users, produccing
large volumes of online useer data. Older users have been shown to engage less witth a
more infrequent online soccial network engagement. Through the plotting of a usser's
online history, results showw that the behavioural patterns of a user can be observved,
particularly in regard to hig
gh, low and non engagement.
6 Summary of Results, Future Work and Limitations
In determining the activity frequency of a user’s data, and its subsequent

visualisation, it has helped further the understanding of user’s behaviour, detailing
when and how users engage, identifying patterns of early, mid and current
engagement. Importantly it defines high and low periods of activity. As with just one
user example, it may be possible to hypothesize as to the reasoning of activity
patterns. They are observed throughout the spring and early summer months as having
a concerted period of activity. One tentative hypothesis is that behavioural patterns
may have been stimulated by the reaction to increased real world social activities
(holidays, etc.). Although conjecture for this particular user, patterns may realistically
be applied to specific user types for monitoring social engagement, as early warnings
relating to increasing social isolation of vulnerable individuals or groups. Online
social networks have been shown as a means to track a user’s social engagement
levels, which may be applied in epidemiology for pattern anomaly detection.
It is envisioned that our future work in this area will extend current research,
establishing if correlations exist between online activity and user well being.
Limitations of current work are concerned with user data, specifically that the study is
limited to only users that defined privacy settings to 'everyone'. It is hoped that a tool
or process can be developed to overcome this issue, whilst acquiring a larger data set.
References
1. Miniwatts Marketing Group, Internet World Stats. Internet (March 2011),
http://www.internetworldstats.com/stats4.htm (accessed July 12, 2011)
2. Charron, C., Favier, J., Li, C.: Social Computing, http://www.forrester.com
(accessed April 02, 2010)
3. Lewis, K., Kaufman, K., Gonzalez, Christakis, M.: Tastes, ties and time: A new social
network dataset using facebook.com. Social Networks 30(4), 330–342 (2008)
4. Porter, M.A., Onnela, J.P., Mucha, P.J.: Communities in networks. Notices of the American
Mathmatical Society 56(9), 1082–1097 (2009)
5. eBizMBA. Top 15 most popular Social Networking Sites. Internet,
http://www.ebizmba.com/articles/social-networking-websites
(accessed July 12, 2011)
6. Massari, L.: Analysis of MySpace user profiles. Information Systems Frontiers 12(4), 357–
359 (2010)
7. Huang, L., Xia, L.: User Character and Communication Pattern Detecting on Social
Network Site. In: International Conference on Engineering Computation, ICEC 2009, May
2-3, pp. 261–264 (2009)
8. Wilson, C., Boe, B., Sala, A., Puttaswamy, K.P.N., Zhao, B.Y.: User interactions in social
networks and their implications. In: Proceedings of the 4th ACM European Conference on
Computer Systems (Eurosys 2009), pp. 205–218 (2009)
9. Facebook. Facebook Press Room, About Facebook. Internet,
http://www.facebook.com/press.php (July 12, 2011)
An Agile Methodology
for IHE-Enabled Deployments
Bruno Alves and Michael Schumacher
University of Applied Sciences Western Switzerland, 3960 Sierre, Switzerland

bruno.alves@hevs.ch
Abstract. Recent history of healthcare software is littered with fail-

ures, mostly attributable to bad design and the inability to capture user
community’s needs and workflows[1]. This paper presents a new agile
methodology aiming at improving engineering practises in Integrating
the Healthcare Enterprise (IHE)-based projects in small-to-medium scale
hospitals and healthcare organizations. The work described here is a com-
pilation of experiences in the field and is based on the successfull Scrum
methodology, centered on IHE specifications that aims at providing value
by making IHE developments easier and more predictable.
Keywords: Integrating the Healthcare Enterprise, IHE, security,

methodology, planning, Scrum, interoperability.
1 From Hailing to Failing
The Therac-25 debacle is certainly the most serious case of computer-related

accident to date involving patient death. A study published in [2] pointed to
severe deficiencies in the engineering process. Ross Koppel revealed in an article
published in 2005 [3] that computerized physician order entry systems (CPOE)
can unexpectly and contrary to conventional belief increase the number of med-
ication errors, because they fail at capturing user requirements and because
”machine rules do not correspond to work organization or usual behaviors”.
Often, reponsibility for errors is not only attributable to a bad design, but
also partly to the users, who are ”unlikely to accept blame for their own error
or acknowledge their own inadequacies with respect to using the system[1]. In
1999, Kohn and his colleages suggested in To Err Is Human: Building a Safer
Health Care System [4] that healthcare professionals should increase awareness
on how information technology could be applied to deliver safer care. Bad design,
user misinformation and bad practises along with uncontrolled system growth
and fast moving technology can actually lead to a new range of unexpected
consequences. Awareness to the value of healthcare information exchange and
interoperability[5] is probably the key to building better and safer systems.

c Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2012
18 B. Alves and M. Schumacher
Integrating the Healthcare Enterprise (IHE)1 is an initiative led by profession-

als and the industry to facilitate information sharing in healthcare. IHE promotes
the coordinated use of well established standards (such as DICOM2 or HL73 )
to address particular clinical needs in specifications known as Profiles described
both in terms of actors and transactions in a Technical Framework[6]. Each Pro-
file solves an unique clinical use-case (document sharing, patient management,
...), but can be integrated in more general framework. Profiles are systematic
implementation guides that allow developers to focus more on the actual user
experience and less on the integration problems, leading thus to less frustration
for the developers and more satisfaction to the users. The amount of available
Profiles however and the high number of elements to care about may be a bit
overwhelming as more Profiles are integrated into one product. The problem can
be largely mitigated with a careful and methodic development approach.
Healthcare software has usually a rigid set of fixed requirements and one may
think development in healthcare does not require the agility that is essential
in more business-oriented software. Healthcare IT requires high connectivity, a
great number of overlapping subsystems and presents challenging integration
issues, difficult to foresee. Development teams are expected to be extremely
reactive to changes and may have to respond to notable modifications in the
environment variables (timeframe, team, ...). Agile methods provide the kind of
flexibility necessary to handle these cases, and especially Scrum.
Scrum [7] aims at delivering as much quality software as possible in small
frames of time. It improves on existing engineering practises by involving fre-
quent management activities tracking any deficiencies or impediments in the
development process. Scrum particularly adapted to reduced-size regional or-
ganizations, because its process does not require large teams. And, since the
customer is usually the organization itself, the proximity allows for a better
communication and helps in better serving its needs.
The IheTools project is a follow up of the Medicoordination project published
in [8],[9] and [10]. Its main objective is to take the exploration of IHE Profiles
started with Medicoordination a step further. This exploratory research is nec-
essary to gain enough knowledge so to be able to assist regional hospitals and
healthcare organizations in their IHE-related tasks.
This paper presents a new methodology based on the successful Scrum that
aims at assisting small-to-medium scale regional hospitals and healthcare orga-
nizations in their IHE-related development tasks, by providing a consistent and
systematic approach on building up frameworks based on IHE Profiles. A moti-
vation case will be first introduced; the methodology will be then presented and
discussed at the end of this work.
1
IHE International, http://www.ihe.net
2
DICOM Homepage, http://medical.nema.org/
3
Health Level Seven International,http://www.hl7.org/
An Agile Methodology for IHE-Enabled Deployments 19
2 Motivation Case Study
2.1 On the Importance of a Good Plan
Our experience with IHE technologies was mostly acquired in the context of
Medicoordination project. Although the project was successful in several aspects
and the experiment was positively evaluated, the first prototype suffered from a
few important design flaws that could have been avoided through a more careful
design. This experience emphasized the importance of having good planning and
a good methodology. Here are some experiences from the field:
As we came to learn, security is a topic that needs planning. It is in general
a good practice to consider it already in the early stages of the product’s de-
velopment lifecyle. Although it may sound like a good idea to first develop an
unsecured version and then add security to it, it is not. Security planning has
many implications on the very foundations of the architecture and can lead to the
creation of new unplanned breaches. IHE solves the problem partly, by providing
a methodic and well documented security section in every Profile description.
Common security concerns and elements to care about are well documented.
Furthermore, because healthcare software is usually provisionned with patient-
centric sensitive data, not having a risk assessment and mitigation plan is un-
acceptable. Without it, the risk implications are not well understood, users not
trained at is consequences and the final product may go well against local poli-
cies, against law or even against ethics.
Finally a mistake commonly found in research projects is to believe that a
product will integrate flawlessly in the target operating environment. People
often tend to believe there’s no need to take the environment in consideration,
even more in the early stages of the development. Murphy’s Law4 states that
”if something can go wrong, it will eventually go wrong”. Taking the operating
environment into account in the planning allows avoiding most nasty unforseen
consequences.
2.2 Price Is What You Pay. Value Is What You Get5
Small-to-medium scale regional hospitals and healthcare organizations often have

limited financial resources and assets. Although Return on Investment (ROI) is
an important metric in decision making, the primary focus is on the value. Agile
methods concentrate on the value by delivering functional iterations in short
periods of time and also by placing the customer to the center of the discussion.
IHE can also provide value. A set of Open-Source IHE Profile implementations
and frameworks was evaluated and a state of the art was published in [11]. All
the selected implementations are free and IHE-compliant (most of them provide
compliance statements) and may prove of good value for enterprises not willing
to spend millions of euros in commercial licenses.
4
Murphy’s Law, http://en.wikipedia.org/wiki/Murphy’s law/
5
Warrent Buffet 1930, American Investment Entrepreneur.
2.3 The Waterfall Falls, The Scrum Scrums
Unlike the traditional waterfall development model, agile methods do not assume
that the requirements will not change. On the contrary, they embrace change as a
mean to constantly improve the software architecture. Scrum was first introduced
Fig. 1. The Scrum methodology process
in 1986 by Hirotaka Takeuchi and Ikujiro Nonaka [12] as a new approch to

commercial product development. The primary objective was to increase the
speed and flexibility of developments, by enforcing a single team with cross-
functional skills to the whole process.
The Scrum methodology contains a set of predefined roles:
1. the ”ScrumMaster” makes sure the process goes as intended

2. the ”Product Owner” represents the customer
3. the ”Team” works on the delivery of the product
The Scrum methodology does not make assumptions on the model used for the
implementation, but rathers limits itself to the description of how to drive the
development. Scrum consists in three main phases: pre-game, development and
the postgame.
The pre-game phase starts with a meeting to define a set or prioritized high-
level requirements, called product backlog list. A thorough planning is achieved,
including risk assessment, security consideration, project team and so on. In
a second step, an architecture model is designed from the backlog items and
problems, which may arise are identified and mitigated.
The development phase undergoes successive iterations, called sprints that can
last up to 30 days until the release is ready for distribution. Each sprint starts
with a meeting in which several items are taken from product backlog list to work
on. Items are often splitted in smaller tasks and inserted into the sprint backlog.
Some environment variables, such as timeframe, quality, resources and so on are
constantly re-evaluated. The Scrum Master is tied to management activities to
assess that no impediment will ever affect the ongoing development.
The post-game phase leads to a fully working release. The development enters
this phase as soon as all the environment variables (requirements) are com-
pleted. The system is ready to be released and preparation includes tasks such
as integration, user training and documentation.
3 Description of the IHE-Agile Methodology
This section describes the three phases of the Scrum methodology including
elements from the IHE documentation.
3.1 Preparing the Game
The pre-game phase contains all the required groundwork before the actual de-
velopment can start. The phase starts with a planning, in which the central
element is the generation of items in the product backlog by the members of the
whole team, including customers, sales and marketing division, software devel-
opers and a few other roles, that may vary from one project to the other. The
initial planning generally includes elements from security, risk assessment, team
foundation, training needs, integration and testing plans and many others that
have been detailed in other works on agile development.
IHE documents state that a risk assessment and mitigation plan should be
written for each profile. IHE provides a Security Considerations section for every
Profile, which is based on the mitigations identified in each risk assessment. The
Security Considerations section is not a thorough standalone security assessment,
but just deals with issues specifically relevant to interoperability. There is no pre-
cooked recipe for a risk assessment and mitigation plan, since risks usually vary
from one project to the other. However, elements of the method used to write
the Security Considerations section, which is described in the IHE Cookbook:
Preparing the IHE Profile Security Section[13] can be used to help writing the
global risk assessment plan. The method consists in identifying lists of risks, by
imagining different scenarios and then assess their level of impact and probability
of occurrence. Finally, mitigation of relevant risks for each profile is proposed.
Ethical and legal considerations need to be accounted for in the early stages
of the development process. Ethical issues have to be well understood and com-
municated to the future users of the system. It should be clearly stated what is
going to be done with sensitive data, who has access to it, what are the mech-
anisms to protect the data and what should be done in case of unauthorized
disclosure. Legal aspects need to be analyzed transversely taking into account
the laws of the country as well as the laws and policies of the lower levels (region,
organization or business unit).
Security planning should be also partly done in the early phases of the develop-
ment and improved during the subsequent sprints as the technical requirements
become more and more clear. Some risk mitigation solutions proposed in the
Technical Framework, must be analyzed and described in the security planning.
IHE proposes some good articles on Profile-centric access control mechanisms[14]
and on the management of security and privacy. The Template for XDS Affin-
ity Domain Deployment Planning Handbook [15] is usually used in planning the
deployment of a XDS domain, but can also be used to plan organizational, op-
erational and membership rules, as well as patient privacy and consent matters.
System developers should not forget to attach the audit logging planning to the
security section.
3.2 The Scrum Process

The subsections below describe the different phases of a single sprint. This is
where the process shows agility. Indeed, in each sprint, environmental variables,
such as requirements, technology and timeframes, are tracked and the develop-
ment adapted if necessary.
Requirements. During the Sprint Planning Meeting punctuating the begin-

ning of the next sprint, the Scrum Master, customers, users, Product Owner
and Scrum team decide upon the goals and functionality. A second phase of the
meeting gathers the Scrum Master and Scrum Team to decide how the product
increment is going to be implemented and integrated into the existing product.
Analysis. The analysis phase includes elements such as security planning, us-
ability and testing procedures.
Risk assessment must typically be done for every profile integrated in the final
product, because different profiles generate different risks. Furthermore integrat-
ing with other Profiles generates even more risks.
Security planning must include all important information to help making se-
curity decisions. The team should be aware of topics such as actors, roles, au-
thorizations, secure protocols, privacy, confidentiality and patient consent (in-
volvement). It is important to define here what type of data is stored, where it is
stored, who as access to it, for how long. This section may also contain emergency
and bypass protocols as well as audit logs format and auditing placement.
Design. A typical IHE design should include low level aspects, such as IHE
actors, transactions as well as security, interoperability annotations, communi-
cation protocols, encryption standards, protection mechanisms, role-based access
controlling schemes and so on. All the links between the actors sould be clearly
identified and annotated with the type of security they require (mutual TLS over
HTTP, for example) and with used protocols at both ends. For example, if one
link is used to send data from a system that produces HL7 CDA R2 documents,
it is a good idea to write HL7 CDA R2 next to the source actor. This kind of
annotations helps making sure that two integrated systems are talking the same
language and the same version.
Implementation. Demos are important, because they help them assessing the
progress of an ongoing development. The implementation process may consist
in writing modules or deploying existing frameworks, but the general rule is
that implementation efforts should always result in output, which is visible and
presentable at the end of the sprint. Customers do not live on promises, they
want value. IHE projects often tend to mix both coding and deployments. It is
thus important to limit the scope of the current backlog item in order to produce
some visible output until the end of the next deadline, even if that means to
split the item for the next iterations.
Testing. Unit testing is a great tool for assessing the correct behaviour of the
iteration, even though it is far from sufficient. There are other key aspects that
must be tested and validated such as: interoperability, security and usability.
Interoperability testing comes down to assessing that your system is using
the right communication protocols and standard. Sometimes, using a different
version of a same protocol, for example HL7 v2.5 over HL7 2.4, may invalidate
your efforts. Interoperability validation ensures that messages sent by your piece
of software are well understood by the surroundings. IHE Profiles inform about
the expected outcome of a particular transaction. The software tests must use
this data to make sure the current products’ iteration behaves as expected.
Furthermore, IHE Profiles rely heavily on proven standards and are generally
interoperable. It may happen, however, that some specific features or options
are not implemented in one particular IHE Profile implementation. Hence, it is
necessary to perform interoperability validation before proceeding to the next
phase. There exist some IHE testing frameworks, such as the NIST XDS Test
Suite6 or the now famous IHE Connectathons7 , where teams can test their IHE-
enabled products against others and possibly receive compliance statements.
Security testing is a delicate matter, since in most small projects, there is no
true security expert. Testing procedures must assess the security of the software
module for scenarios designed in the planning phase. For safety critical systems,
availability has also to be tested, by putting systems under heavy charges and
inserting deficient nodes.
Usability is one aspect that can make or destroy your project. It is important
to involve user judgment early in the development lifecyle. IHE Profiles are
designed to integrate user’s workflow without much disturbances.
3.3 The Game’s Isn’t over Until It’s over8
The post-game phase represents the end of a release. The project enters in the
post phase upon agreement that all environment variables are completed. The
system is typically finished and release preparation tasks such as integration and
delivery are done at this step.
6
http://ihexds.nist.gov/
7
http://gazelle.ihe.net/content/ihe-europe-2011-connectathon
8
Yogi Berra, American professional baseball player and manager.
Integration. Integration is the operation, which consists in inserting the current

release of the product in the operating environment. For implementations of
typical IHE use-cases, IHE simplifies this integration, by providing guides and
illustrations. Part of the integration work is about configuring the environment to
line up with product’s final specifications. Even through everything was planned
in the pre-game phase, unexpected conditions may arise and a return in another
sprint cycle may be necessary. The integration phase also includes extensive
testing and quality assurance.
Delivery. At this point the product is ready to be delivered to the customer.

In this phase, an extensive documentation has to be created and information
procedures set up, including user training.
Postmortem: Lessons Learned. In almost all developments, some processes

may go straight and others just don’t. Recording in a written and accessible
form what went wrong and what went right allows making continous process
improvements. The recording usually happens in the Sprint Retrospecive meet-
ing. Knowing what can adversely affect an ongoing development effort and how
to respond to it, helps making future developments more predictable and com-
fortable for the team.
4 Discussion
The methodology presented above builds on the success of agile methodologies
and on the rigor and consistency of IHE Profiles. It simplifies the development
of projects in small-to-medium scale regional hospitals and healthcare organiza-
tions, by providing a guide about subjects that require a special care. Further-
more, this methodology also provides links to helpful IHE documents facilitating
the process of writing down good risk assessement plans, creating SOA architec-
tures based on IHE Profiles or preparing the security of the product.
This methodology is meant to be used as a supplementary tool and not as a
all-in-one guide. Is was created as a response to problems we had in our previous
developments. No standard evaluation was performed using existing frameworks,
which may restrain the scope of usability of this methodology to non-safety
critical and non-business critical projects . However, concepts presented here are
taken from our past experiences and may still prove valuable in order to avoid
the same mistakes again and again.
5 Conclusion
We presented here a new agile methodology centered on IHE Profiles that allows
assisting small-to-medium scale healthcare development tasks based on IHE. The
methodology presented here builds on the agility principles of the successful
Scrum agile method and is primarily intended to be used by small-to-medium
scale regional hospitals and healthcare organizations where teams are sufficiently
small and proximity is good.
References
1. Laplante, P.A., Nell, C.J., Sangwan, R.S.: Healthcare Professionals’ Perceptions of
Medical Software and What to Do About it. J. IEEE Computer 39, 26–32 (2006)
2. Leverson, G.S., Turner, C.S.: An Investigation on the Therac-25 Accidents. Com-
puter, 18–41 (1993)
3. Koppel, R., et al.: Role of Computerized Physicain Order Entry Systems in Facil-
itating Medication Errors. JAMA 9, 1197–1202 (2005)
4. Kohn, L.T., Corrigan, J.M., Donaldson, M.S.: To Err Is Human: Building a Safer
Health System, p. 80. Nat’l. Academies Press (2000)
5. Walker, J., Pan, E., Johnston, D., Adler-Milstein, J., Bates, D.W., Middleton,
B.: The value of health care information exchange and interoperability. J. Health
Affairs, hlthaff–w5 (2005)
6. IHE IT Infrastructure (ITI) Technical Framework. IHE International, vol. 1–3,
Revision 7 (2010)
7. Schwaber, K., Beedle, M.: Agile software development with Scrum. Prentice Hall
PTR (2001)
8. Alves, B., Müller, H., Schumacher, M., Godel, D., Abou Khaled, O.: Interoper-
ability prototype between hospitals and general practitioners in Switzerland. In:
Medinfo 2010, Cape Town, South Africa, pp. 366–370. IOS Press (2010)
9. Müller, H., Schumacher, M., Godel, D., Abou Khaled, O., Mooser, F., Ding, S.:
MediCoordination: A practical approach to interoperability in the Swiss health
system. In: The Medical Informatics Europe Conference (MIE 2009) (2009)
10. Alves, B., Schumacher, M., Godel D., Richard, P. Abu Khaled, O., Müller, H.:
Prototype d’interoprabilit entre hpitaux et mdecins traitants. In: Actes de GISEH
2010, Conference Francophone “Gestion et Ingnierie des Systmes Hospitaliers”,
Clermont-Ferrand, France (2010)
11. Alves, B., Schumacher, M.: The Open-Source Integrating the Healthcare Enterprise
(IHE). In: Proceedings of the IADIS e-Health Conference, Rome, Italy (2011)
12. Takeuchi, H., Nonaka, I.: The new new product development game. J. Harvard
Business Review 64, 137–146 (1986)
13. IHE Cookbook: Preparing the IHE Profile Security. IHE International (2008)
14. Caumanns, J., Kuhlisch, R., Pfaff, O., Rode, O.: IHE ITI Infrastructure White
Paper: Access Control. IHE International (2009)
15. IHE IT Infrastructure Technical Commitee White Paper: Template for XDS Affin-
ity Domain Deployment Planning. IHE International, Version 15 (2008)
Obstetric Medical Record Processing and
Information Retrieval
Miroslav Bursa1 , Lenka Lhotska1 , Vaclav Chudacek1 , Michal Huptych1 ,

Jiri Spilka1 , Petr Janku2 , and Martin Huser2
1
Department of Cybernetics, Faculty of Electrical Engineering,
Czech Technical University in Prague, Czech Republic
miroslav.bursa@fel.cvut.cz
http://bio.felk.cvut.cz
2
Obstetrics and Gynaecology clinic,
University Hospital in Brno, Czech Republic
Abstract. This paper describes the process of mining information from

loosely structured medical textual records with no apriori knowledge. In
the paper we depict the process of mining a large dataset of ∼50,000–
120,000 records × 20 attributes in database tables, originating from the
hospital information system (thanks go to the University Hospital in
Brno, Czech Republic) recording over 10 years. This paper concerns
only textual attributes with free text input, that means 613,000 text
fields in 16 attributes. Each attribute item contains ∼800–1,500 charac-
ters (diagnoses, medications, etc.). The output of this task is a set of
ordered/nominal attributes suitable for rule discovery mining and auto-
mated processing.
Information mining from textual data becomes a very challenging task
when the structure of the text record is very loose without any rules.
The task becomes even more difficult when natural language is used
and no apriori knowledge is available. The medical environment itself is
also very specific: the natural language used in textual description varies
with the personality creating the record (there are many personalized
approaches), however it is restricted by terminology (i.e. medical terms,
medical standards, etc.). Moreover, the typical patient record is filled
with typographical errors, duplicates, ambiguities and many (nonstan-
dard) abbreviations.
Note that this project is an ongoing process (and research) and new
data are irregularly received from the medical facility, justifying the need
for robust and fool-proof algorithms.
Keywords: Swarm Intelligence, Ant Colony, Textual Data Mining,

Medical Record Processing, Hospital Information System.
1 Introduction
1.1 Motivation
In many industrial, business, healthcare and scientific areas we witness the
boom of computers, computational appliances, personalized electronics, high-

Medical Record Information Retrieval 27
speed networks, increasing storage capacity and data warehouses. Therefore a

huge amount of various data is transferred and stored, often mixed from differ-
ent sources, containing different data types, unusual coding schemes, and seldom
come without any errors (or noise) and omissions. Massively parallel distributed
storage systems are used nowadays to provide computational nodes with data
in reasonable time.
1.2 Nature Inspired Methods

Nature inspired metaheuristics play an important role in the domain of artificial
intelligence, offering fast and robust solutions in many fields (graph algorithms,
feature selection, optimization, clustering, etc). Stochastic nature inspired meta-
heuristics have interesting properties that make them suitable to be used in data
mining, data clustering and other application areas.
With the boom of high-speed networks and increasing storage capacity of
database clusters and data warehouses, a huge amount of various data can be
stored. Knowledge discovery and Data mining is not only an important scien-
tific branch, but also an important tool in industry, business and healthcare.
These techniques target the problematic of processing huge datasets in reason-
able time – a task that is too complex for a human. Therefore computer-aided
methods are investigated, optimized and applied, leading to the simplification
of the processing of the data.
Ant Algorithms. Ant colonies inspired many researchers to develop a new

branch of stochastic algorithms: ant colony inspired algorithms. Based on the ant
metaphor, algorithms for both static and dynamic combinatorial optimization,
continuous optimization and clustering have been proposed. They show many
properties similar to the natural ant colonies, however, their advantage lies in
incorporating the mechanisms, that allowed the whole colonies to effectively
survive during the evolutionary process.
1.3 Knowledge Extraction

Several techniques to extract knowledge from raw data have been developed in
the past. These techniques have various and multiple origins: some result from
the statistical analysis of the data, the regressions, decision trees, etc.; some
resulting from the artificial intelligence such as the expert systems, intelligent
agents, fuzzy logic, etc.
Text Extraction. The accuracy for relation extraction in journal text is typ-
ically about 60 % [1]. A perfect accuracy in text mining is nearly impossible
due to errors and duplications in the source text. Even when linguists are hired
to label text for an automated extractor, the inter-linguist disparity is about
30 %. The best results are obtained via an automated processing supervised by
a human [2].
28 M. Bursa et al.
Ontologies have become an important means for structuring knowledge and

building knowledge-intensive systems. For this purpose, efforts have been made
to facilitate the ontology engineering process, in particular the acquisition of
ontologies from texts.
2 Input Dataset Overview
The dataset consists of a set of approx. 50 to 120 thousand records (structured in

different relational DB tables; some of them are not input, therefore the range is
mentioned) × approx. 20 attributes. Each record in an attribute contains about
800 to 1500 characters of text (diagnoses, patient state, anamneses, medications,
notes, references to medical stuff, etc.). For textual mining, 16 attributes are
suitable.
The overview of one small (in field length) attribute is visualized in Fig. [2].
Only a subsample (about 5 %) of the dataset could be displayed in this paper,
as the whole set would render into a uncomprehensible black stain. The vertices
(literals) are represented as a green circle, the size reflects the literal (i.e. word)
frequency. Edges represent transition states between literals (i.e. the sequence of
2 subsequent words in a sentence/record); edge stroke shows the transition rate
(probability) of the edge. The same holds for all figures showing the transition
graph, only a different visualization approach has been used.
It is clear, that human interpretation and analysis of the textual data is very
fatiguing, therefore any computer aid is highly welcome.
3 Graph Explanation
In this paper we describe transition graphs. These are created for each attribute.
An attribute consists of many records in form of a sentence. By sentence we
hereby mean a sequence of literals, not a sentence in a linguistic form. The
records are compressed – unnecessary words (such as verbs is, are) are omitted.
In this paper, only the atribute delivery anesthetics is visualized, as it is the
simplest one.
Vertices of the transition graph represent the words (separated by spaces) in
the records. For each word (single or multiple occurence) a vertex is created and
its potence (number of occurences is noted). For example, the words mesocaine,
anesthetics, not, mL form a vertex. Note that also words as mesocain, mezokain
and other versions of the word mesocaine are present. For a number (i.e. sequence
of digits) a special literal NUMBER is used.
Edges are created from single records (sentences entered). For example the
sentence mesocaine 10 mL would add edges from vertex mesocaine to vertex
NUMBER and from vertex NUMBER to the vertex mL (or the edge count is
increased in case it exists). For all records, the count of the edges is also useful.
It provides an overview on the inherent structure of the data – the most often
word transitions.
4 Motivation
The task of this work is to provide the researchers with a quick automated or
semi-automated view on the textual records. Textual data are not easy to visu-
alize. The word frequency is inapropriate, although it is very simple. Therefore
we decided to extract information in the form of a transition graph.
Using these graphs a set of rules for information retrieval is bein created
(defined). These rules serve for extraction of (boolean) attributes from the tex-
tual rules. These attributes are used in automated rule discovery and can be
further used for recommendation.
5 Nature Inspired Techniques
Social insects, i. e. ant colonies, show many interesting behavioral aspects, such
as self-organization, chain formation, brood sorting, dynamic and combinatorial
optimization, etc. The coordination of an ant colony is of local nature, composed
mainly of indirect communication through pheromone (also known as stigmergy.
The high number of individuals and the decentralized approach to task coor-
dination in the studied species means that ant colonies show a high degree of
parallelism, self-organization and fault tolerance. In studying these paradigms,
we have high chance to discover inspiration concepts for many successful meta-
heuristics.
5.1 Ant Colony Optimization
Ant Colony Optimization (ACO) [3] is an optimization technique that is in-

spired by the foraging behavior of real ant colonies. Originally, the method was
introduced for the application to discrete and combinatorial problems.
Ant Colony Methods for Clustering. Several species of ant workers have
been reported to form piles of corpses (cemeteries) to clean up their nests. This
aggregation phenomenon is caused by attraction between dead items mediated
by the ant workers. The workers deposit (with higher probability) the items in
the region with higher similarity (when more similar items are present within
the range of perception). For example, the Messor sancta ants organize dead
corpses into clusters; brood sorting has been studied in ant colony of Leptothorax
unifasciatus.
This approach has been modeled in the work of Deneubourg et al. [4] and in
the work of Lumer and Faieta [5] to perform a clustering of data.
Methods using pheromone also exist, namely A2 CA [6]. Another approach can
be found the work of J. Handl in [7] (an ATTA algorithm), which introduce mod-
ified neighborhood function (penalizing high dissimilarities), short-term memory
with lookahead (jumping ants), increasing radius of perception, time-dependent
modulation of the neighborhood function.
30 M. Bursa et al.
ACO DTree Method. The ACO DTree method is a hybrid evolutionary

approach for binary decision tree construction [8,9]. The tree is induced using
the known data and it can be used for unsupervised clustering later: each leaf
of the classification tree can be interpreted as a cluster. The algorithm uses a
population of classification trees that is evolved using an evolutionary approach.
Creation of the trees is driven by a pheromone matrix, which uses the ACO
paradigm. The high number of individuals and the decentralized approach to
task coordination in the studied species means that ant colonies show a high
degree of parallelism, self-organization and fault tolerance.
This approach has been utilized (with improvements) to simplify the structure
of the vast dataset by finding the most important state transitions between
literals, producing a probabilistic transitional model. The output structure is
presented to the analyst for further processing/iteration.
For clustering, the ACO DTree method [8,10] and ACO inspired clustering [5]
variations have been successfully used.
6 Automated Processing
Automated layout of transition graph is very comfortable for an expert, how-
ever the contents of the attribute is so complicated, that a human interven-
tion is inevitable. Examples of automated layout can be seen in Fig. [1] and
Fig. [2].
The figure Fig. [1] shows a transitional graph where only positioning based on
the word distance from the sentence start is used. Althoug it migh look correct,
note that the same words are mispositioned in the horizontal axis.
Fig. 1. A fully automated transition graph showing the most important relations in
one textual attribute. No clustering has been used. The layout is based on the word
distance from the start of the sentence. Note the mis-alignment of the similar/same
words. Refer to section [2].
Fig. 2. A fully automated transition graph (sub-graph) showing the most important
relations in one textual attribute. The ACO approach has been used to cluster the
corresponding vertices. Refer to section [2].
7 Expert Intervention
A human intervention and supervision over the whole project is indiscutable.
Therefore also human (expert) visualization of the transition graph has been
studied.
The vertices in a human-only organization are (usually) organized depending
on the position in the text (distance from the starting point) as the have the
highest potence. Number literal (a wildcard) had the highest potence, as many
quantitative measures are contained in the data (age, medication amount, etc.).
Therefore it has been fixed to the following literal, spreading into the graph via
multiple nodes (i.e. a sequence mesocain 10 mL become two vertices – meso-
cain NUMBER and mL). This allowed to organize the chart visualization in
more logical manner. Time needed to organize such graph was about 5–10 min-
utes. The problem is that the transition graph contains loops, therefore the
manual organization is not straigthforward.
An aid of a human expert has been used in semi-automated approach (see
Fig. [3] where the automated layout has been corrected by the expert. The
correction time has been about 20–30 seconds only.
8 Results and Conclusion

The main advantage of the nature inspired concepts lies in automatic finding
relevant literals and group of literals that can be adopted by the human analysts
and furthermore improved and stated more precisely. The use of induced proba-
bilistic models in such methods increased the speed of loosely structured textual
attributes analysis and allowed the human analysts to develop lexical analysis
grammar more efficiently in comparison to classical methods. The speedup (from
32 M. Bursa et al.

$

$

$

$

$
$
$"
$"
$"

&'('
$"
!

"

")
%#
" %#
" #
Fig. 3. A semi-automated (corrected by a human expert) organized transition graph

showing the most important relations in one textual attribute. Refer to section [2].
about 5–10 minutes to approx 20–30 seconds) allowed to perform more iterations,
increasing the yield of information from data that would be further processed
in rule discovery process. However, the expert intervention in minor correction
is still inevitable. The results of the work are adopted for rule discovery and are
designed to be used in expert recommendation system.
9 Discussion and Future Work

The future work is to evaluate the DB analyst’s utilization and aid of such graphs
in more accurate way. The graphs serve as a bases for extraction rule proposal.
However the only relevant measure is the time to reorganize the transitional
graphs. The subjective opinion is very expressive and is not coherent. Next, the
semantic meaning of the attributes will be extracted and verified followed by
rule discovery mining.
Acknowledgements. This research project has been supported by the research

programs number MSM 6840770012 ”Transdisciplinary Research in the Area of
Biomedical Engineering II” of the CTU in Prague, sponsored by the Ministry of
Education, Youth and Sports of the Czech Republic and by the project number
NT11124-6/2010 ”Cardiotocography evaluation by means of artifficial intelli-
gence” of the Ministry of Health Care. This work has been developed in the
BioDat research group http://bio.felk.cvut.cz.
References
1. Freitag, D., McCallum, A.K.: Information extraction with hmms and shrinkage.
In: Proceedings of the AAAI Workshop on Machine Learining for Information
Extraction (1999)
2. Lafferty, J., McCallum, A., Pereira, F.: Conditional random fields: Probabilistic
models for segmenting and labeling sequence data. In: Proceedings of the ICML,
pp. 282–289 (2001); Text processing: interobserver agreement among linquists at
70
3. Dorigo, M., Stutzle, T.: Ant Colony Optimization. MIT Press, Cambridge (2004)
4. Deneubourg, J.L., Goss, S., Franks, N., Sendova-Franks, A., Detrain, C., Chretien,
L.: The dynamics of collective sorting robot-like ants and ant-like robots. In: Pro-
ceedings of the First International Conference on Simulation of Adaptive Behavior
on From Animals to Animats, pp. 356–363. MIT Press, Cambridge (1990)
5. Lumer, E.D., Faieta, B.: Diversity and adaptation in populations of clustering ants.
In: From Animals to Animats: Proceedings of the 3th International Conference on
the Simulation of Adaptive Behaviour, vol. 3, pp. 501–508 (1994)
6. Vizine, A.L., de Castro, N.L., Hruschka, E.R., Gudwin, R.R.: Towards improving
clustering ants: An adaptive ant clustering algorithm. Informatica 29, 143–154
(2005)
7. Handl, J., Knowles, J., Dorigo, M.: Ant-based clustering and topographic mapping.
Artificial Life 12(1), 35–61 (2006)
8. Bursa, M., Lhotska, L., Macas, M.: Hybridized swarm metaheuristics for evolution-
ary random forest generation. In: Proceedings of the 7th International Conference
on Hybrid Intelligent Systems 2007 (IEEE CSP), pp. 150–155 (2007)
9. Bursa, M., Lhotska, L.: Ant colony cooperative strategy in biomedical data clus-
tering. SCI. Springer, Heidelberg (2007)
10. Bursa, M., Huptych, M., Lhotska, L.: Ant colony inspired metaheuristics in biolog-
ical signal processing: Hybrid ant colony and evolutionary approach. In: Biosignals
2008-II, vol. 2, pp. 90–95. INSTICC Press, Setubal (2008)
Health Care Performance Monitoring Using an Inpatient
Reference Process Model and Clinical KPIs
Eva Gattnar1,2, Okan Ekinci1, and Vesselin Detschew2

1 Clinical Competence Center Cardiology, Siemens Healthcare
Allee am Röthelheimpark 3a, 91052 Erlangen, Germany
{eva.gattnar,okan.ekinci}@siemens.com
2 Institute of Biomedical Engineering and Informatics, Ilmenau Technical University
POB 100565, 98684 Ilmenau, Germany

vesselin.detschew@tu-ilmenau.de
Abstract. Health care processes are nowadays heavily dependent on

Information Technology (IT). On the other side, assessing a higher quality of
health care to patients has become increasingly important. Effective health care
processes are thereby heavily dependent on a comprehensive IT-support.
Therefore, a thorough understanding about both the hospital IT-systems and
clinical pathways is required, to identify quality problems. Improving quality of
care through process performance measurement in hospitals and the
identification of bottlenecks in performed workflows is thereby a promising
concept. On this account we propose a novel approach based on an inpatient
reference process model aiming a standardized and automated retrieval of
quality and performance metrics along the clinical pathways of time-critical
diseases in the context of various clinical standards and systems.
Keywords: performance monitoring, clinical key performance indicators,

patient process, clinical workflow, clinical process modeling, events.
1 Introduction
With the enormous impact of Information Technology (IT), there is a major demand
for standardization in health care nowadays. On the other side, assessing a higher
quality of health care to patients has become increasingly important. Effective health
care processes are thereby heavily dependent on a comprehensive IT-support.
Therefore, a thorough understanding about both the hospital IT-systems and clinical
pathways is required, to identify quality problems.
In recent years, hospital performance assessment projects and health care quality
reporting have become therefore common worldwide [1], resulting in numerous
initiatives aiming the development of clinical quality indicators to measure health care
quality including hospital care performance measurement [2]. Improving quality of care
through process performance measurement in hospitals and the identification of
bottlenecks in performed workflows is thereby a promising concept. Process
performance measures can assist hospitals in assessing their workflows and identifying
areas for improvements in the field of diagnostics and treatment.
Health Care Performance Monitoring Using an Inpatient Reference Process Model 35
Several health care organizations recognized the importance of using performance

metrics to monitor the clinical workflows. However, there is a lack of approaches
aiming also the “real-time” end-to-end patient process monitoring, which can provide
the possibility, to assist the clinicians in improving the quality of care already during
the diagnostics and treatment and not only afterwards. Event-based approaches using
workflow-events are a promising opportunity to measure the process performance and
to monitor the clinical pathways, to improve the quality of care and benchmark the
healthcare organization. On this account we propose a novel approach based on an
inpatient reference process model aiming the standardized and automated retrieval of
quality and performance metrics like Key Performance Indicators (KPIs) along the
clinical pathways of time-critical diseases in the context of various clinical standards
and systems. Beside the definition of the new model we develop the required clinical
quality metrics systematically and analyze the hospital IT-systems, devices and
communication standards regarding the process-based extraction of timestamps. In
this paper we describe the conditions behind the approach and the results achieved.
2 Health Care Process Performance Monitoring

Processes are significant components within a health care organization. Process
orientation is among others the most important method to support patient-oriented
high quality of care. The economical benefit of recognized, well-defined, optimized,
IT-supported and hence resulting mature processes is approvable too. Following the
international Detecon study [3], a higher maturity of an organization’s processes
results directly in an improved Return on Equity (ROE) as well as in an improved
Return on Assets and Return on Investment. The resulting positive financial and
quality-based effects are exactly measurable [4]. Consequently, to be successful and
to provide an efficient and effective diagnostics and treatment, health care
organizations have to apply optimization approaches aiming a higher process maturity
and improved process performance [4][5].
Health care performance correlates often directly with process performance [5].
Therefore, processes should be applied as an important basis for health care quality
improvement. Process models can help to analyze the process quality using structured
process descriptions to allow process assessments. Secondly, they provide the basis for
improvement steps to achieve a better process performance and in the following a higher
quality of the processes within the health care organization. In general, process
performance monitoring initiates and supports process improvement within an
organization [5].
Thereby, hospitals need reliable and valid indicators for benchmarking and process
improvement [6][7]. Following [6], health care process monitoring is impossible
without using clinical indicators. Various types of such indicators can be either
generically defined to support the measurement of several diseases or specifically
defined to provide disease-specific measures [6]. Thereby, process-related indicators
enable the performance measurement of the activities and tasks in patient’s “episodes
of care” [8]. Based on the performance measurement theories and methodologies
derived from industry, they are commonly called Key Performance Indicators (KPIs).
36 E. Gattnar, O. Ekinci, and V. Detschew
In particular, the development of reliable and valid clinical indicators requires a

rigorous scientific design, definition and implementation [9]. Numerous
organizations, such as the Institute of Medicine (IOM) [10], accrediting organizations
(JCAHO, NCQA), government healthcare research and quality agencies (AHRQ,
CMS) and various research groups devise and publish clinical indicator sets for
measuring purposes in various domains of quality of care [11]. However, not all of
these indicators are considered by physicians or hospitals to be valuable or relevant in
their practice. Furthermore, health care systems are complex and therefore indicators
may not reflect accurately all aspects of quality [12]. Another problem is that their
development is frequently driven by feasible availability of data or ease of
measurement, rather than by an objective definition of important clinical indicators
[5]. Consequently it is essential to define indicators involving all stakeholders like
physicians and IT-experts and additionally to consult approved guidelines, to ensure
the indicator relevance, reliability, reproducibility and validity [11] and provide
the possibility to develop both generic and disease-specific indicators.
We focus on clinical quality of care indicators and relate to the processes, which
are performed in the care of time-critical diseases. The usage of time-based indicators
for quality assessment in time-critical workflows represents a possibility for
measurement, documenting and improving the quality of care [13][15]. Aiming the
collection of time-based indicators in hospitals required IT-timestamps have to be
retrieved. IT-systems act in our approach as enabler for performance monitoring,
clinical quality improvement as well as risk and cost reduction [14][16]. The involved
IT-systems are the Hospital Information System (HIS) and in the radiology
department: the Radiology Information System (RIS) and imaging devices
(Modalities). Outside the radiology department the Laboratory Information System
(LIS) is used. Because several systems and communication standards (DICOM and
HL7) (see Fig.1) are involved in the clinical workflow, we developed a new inpatient
process model to enable performance monitoring across several clinical systems [16].
Fig. 1. IT-systems and clinical standards involved in the diagnostic workflow in hospitals
Fig.1 shows the involved IT-systems and the timeline during the diagnostics of an
inpatient (i.e. a patient who stays at hospital for diagnostics and treatment).
Additionally, timestamps (t) and events (e) are exemplary marked in the figure. Event
(e1) means the patient’s admission at hospital. The two following events mark the
beginning and the end of the laboratory procedure, i.e. the KPI “order to lab test”. The
fourth and fifth events allow the imaging procedure monitoring at a modality (i.e.
imaging device). Here the KPI “order to imaging” can be determined. Thereby, the
timestamps, these events are based on, are extracted from the logfiles stored at the
modalities. Using modality logfiles for monitoring purposes is a novel approach we
introduce to enable process monitoring at the lowest process level. Finally, the last
event indicates the finishing of the diagnostic process flow. The difference between
the first and last event timestamps provides the KPI “door to diagnosis”, which is very
important in time critical diseases like heart attack and stroke [17].
3 Results and Validation
We developed a new generic inpatient reference process model to provide the

possibility of standardized and comprehensive performance process monitoring in
hospitals in the field of time-critical diseases. In order to monitor the process cycle
times, health care organizations are up to date faced with the problem of extracting
IT-timestamps from several not well-connected IT-systems, which are not related to a
specific inpatient process. Consequently, it is not possible to monitor the whole end-
to-end inpatient process-flow starting with the patient’s admission at hospital until the
patient’s discharge. Secondly, valid and reliable performance monitoring requires
standardized, meaningful and well-defined timestamps and events, which have to be
monitored in order to measure the process cycle times. We provide these possibilities
and fulfill both requirements bridging the gap between the IT and the clinical context
proposing our model and KPIs [16]. As the definition of such events and the
indicators based on them is very time-consuming and difficult [13], a predefined
generic approach - as we present it - is desirable by the stakeholders. It can be used
already in an early phase during the assessment in hospital to analyze the pathways
and to define the measure points based on the IT-system architecture [14].
The first model development phase comprised the construction of our newly
clinical reference process model using events, functions, connectors, control flows
and process path elements. We selected the relevant process tasks based on interviews
we performed with sophisticated clinicians. This group was built up of physicians like
cardiologists, neurologists and radiologists who have huge and long-time experience
in hospitals in heart attack and stroke treatment. Individually observations of the
workflows during diagnostics and treatment of heart attack and stroke patients in
several hospitals built up the second information source. After finishing the modeling
tasks the physicians were consulted again in order to approve the model, the
acceptance of the chosen modeling technique as well as the clinical correctness. In
further steps we enhanced the model analyzing clinical pathways, as they are used in
best practice hospitals. “Best practice” means in this context, that the considered
hospitals are best in treating heart attack resp. stroke regarding quality of care.
In the following, we conducted interviews with involved physicians again, to

derive relevant, valid, reliable and feasible clinical indicators for process monitoring.
Additionally we consulted clinical guidelines for stroke and heart attack treatment.
We derived sets of time-based indicators following [18],[19], which we approved
again with experienced clinicians. Afterwards we generalized the indicators to use
them in our generic reference process model. Finally we investigated the IT-
infrastructures, which are nowadays commonly installed in hospitals.
To provide process-based and integrated performance measurement support we
selected the event-based process chain (EPC)-method for modeling. EPCs have
become widely-used for business process modeling in Europe, in particular in
countries where SAP acts as a leading Enterprise Resource Planning (ERP) system
[20]. EPCs are inspired from petri nets, incorporate role concepts and data models like
ER-models or UML class diagrams [20]. According to [21], the EPC belongs to the
most advanced business process modeling languages beside the UML and is one of
the most used languages in the measurement researching community [22].
We evaluated and verified our approach with several clinicians and medical
specialists applying it in the hospital environment. Therefore we simulated the clinical
pathways of several hospitals using our EPC-model in the ARIS-Toolset. ARIS
provides the possibility to test and to simulate the modelled processes already at
design time, i.e. before an implementation in the clinical IT-environment is started or
performed. An integrated “Process Performance Manager“ (ARIS PPM) can be used
for process cycle times analysis purposes. ARIS PPM extracts information from
logfiles and shows them graphically including process bottlenecks. The second
evaluation step involved the mapping of the IT-systems, which were used in the
hospitals in question to the modelled events. Thereby, the events, which can be
automatically extracted, were identified. Finally the derived clinical indicators were
evaluated again in discussion with clinicians as well as through a comprehensive
literature study.
The objective of this practical validation was first to demonstrate the applicability
of the approach in the hospital environment. This included the approval of the
appropriateness of the modelling approach. First, the used EPC-methodology was
approved as understandable for clinicians and as practicable for process monitoring in
hospitals. As we modelled and use separate EPC-modules in our approach, which are
loosely connected using process interfaces, we assure the flexibility that is required in
such a highly flexible clinical environment. This means that several diagnostic and
treatment procedures as well as performance measures can be repeated and performed
in parallel if it is necessary. Finally, our modelled generic indicators were approved as
relevant, valid, reliable and feasible. They are relevant because they relate to
clinically important, evidence-based aspects of care over which clinicians have
significant control. Furthermore they are valid as they correlate strongly with the
measurement goals. Very important was the approval of the indicator reliability as the
indicators are mapped to selected model events and therefore valid for comparison
and standardized recording. Finally, the indicators are feasible as they are sustainable
over time. In future work we will formalize our approach stronger and consider
a graph valuated mechanism therefore. As we focus on time-critical diseases, a
statistical evaluation would require a comprehensive study involving several

hospitals. In near future, statistical results will be provided therefore using simulation.
4 Related Work
So far, there is no process model published that describes the integration of clinical
performance metrics for acute diseases taking in particular into account the
possibilities for their automatic extraction using clinical information systems during a
patient’s stay in hospital and additionally considering the characteristics of the clinical
environment, such as flexible workflows and modular process structure. Nevertheless,
focusing on classical workflow analysis, event-based process monitoring and mining
approaches with the objective to optimize the workflows are presented in [23], [24],
[25] and [26]. However, they were not designed for health care purposes and focus on
more standardized processes as they are usually performed in clinical domain. An
identification approach aiming the clinical context identification and categorization is
proposed in [27].
A KPI-based framework for process-based benchmarking of hospital information
systems is presented in [28]. In contrast to our approach, documentation processes
and outcome criteria related to the process flow and underlying structures in the area
of the Hospital Information System (HIS) aiming benchmarking possibilities are there
focused on. An approach using the derivation of hospital-specific clinical guidelines
is described in [29]. Thereby, lifecycle support for medical guidelines and pathways is
the objective.
Further clinical (reference) process models are presented in [30], [31], [32], [33],
[34]. In [35] the authors present a modeling approach, which introduces principles of
process modeling in healthcare using EPCs and employ these principles to existing
medical information systems by implementing them using a Workflow Management
System (WFMS). However, the possibility for event-based clinical performance
measurement in time-critical diseases, as we introduce it in our approach, is not
provided. Additionally, the model-based generic and disease-specific performance
monitoring is also not supported.
5 Conclusion and Outlook
Especially in the field of time-critical diseases, performance monitoring and

measurement is important regarding process optimization and quality improvement in
hospitals. Our approach comprehends performance monitoring using standardized
time-based performance indicators based on clinical IT-systems and standards as well
as the developed inpatient clinical process model. In future work we will investigate
the usage of our approach in clinical workflow engines to provide the possibility to
manage and improve the IT-based workflow execution and to enable process mining.
References
1. Groene, O., Skau, J., Frolich, A.: An international review of projects on hospital
performance assessment. International Journal for Quality in Health Care 20(3), 162–171
(2008)
2. Copnell, B., et al.: Measuring the quality of hospital care: an inventory of indicators.
Internal Medicine Journal 39(6), 352–360 (2009)
3. n.e.: Der Einfluss der Prozessreife auf den Erfolg von Telekommunikationsunternehmen,
Detecon study (2009), http://www.detecon.com/bpm
4. Gattnar, E., Ekinci, O.: Process Assessment and Improvement Using Capability Maturity
Models. In: Podium der Wirtschaft, Recht & Soziales, vol. 22, pp. 121–148. VHK, Altdorf
b. Nbg (2011)
5. Gattnar, E., Ekinci, O.: Qualitätsmanagement im Krankenhaus – Konzepte,
Rahmenbedingungen und Standards. In: Podium der Wirtschaft, Recht & Soziales, vol. 21,
pp. 67–130. VHK, Altdorf b. Nbg (2011)
6. Mainz, J.: Developing evidence-based clinical indicators: a state of the art methods primer.
International Journal for Quality in Health Care 15(supp. 1), i5–i11 (2003)
7. de Vos, M., et al.: Quality measurement at intensive care units: which indicators should we
use? Journal of Critical Care 22, 267–274 (2007)
8. Rubin, H.R., Pronovost, P., Diette, G.B.: The advantages and disadvantages of process-
based measures of health care quality. International Journal for Quality in Health
Care 13(6), 469–474 (2001)
9. Rubin, H.R., Provonost, P., Diette, G.B.: From a process of care to a measure: the
development and testing of a quality indicator. International Journal for Quality in Health
Care 13(6), 489–496 (2001)
10. Crossing the Quality Chasm: A new Health System for the 21st Century. Committee on
Quality of Health Care in America, Institute of Medicine (IOM), 4th edn. National
Academy Press, Washington (2003)
11. McGlynn, E.A.: Identifying, Categorizing, and Evaluating Health Care Efficiency
Measures. Final Report (prepared by the Southern California Evidence-based Practice
Center—RAND Corporation, under Contract No. 282-00-0005-21). AHRQ Publication
No. 08-0030. Agency for Healthcare Research and Quality, Rockville, MD (2008)
12. Pencheon, D.: The good indicators guide: understanding how to use and choose indicators.
NHS Institute for Innovation and Improvement, Association of Public Health
Observatories (APHO) (2007)
13. Gattnar, E., Ekinci, O., Detschew, V.: Event-based Workflow Analysis in Healthcare. In:
Proceedings of the 9th International Workshop on Modeling, Simulation, Verification and
Validation of Enterprise Information Systems (MSVVEIS 2011), in conjunction with the
13th International Conference on Enterprise Information Systems (ICEIS 2011), pp. 61–
70. SciTePress (2011)
14. Gattnar, E., Ekinci, O., Detschew, V.: Clinical Process Modeling and Performance
Measurement in Hospitals. In: Proceedings of the 5th International Workshop on Advances
in Quality of Service Management (AQuSerM 2011), in conjunction with 15th IEEE
Enterprise Computing Conference (EDOC 2011), pp. 132–140. IEEE Computer Society
Press, Los Alamitos (2011)
15. Gattnar, E., Ekinci, O., Detschew, V.: Interoperable Process Monitoring using Clinical IT-
Standards and Healthcare Information Systems. In: Proceedings of the IADIS International
Conferences Informatics 2011, Wireless Applications and Computing 2011,
Telecommunications, Networks and Systems 2011, pp. 59–66. IADIS Press (2011)
16. Gattnar, E., Ekinci, O., Detschew, V.: An Integrated Approach to Standardized and
Automated Extraction of Clinical Process Cycle Times. In: Proceedings of the IADIS
International Conferences eHealth 2011, pp. 128–136. IADIS Press (2011)
17. Gattnar, E., Ekinci, O., Detschew, V.: A Novel Generic Clinical Reference Process Model
for Event-Based Process Times Measurement. In: Abramowicz, W., Maciaszek, L., Węcel,
K. (eds.) BIS 2011 Workshops. LNBIP, vol. 97, pp. 65–76. Springer, Heidelberg (2011)
18. Wollersheim, H., et al.: Clinical indicators: development and applications. The
Netherlands Journal of Medicine, Van Zuiden Communications B.V. 65(1), 15–22 (2007)
19. Scott, I., Phelps, G.: Measurement for improvement: getting one to follow the other.
International Medicine Journal, Royal Australasian College of Physicians 39, 347–351
(2009)
20. Korherr, B., List, B.: Extending the EPC and the BPMN with Business Process Goals and
Performance Measures. In: Proceedings of the 9th International Conference on Enterprise
Information Systems (ICEIS 2007) (2007)
21. List, B., Korherr, B.: An Evaluation of Conceptual Business Process Modelling
Languages. In: Proceedings of the 21st ACM Symposium on Applied Computing (SAC
2006). ACM Press (2006)
22. Sánchez Gonzáles, L., Rubio, F.G.: Measurement in business processes: a systematic
review. Business Process Management Journal 16(1), 114–134 (2010)
23. zur Mühlen, M., Rosemann, M.: Workflow-based process monitoring and controlling-—
technical and organizational issues. In: Sprague, R. (ed.) Proceedings of the 33rd Hawaii
International Conference on System Science (HICSS-33), pp. 1–10. IEEE Computer
Society Press, Los Alamitos (2000)
24. van der Aalst, W.M.P.: Business Alignment: Using Process Mining as a Tool for Delta
Analysis. Requirements Engineering 10(3), 198–211 (2005)
25. van der Aalst, W.M.P., et al.: Workflow mining: A survey of issues and approaches. Data
& Knowledge Engineering 47(2), 237–267 (2003)
26. van der Aalst, W.M.P., ter Hofstede, A.H.M., Weske, M.: Business Process Management:
A Survey. In: van der Aalst, W.M.P., ter Hofstede, A.H.M., Weske, M. (eds.) BPM 2003.
LNCS, vol. 2678, pp. 1–12. Springer, Heidelberg (2003)
27. Ghattas, J., Peleg, M., Soffer, P., Denekamp, Y.: Learning the Context of a Clinical
Process. In: Rinderle-Ma, S., Sadiq, S., Leymann, F. (eds.) BPM 2009. LNBIP, vol. 43, pp.
545–556. Springer, Heidelberg (2010)
28. Jahn, F., Winter, A.: A KPI Framework for Process-based Benchmarking of Hospital
Information Systems. In: Proceedings of the 23th International Conference of the European
Federation for Medical Informatics (MIE 2011), pp. 542–546. IOS Press (2011)
29. Lenz, R., Reichert, M.: IT support for healthcare processes – premises, challenges,
perspectives. Data & Knowledge Engineering 61, 39–58 (2007)
30. Eisentraut, K., et al.: Abbildung klinischer Behandlungspfade als Modelle in der Unified
Modeling Language. 53. Jahrestagung der Deutschen Gesellschaft für Medizinische
Informatik, Biometrie und Epidemiologie (gmds) (2008)
31. Mauro, C., Happle, T., Sunyaev, A., Krcmar, H., Leitmeister, J.M.: From Medical
Processes to Workflows – Modeling of Clinical Pathways with the Unified Modeling
Language. In: Proceedings of 3rd International Conference on Health Informatics
(HealthInf 2010), pp. 388–391. INSTICC Press (2010)
32. McGregor, C., et al.: A Structured Approach to Requirements Gathering Creation Using
PaJMa Models. In: Proceedings of the International Conference of IEEE Engineering in
Medicine and Biology Society, Vancouver, Canada, pp. 1506–1509 (2008)
33. Winter, A., Haux, R.: A Three Level graph-based Model for the Management of
Computer-Supported Hospital Information Systems. Methods of Information in
Medicine 34(4), 378–396 (1995)
34. Buchauer, A., et al.: 3LGM: Method and Tool to support the management of
heterogeneous hospital information systems. Computers in Medicine, Polish Society of
Medical Informatics, 72–82 (1997)
35. Becker, J., Fischer, R., Janiesch, C.: Optimizing U.S. Health Care Processes - A Case
Study in Business Process Management. In: Proceedings of the 13th Americas Conference
on Information Systems (AMCIS 2007) (2007)
A Model for a Motivational System Grounded
on Value Based Abstract Argumentation
Frameworks
Eugenio Di Tullio1 and Floriana Grasso2

1
Dipartimento di Informatica, University of Bari, Italy
ditullioeugenio@gmail.com
2
Department of Computer Science, University of Liverpool, UK
floriana@liverpool.ac.uk
Abstract. Digital interventions for promoting behavioural changes have

become more and more prevalent, due to the ubiquitous role played by
the Web 2.0. Not many of these programmes are however grounded on
well established theories, both from a psychological and from a dialogue
perspective. We present in this paper a model for incorporating a general
framework for abstract argumentation, into a motivational intervention
based on the Transtheoretical Model of Behaviour Change. A preliminary
implementation of the model as a proof of concept has been carried out
on the domain of healthy eating.
Keywords: Motivation, health promotion, argumentation systems.
1 Introduction
Digital interventions to promote healthy lifestyles are more and more ubiquitous.
[13] offers a detailed overview of Internet based interventions, and conclude that,
as their effect can vary substantially, it is important to identify the parameters
that can contribute to their success. Their analysis seems to suggest that more
successful interventions are strongly grounded on theory, especially if the the-
ory deals with planned behaviour, and that using a variety of techniques which
could impact different stages of the changing process is crucial, as well as the
use of different styles of interactions. In this paper we show how insights from
a relatively new research field, Argument and Computation [10], can be utilised
to augment the design of a motivational system, based on the Transtheoretical
Model of Change. A small prototype has been implemented, in the domain of
healthy eating. This paper, after introducing some relevant background litera-
ture, will discuss the design of the system, and show a walked-through example
to demonstrate the way we envisage this system to work.
2 Background
The Transtheoretical model of Change [9] is a widely accepted theory that at-
tempts to model how people modify their behaviour. The model suggests that

44 E. Di Tullio and F. Grasso
people progress through “stages of change”, from a first precontemplation stage,

when people see no problem with their current behaviour, up to a maintenance
stage, where the new behavior is continued on a regular basis. In each of the
stages, an advisor can use various strategies to foster movement to the next
stage. In particular, when people have to move from precontemplation to con-
templation, a stage where it is clear that there is a problem to address, infor-
mation alone is not sufficient, and it has been argued that appeal to persua-
sion or argumentation could be beneficial [3,11]. Therefore it makes sense to
explore what research fields like persuasive technologies or argument and com-
putation can contribute to the problem specification. In particular, we were
interested in those theories which explicitly model extra-rational factors in the
argumentation process, such as perspectives and values. A recent work [12] in
particular attempts a systematic definition of the concept of “value” and the
way this can be used in pragmatic argumentation to reason about goals, aug-
menting the “Value-Based Argumentation Framework” as defined in [2]. The
system presented in this paper is based on this work, and uses the definitions
of values and perspectives as described in the paper. This section gives a brief
overview of these notions, but we refer to the paper for a more extensive de-
scription.
Perspectives
A state of affair can be evaluated from different points of view, or perspectives.
This allows to express concepts like: from the “health” perspective t is prefer-
able to s, while from a “travel comfort” perspective, the opposite is true. A
perspective is defined by [12] as a pre-order on states in a certain domain,
and is denoted with ≤ p. We use p,q,r to denote perspectives. When s ≤p t
(where s and t are states), we say that t is at least as preferred as s from
perspective p.
The same state of affair can be evaluated from different perspectives. In some
cases it is not known which of two states is preferred from a given perspec-
tive, but one can assume that one perspective is influenced by another. For
example one can assume that perspective p =“Being successful” positive in-
fluences perspective q =“Happiness” thus if a state is preferred from p, it will
be preferred from q too. We denote that a perspective p positively influences
the perspective q with notation p ↑ q and similarly we use p ↓ q for a negative
influence.
Perspectives influencing each other may create influence chains. These chains
can be represented with a directed graph (Fig. 1) where the dashed arrows indi-
cate a negative influence, while the solid ones a positive influence. The left graph
denotes how perspective p positively influences q, which in turn negatively influ-
ences r. It follows that p negatively influences r (right graph).
A Model for a Motivational System 45
Fig. 1. Perspective Influence Chain
Values
Perspectives are subjective points of view and are not necessarily shared by every
agent in the dialogue. The way in which an agent reasons with perspectives is
based on the agent’s set of values and preferences. An agent preference is a
pre-order on states denoted with <α . When s <α t we say that agent α prefers
state t to state s. Therefore agent preferences are perspectives which an agent
considers valid. If one or more preferences maximise a particular perspective,
this perspective is considered a value for that agent. A value is therefore a
perspective maximised from one or more agent’s preferences. Starting from an
agent’s values, it is possible to infer other agent’s preferences, kept implicit in
the user model, by finding those which maximise his values.
Values can also be in an order relationship, to allow for situations in which, for
example, a user may prefer the value Health to Happiness, and thus will choose
a state transition that maximises Health over one that maximises Happiness. A
Value System is a pre-order on values, and is denoted with ≺α . When W ≺α V
we say that agent α prefers to promote the value set V over the value set W .
Transitions may promote or demote a particular value set:
– pro: function pro : S × S → V determines the values promoted by a tran-
sition from a state to another. This function is defined as: pro(s, t) = {v ∈
Vα | t <v s}. When pro(s, t) = V we say that the transition from s to t
promotes V .
– dem: function dem : S × S → V determines the values demoted by a
transition from a state to another. This function is defined as: dem(s, t) =
pro(t, s) = {v ∈ Vα | t <v s}. When dem(s, t) = V we say that the transition
from s to t demotes V .
– Neutral transition: when a transition from s to t neither promotes nor
demotes a value (v ∈ / (dem(s, t) ∪ pro(s, t)) we say that that transition is
neutral for that value.
By means of the value ordering and the functions pro and dem, it is possible to
derive α preferences.
3 System Description
The system’s architecture is shown in Fig. 2. The jVS dialogue manages the
interaction with the user. It is build on top of an ASPIC type dialogue [1], which
is an implementation of a dialogue game based on an abstract argumentation
Fig. 2. System Architecture
framework. The jVS interfaces the dialogue component with the reasoning en-
gine, a set of prolog libraries which reason about the value systems according to
the theory at [12]. The libraries use information from a set of ontologies describ-
ing the stereotypical model of a user in each of the 6 stages of change, according
to the Transtheoretical model, as well as the set of values and perspectives which
can be applied to a state of the world. The system maintains one belief model,
in form of an ontology, for each user of the system, as well as the system’s own.
Aside from domain related notions, the model of values and preferences in the
ontology is shown in Fig. 3. The User Modelling component chooses the appro-
priate stereotype for the user, from information retrieved during the interactions,
on the basis of a widely used Transtheoretical model questionnaire. The Plans
component manages the plans used by the system to build the interaction with
the user in each stage.
The interaction is driven by discourse plans describing the evolution of the
motivational strategy. A plan consists of domain knowledge, eliciting questions
to build the user model of their values, suggestions to the user depending on
which stage the user is in, and motivational expressions to operate some posi-
tive reinforcement when the user succeeds in achieving a goal. The plan content
depends on the stage of change of focus at any one time, as to each stage cor-
responds a different set of goals. For instance in the precontemplation stage one
goal might be raising awareness of the problem the user is facing.
Fig. 3. Value/Preference Ontology
The main drive for deciding what to say is the user’s stage of change. The
discourse plan, on the basis of the different stages of change, identifies appro-
priate argumentation paths in the VS. One of the paths is chosen on the basis
of the user model (taking into account what the user already said and whether
the user had already ’accepted’ some argumentations beforehand). The use of a
VS path allows to detect inconsistencies in the user’s system of values. In this
paper we show the value systems using a very simple, preliminary, formalisation,
not comprising the stage of change user profile, that has mainly the purpose of
better explain how the system manipulates values and how they are connected
together. In the remainder of the section, an example will help understand this
mechanism.
3.1 A Walked through Example

Let us suppose the system contains the states: (i) Eating junk food less than 4
times a month; (ii)Eating junk food between 4 and 8 times a month; (iii) Eating
junk food more than 8 times a month, while the perspectives to evaluate the
states are Healthy eating; Fitness; Health; Social life. Let us also suppose that
the following pre-order applies in, respectively, the system’s set of beliefs (left)
and the user’s set of beliefs (right):
System’s VS User’s VS
Healthy Eating ↑ Fitness -
Fitness ↑ Social life Fitness ↑ Social life
Fitness ↑ Health -
Social life ≺ Health Social life ≺ Health
At the start of the interaction, the user makes no connection between healthy
eating and health, therefore the user justifies some behaviours, like eating too
much junk food, without loss of coherence in his system. Also, the user values
health and social life, with the former preferred over the latter, therefore if the
system succeeds in showing that some behaviour is negatively affecting health,
the user might agree that it is a behaviour that should be modified. Let us
suppose the dialogue is concerned with the transition:
A = Eating junk food more than 8 times a month is preferred to Eating junk food less than 4 times a month
The user has a preference for the transition, while the system considers A a
transition the user should avoid, as the perspectives that A demotes are more
than those which A promotes. But in the user’s model, the user is not “aware”
of all A’s negative impacts. The situation is therefore:
System’s VS User’s VS
Perspectives that the transition A promotes
Social Life Social Life
User’s preferences
Perspectives that the transition A demotes
Healthy Eating
Fitness
Health
Social Life
The argumentation the system may attempt with the user is therefore:
Eating junk food more than 8 times a week, instead of less than 4 times a week,
is not advisable because your Health would be penalised, because A penalises
healthy eating, which in turn promotes fitness, which in turn promotes health.
The system hopes that by integrating these new connections, the user would
be persuaded that the transition is negative. The VS focuses on the health per-
spective, because the user has indicated this is preferred over other perspectives.
If the user accepts the line of reasoning above, the image of the user’s VS will
be modified as in Fig. 4. A screenshot of the system corresponding to one phase
in this dialogue is shown in Fig. 5.
Fig. 4. User and System’s VS after the interaction in the example

Fig. 5. Screenshot of the system for the example
4 Conclusion
This paper provides a practical, although prototypical, demonstration of a mo-

tivational system that uses the Transtheoretical Model [9] and the Value System
[12] to adapt the interaction on the basis of the user’s state of change and on his
personal values.
Other work has been done on digital intervention for motivation, using simi-
lar theories and ideas. [5] presents a motivational system to stop smoking that
interacts with the user via phone text. The system texts users periodically with
messages that encourage to persevere, and that focus on their success so far.
Users can also text the system in case they crave for cigarettes, receiving mes-
sages that help them to carry on with the quitting attempt. The paper proves
that this way to interact with the user considerably increases the rate of success
of the smoke cessation intervention. In [7] the Transtheoretical Model and vari-
ous motivational techniques are used for motivating people to save energy. They
successfully apply well known theories to new contexts, paying much attention
to the system’s capability of understanding which changing stage the user is in.
The system is however still at a theoretical stage. [6] presents a software aimed
at behavioural researchers, where they can build platforms to test their inter-
ventions on real users and process the results. Researchers are provided with
an end-user development environment where they can build their own interfaces
and facilities that will be used by their test subjects, giving them tailored advice.
Similar tools are intended to be used for the system in this paper.
The system presented in this paper has been implemented as a prototype. The
argumentation component is fully implemented, while the strategic component
is still at a design stage. The role of natural dialogue for interacting with the
system has been left for the moment outside the scope of this work. Testing of
the system showed that it is able to provide very simple argumentation, and it
is envisaged that with more complex ontologies and with an interface with a
NL dialogue system it will be possible to obtain more complex and convincing
natural dialogue with the user. A small evaluation is being designed as a Wizard
of Oz experiment [8], and plans are under way to adapt the system so that it
can be interfaced with an embodied conversational agent developed at the first
author’s research lab [4].
References
1. Amgoud, L., et al.: Project N 002307 ASPIC, Deliverable D2.6 - Final re-
view and report on formal argumentation system, available as Technical Re-
port ULCS-07-005, Department of Computer Science, University of Liverpool,
http://www.csc.liv.ac.uk/research/techreports/tr2007/tr07005abs.html
2. Bench-Capon, T.J.M.: Persuasion in Practical Argument Using Value Based Argu-
mentation Frameworks. Journal of Logic and Computation 13(3), 429–448 (2003)
3. Cawsey, A., Grasso, F.: Goals and Attitude Change in Generation: a Case Study
in Health Education. In: Jokinen, K., Maybury, M., Zock, M., Zukerman, I. (eds.)
Proceedings of the ECAI 1996 Workshop on: Gaps and Bridges: New directions in
Planning and NLG, pp. 19–23 (1996)
4. de Rosis, F., Novielli, N., Carofiglio, V., Cavalluzzi, A., De Carolis, B.: User Mod-
eling And Adaptation In Health Promotion Dialogs With An Animated Character.
International Journal of Biomedical Informatics, 514–531 (2006)
5. Free, C., Knight, R., Robertson, S., et al.: Smoking cessation support delivered via
mobile phone text messaging (txt2stop): a single-blind, randomised trial. Lancet
6. Hare, J., Osmond, A., Yang, Y., Wills, G., Weal, M., De Roure, D., Joseph, J.,
Yardley, L.: LifeGuide: A platform for performing web-based behavioural interven-
tions. In: WebSci 2009: Society On-Line, Athens, Greece, March 18-20 (2009)
7. He, H.A., Greenberg, S., Huang, E.M.: One size does not fit all: applying the
transtheoretical model to energy feedback technology design. In: Proceedings of
the 28th International Conference on Human Factors in Computing Systems (CHI
2010), pp. 927–936. ACM, New York (2010)
8. Kelley, J.F.: An iterative design methodology for user-friendly natural language
office information applications. ACM Transactions on Office Information Sys-
tems 2(1), 26–41 (1984)
9. Prochaska, J., Clemente, C.D.: Stages of Change in the Modification of Problem
Behavior. In: Hersen, M., Eisler, R., Miller, P. (eds.) Progress in Behavior Modifi-
cation, vol. 28. Sycamore Publishing Company, Sycamore (1992)
10. Rahwan, I., Simari, G.R.: Argumentation in Artificial Intelligence, 1st edn. Springer
Publishing Company, Incorporated (2009)
11. Reiter, E., Robertson, R., Osman, L.: Types of Knowledge Required to Personalize
Smoking Cessation Letters. In: [25], pp. 389–399
12. van der Weide, T., Dignum, F., Meyer, J., Prakken, H., Vreeswijk, G.: Practical Rea-
soning Using Values. In: McBurney, P., Rahwan, I., Parsons, S., Maudet, N. (eds.)
ArgMAS 2009. LNCS (LNAI), vol. 6057, pp. 79–93. Springer, Heidelberg (2010)
13. Webb, T.L., Joseph, J., Yardley, L., Michie, S.: Using the Internet to promote
health behavior change: A systematic review and meta-analysis of the impact of
theoretical basis, use of behavior change techniques, and mode of delivery on effi-
cacy. Journal of Medical Internet Research 12(1) (2010)
An Assessment of the Potential for Personalization
in Patient Decision Aids
Øystein Eiring and Laura Slaughter
Norwegian Electronic Health Library, Postboks 7004, 0130 Oslo. Oslo University Hospital
oye@helsebiblioteket.no, lauras@idi.ntnu.no
Abstract. An increasing number ofpatient decision aids are being developed to as-
sist patients in making personalized choices among health care options, but little is
known about the current use of and potential for personalization of web-based de-
cision aids. The purpose of this study is to estimate the potential for personalization
of patient decision aids. We developed a coding scheme for personalization and
analyzed web-based patient decision aids from all relevant developers according to
the scheme. The most relevant subgroups of users and the current representation of
the groups in the tools were identified. We then identified system behavior relevant
to web personalization, and instances of adaptive system behaviors.The decision
aids included in this study exhibit four out of five classes of system behavior eligi-
ble for personalization. With few exceptions, the tools do not contain automatic,
adaptive behavior. Patient decision aids hold potential for web personalization.
Relevant techniques are largely unexplored.
1 Background
Patient decision aids are evidence-based tools designed to help people participate in
making specific, deliberated, personalized choices among health care options, in ways
they prefer. According to a systematic review on the effectiveness on patient decision
aids, the tools «differ from usual health education materials because of their detailed,
specific, and personalized focus on options and outcomes for the purpose of preparing
people for decision making»[1].
The term «personalized» is often included in the definitions and declared pur-
poses of patient decision aids. Personalization can pertain to several aspects of a
decision aid, such as the structure and content of the tool, the decision-making
process, and the resulting choice. Within the domain of information and communi-
cation technology, personalization implies the use of technology to accommodate
the differences between individuals. Technological personalization of patient deci-
sion aids could potentially tailor healthcare and healthcare communication by ad-
justing to the different biological, psychosocial and contextual idiosyncrasies of
patients.
Web personalization is the employment of user features in web systems that adapt
their behavior to the user. The overall aim is to meet the needs, goals and preferences of a
variety of people. The adapted content can be variations regardinginformation, products,
52 Ø. Eiring and L. Slaughter
people, services and activities. An exploration of the techniques in the domain of deci-
sion aids could hold potential for more informed, satisfactory, effective and personalized
decisions. A large inventory of techniques for adaptively selecting, structuring and pre-
senting content in web systems to user features (attributes and interaction data) have
evolved during the last two decades [2].
Despite the fact that decision aids are intended to support the personalization of
care, little is known about the current use of, and potential for web personalization
inherent in the tools.
2 Design and Methods
The objective of this mixed-methods study is to estimate the potential for web perso-
nalization in web-based patient decision aids.
Based on a comprehensive anthology [2] we developed a coding scheme for cur-
rent web personalization techniques. The scheme includes a checklist of possible user
features and a catalog of adaptive system behaviors.
We identified developers of web-based decision aids by hand-searching the quali-
ty-assessed patient decision aids in the Ottawa Inventory[3].Developers of decision
aids only available in PDF format were excluded. One decision aid from each devel-
oper was included for further study. When a developer had produced more than one
decision aid, we selected the decision aid that included the richest functionality
present in the developer′s portfolio of tools.
Applying the coding scheme, we identified explicit and implicit subgroups with
comparable user features in the decision aids. Subgroups existent in the tools that
were not present in the generic coding scheme, were added to the scheme. We identi-
fied and described the linguistic representation of subgroups in the decision aids.
We then mapped the system behavior in the decision aids to the classes and prere-
quisites of adaptable system behavior present in the coding scheme. Finally, we sys-
tematically identified user-adaptive behavior present in the decision aids.
3 Results
259 decision aids developed by 22 producers were found in the Ottawa Inventory.
The 10 producers that met the inclusion criteria were responsible for 223 of the deci-
sion aids included in the inventory.
3.1 A Coding Scheme for Personalization of Patient Decision Aids

According to the coding scheme, the basic components of personalization are the media
content, user features, user model construction and representation, and adaptive system
behavior. We analyzed the decision aids according to the four classes. User features can-
broadly be classified into the user´s knowledge level, interests, preferences, goals/tasks,
background, individual traits and context. Adaptive system behaviors include adaptive
An Assessment of the Potential for Personalization in Patient Decision Aids 53
navigation support, adaptive selection, organization and presentation of content, adaptive

search, adaptive collaboration and personalized recommendations. We adjusted and up-
dated the coding scheme to the features identified in the decision aids. In example, we
included somatic parameters as a subclass of individual traits.
Fig. 1. Screenshot of the coding scheme. The subclass individual traits is expanded.
3.2 Content Types

8 out of 10 decision aids included at least two media types. 8 decision aids could be
classified as hypermedia, by definition present when hyperlinks and at least 2 media
types are integrated in a web system.
3.3 Explicit and Implicit Subgroups

The majority of the implicit and explicit subgroups targeted and represented in the
decision aids could be classified in accordance with the user features subclasses in the
coding scheme. The most frequently found subgroups were coping styles, emotional
reactions, cognitive skills, user´s beliefs, experiences of users, literacy level and so-
matic parameters.
Most of the content that could be differentiated according to subgroups pertained
to the somatic parameters of individual patients. The most frequently used somatic
parameters that defined subgroups of users were risk factors, factors relevant for the
eligibility for treatment, incidences, prevalences, and probabilities, outcomes of deci-
sions, etiology, lab test results and prediction of recovery.
One of the decision aids included an evidence summary that summarized the re-
search underlying the decision aid. 53 % of the sentences in the evidence summary
contained information that described different subgroups. This information was, with
few exceptions, not found in the decision aid.
3.4 Representation of Subgroups

The decision aids embedded a number of different strategies to address user groups
with different features. The most frequently applied representations of subgroups
identified in the decision aids are the following:
• Listing several subgroups and making specific statements true for each subgroup
one by one
• Making statements that are untrue or irrelevant to at least one subgroup
• Alluding to subgroups without specifying the attributes of the subgroups
• Giving an average for all subgroups combined
• Suggesting that a patient belongs to one, particular subgroup
• Listing only some subgroups
• Not acknowledging the existence of relevant subgroups
• Asking user to determine the relevant subgroup her-/himself
• Helping the patient determine the relevant subgroup e.g. through an interactive tool
• Describing how health personnel should determine the relevant subgroup
• Giving general information but acknowledging that subgroups do exist
• All decision aids in the study included information that was true and/or relevant
only to subsets of users with particular user features.
3.5 User Model Construction and Representation

Direct or indirect evidence of the application of user models was not found.
3.6 System Behavior and Adaptation
All decision aids presented selected and organized content. 6 of the decision aids
included a search field, but only 1 included a search field that searched specifically
for content included in the tool only. The search fields in the remaining 5 decision
aids searched the content in all of the website. 1 developer included a tailored search
field to selected websites. With 2 exceptions, the decision aids exhibited static naviga-
tion. None of the producers enabled user collaboration. 1 included a user forum di-
rectly related to the tool. 1 decision aid included automatic recommendation of con-
tent based on implicit interaction data. Users could manually enter personal data in 9
of the 10 decision aids, the most frequent feature being the possibility to adjust text
size.
Table 1. Decision aids included in the study
Clinical condition Developer Hyper Adaptive Corpus

media behavior
Treatment options for Cardiff University Yes Basic Closed
early breast cancer and others, UK
Stroke prevention in Healthwise Inc, US Yes Basic Closed
atrial fibrillation
Statins for prevention of Mayo Clinic, US No Basic Closed
cardiovascular disease
H1N1 vaccination CHEO, Canada Yes Basic Closed
Treatment options for Baylor College of Yes Medium Closed
early breast cancer Medicine, US
PSA testing Health Dialog Inc, Yes Basic Closed
US
Long term feeding tube Ottawa Patient De- Yes Basic Closed
placement in elderly cision Aid Research
patients Group, Canada
Mammography screening Public Health Agen- No Basic Closed
cy of Canada
Premixed insulin for type AHRQ, US Yes Basic Closed
2 diabetes
Treatment options for National Cancer Yes Basic Closed
early prostate cancer Institute, US
4 Discussion
Whereas paper-based and linear media can be personalized to a very limited degree,
web-based hypermedia systems can adapt their content and presentation to individual
users employing an array of techniques. The web-based decision aids included in this
study display 4 out of 5 classes of system behavior eligible for personalization. None
of the systems contain advanced adaptive behavior. All the selected decision aids are
closed corpus systems.
The web-based decision aids analyzed in this study target and address several sub-
groups within what is mainly a one-size-fits-all format. A number of strategies are
applied to relate the content to different subgroups, of which some are potentially
misleading, inappropriate, and might entail increased cognitive burden and unneces-
sary uncertainty. A relatively large amount of content could have been differentiated
according to different patient´s somatic parameters.
In the case where the evidence supporting the decision aid was provided, our anal-
ysis indicates that most of the information could be differentiated to individual
patients. This information was only to a limited degree reflected in the decision aid.
As decision aids are developed to support patient´s personalized choices, the gen-
eral absence of personalization in current decision aids is a paradox. According to our
findings, a significant amount of content in both the decision aids and their underlying
evidence base could be tailored to the somatic parameters or other features of individ-
ual subgroups.
In addition to personalized selection and organization of content, system behavior
could adapt to users in a variety of ways. To give an idea of the potential, personaliza-
tion techniques could include:
• Presentation of the content most relevant to the user by priority-on-context tech-
niques such as coloring or scaling
• Supported navigation by augmenting links with annotations that give visual cues,
for instance of progress
• Recommendation of peers with similar features that is treated at the same hospital,
to the patient
• Personalized web search adapted to the health literacy of the user
• Personalized collaboration where users e.g. build a shared list of questions to ask
their caregiver or physician
Fig. 2. In the patient decision aid Treatment choices for men with early prostate cancer, infor-
mation about different tumor stages is presented identically to all users. The presentation could
be personalized utilizing techniques that highlight the tumor stage relevant to the individual
patient. Content fragments could be dimmed, colored, scaled or sorted. Reprinted with permis-
sion from the National Cancer Institute.
5 Conclusion
Existing web personalization techniques constitute a rich resource of developmental

possibilities for various aspects of decision aids that is largely unexplored. By adapt-
ing system behavior already present in current web-based decision aids, the content
and functionality could potentially be tailored to a range of user groups. The time and
effort in using the techniques must be weighted against the possible benefits.
Advanced personalization techniques require the representation of user features in
user models, which is unlikely to be worth the cost when decision aids are stand-alone
applications. User models could potentially be built utilizing input from the electronic
patient journal. Embedding decision aids in patient portals holds promise of enabling
adaptive behavior of decision aids. The advance of semantic web technologies such as
RDF and OWL, and the use of an open corpus knowledgebase, could bring added
possibilities.
References
1. O’Connor, A.M., Bennett, C.L., Stacey, D., Barry, M., Col, N.F., Eden, K.B., Entwistle,
V.A., Fiset, V., Holmes-Rovner, M., Khangura, S., Llewellyn-Thomas, H., Rovner, D.:
Decision aids for people facing health treatment or screening decision. Cochrane Database
Syst. Rev. (3), CD001431 (2009)
2. Brusilovsky, P.: Adaptive Navigation Support. In: Brusilovsky, P., Kobsa, A., Nejdl, W.
(eds.) Adaptive Web 2007. LNCS, vol. 4321, pp. 263–290. Springer, Heidelberg (2007)
3. http://decisionaid.ohri.ca/AZinvent.php (acessed July 20, 2011)
An Analysis of Twitter Messages
in the 2011 Tohoku Earthquake
Son Doan, Bao-Khanh Ho Vo, and Nigel Collier
National Institute of Informatics

Chiyoda-ku, Hitotsubashi, Tokyo, Japan
{doan,khanhvo,collier}@nii.ac.jp
Abstract. Social media such as Facebook and Twitter have proven to be a

useful resource to understand public opinion towards real world events. In this
paper, we investigate over 1.5 million Twitter messages (tweets) for the period
9th March 2011 to 31st May 2011 in order to track awareness and anxiety levels
in the Tokyo metropolitan district to the 2011 Tohoku Earthquake and
subsequent tsunami and nuclear emergencies. These three events were tracked
using both English and Japanese tweets. Preliminary results indicated: 1) close
correspondence between Twitter data and earthquake events, 2) strong
correlation between English and Japanese tweets on the same events, 3) tweets
in the native language play an important roles in early warning, 4) tweets
showed how quickly Japanese people’s anxiety returned to normal levels after
the earthquake event. Several distinctions between English and Japanese tweets
on earthquake events are also discussed. The results suggest that Twitter data
can be used as a useful resource for tracking the public mood of populations
affected by natural disasters as well as an early warning system.
Keywords: Twitter, social media, earthquake, surveillance, natural language

processing.
1 Introduction
Social media such as Facebook and Twitter have proven to be useful resources for
understanding public opinion towards natural disaster events. Such resources can be
used to detect general events in politics, e.g., elections [10], and finance, e.g., stock
market changes [2,10] and oil price changes [10], as well as in alerting disasters such
as earthquakes and typhoons [12]. Other social data such as search queries have been
successfully used in public health to build bio-surveillance systems for early warning
of influenza-like illness [7,8,11], showing high correlations with Centers for Disease
Control and Prevention (CDC) reports. Within the wider Web, the BioCaster project
has worked on detecting and tracking infectious diseases using newswire reports [5].
Twitter, the largest micro-blogging service with about 200 milion users as of March
2011 [1], can generate 200 million tweets a day. Tweets are short but condensed
personal messages with a 140 character limit designed for rapid reporting from
mobile devices. Several applications using Twitter messages in biosurveillance
An Analysis of Twitter Messages in the 2011 Tohoku Earthquake 59
systems have been developed. For example, Flu Detector used Twitter messages to
detect the ILI rate in the United Kingdom [9], and DIZIE, which is part of the
BioCaster project, is an experiemental syndromic surveillance system [6].
The great Tohoku earthquake happened on 11th March 2011 was the most
powerful known earthquake to have hit Japan, and one of the five most powerful
earthquakes in the world overall since modern record-keeping began in 1900 [3]. The
earthquake triggered a tsunami, causing massive loss of life and destruction of
infrastructure, and in turn lead to a number of nuclear accidents in Fukushima
prefecture, affecting hundreds of thousand of residents. This was described as “the
toughest and the most difficult crisis for Japan” by the Japanese Prime Minister [4].
Greater understanding of social responses during such disaster periods should help
metropolitan governments and public health agencies to gain greater insights for
preparedness and response. Twitter data, being real time and large-scale, offers a
unique insight into public opinions as the disaster develops.
In this paper, we analyzed over 1.5 million Twitter messages for the period starting
9th March 2011 until 31st May 2011 – the time when the main crisis happened - in
order to review social attitudes during the time when the earthquake occurred. We
focused on tracking keywords related to three main topics: earthquake/tsunami,
radiation and public anxiety for the Twitter user population in the metropolitan Tokyo
area; an area that experienced severe tremors, social anxiety and mild radiation but no
major loss of life. To gain greater insights into differing attitudes between local and
foreign residents we explicitly differentiated English and Japanese tweets. Our results
show high correlations between Twitter data and real world events as well as how
quickly Japanese people’s anxiety returned to a stable level after the disasters. To the
best of our knowledge, this is the first such study on Twitter data during the 2011
Tohoku earthquake.
2 Methods
2.1 Twitter Corpus
We collected Twitter data for three months, starting from March 9th 2011 to May 31st
2011 using Twitter API (http://dev.twitter.com/) with the geolocation feature set to
track messages originating within Tokyo. The resulting corpus had a total of 48,870
tweets in English and 1,611,753 tweets in Japanese. The details of tweets by dates in
both English and Japanese are depicted in Figure 1. In order to analyze the data, we
wrote a simple parser in Python to parse the text and used relevant keywords for
filtering.
2.2 Earthquake Events and Relevant Keywords
Our empirical analysis focuses on the events during the 2011 Tohoku earthquake.
Within the stream of Twitter messages we studied three indicators of public response:
1) earthquake and tsunami, 2) radiation caused by the Fukushima Daiichi plant’s
60 S. Doan, B.-K.H. Vo, and N. Collier
meltdown, and 3) public anxiety. The first two types of indicators are aimed at
showing people’s awareness of the earthquake, tsunami and radiation and the last
indicators looks at how people in Tokyo are anxious about these events. Essentially,
the events happened as the sequences as follows: The first is the earthquake occurring
at 05:46:23 UTC on Friday, 11th March 2011. The second is the tsunami which
happened after the earthquake a few minutes. The third is the nuclear explosion at the
Fukushima Daiichi plant which the first explosion at reactor 1 happened at 6:36 UTC
th
on 12 March 2011.
Fig. 1. Tweet numbers by dates in English (left) and in Japanese (right)
We manually investigated both English and Japanese tweets and constructed lists
of key terms that are relevant to the events. The lists of English and Japanese terms
indicating earthquake and tsunami, radiation and public anxiety are shown in Table 1.
Although the number of terms might not be comprehensive we believed that they are
good enough for our investigation. The reason we included earthquake and tsunami as
a single event was because we found there were few English tweets about tsunami in
our corpus.
2.3 Data Analysis

Keyword filtering has shown to be a simple, but effective way to filter tweets for
relevant topics [7,8,9,10,11]. In our study, we filtered tweets by event keywords in
Table 1. We normalized tweets per day by dividing the total number of filtered tweets
to the total number of tweets per day, i.e.,
# filtered tweets per day
f(event) = ,
# tweets per day
where f(event) is the relative frequency of events per day.

Table 1. List of relevant keywords for the Earthquake and Tsunami, Radiation, and Anxiety
events
English terms Japanese terms
Earthquake and Tsunami event
大地震大震災
(major earthquake), (great
震災地震
earthquake, quake, quaking, post-quake,
earthquake), (earthquake disaster),
余震揺れ
shake, shaking, shock, aftershock, temblor,
tremor, movement, sway, landslide seismic, (earthquake), (aftershock), (quake/tremor),
seismography, seismometer, seismology, 震度震源マグニチュ
(seismic intensity), (epicenter),
epicenter, tsunami, wave ード津波 (magnitude) (tsunami)
Radiation event
放射放射線
(radiation), 放射能 (radiation ray),
放射性物質
radiation, radioactivity, radioactive, nuclear, (radioactivity), 原発(radioactive material),
東京電力
(nuclear power plant), メルト (TEPCO),
ダウンマイクロシーベルト
power plant, reactor, iodine, TEPCO,
meltdown, sievert, micro sievert, iodine, (meltdown), (micro
isodine, explosion, caesium, strontium, ヨウ素
sievert), イソジンヨウ化
(iodine), (isodine),
plutonium, uranium カリウム炉心溶融
(potassium iodide), (core
爆発
meltdown), (explosion)
Anxiety event
死亡 (death), 死ぬ (die), やばい, やばかった, ヤ
バい, やばっ, やべ (risky; dangerous), 怖い, 怖かっ
た, 怖っ, 恐れ (scary, scared), すごい, すげえ, すげ
die, death, risky, scary, scared, incredible,
ー, すっげー (incredible), びびる, びびった (freaked

freaked out, chaos, evacuate, help, unable to
contact, bad, worrying, worried, anxious,
annoying out), 混乱 (chaos) , 避難 (evacuation), 助けて (help),
連絡とれない (unable to contact), 大変 (bad; oh, my
God), 心配 (worrying), 船酔い (seasick)
3 Results and Discussions
Figure 1 supports Sakaki et al.’s [12] observations that the number of earthquake
tweets increases significantly directly after a major earthquake. The data indicates that
Twitter users would like to broadcast their experience immediately.
Event 1: Earthquake & tsunami event
Earthquake and tsunami keyword frequencies for both English and Japanese are
shown in Figure 2. We noticed that there is a sharp and sudden rise in the number of
tweets immediately preceding the first major tremor. Note that the earthquake
happened at 14:46:23 JST (05:46:23 UTC) on Friday, 11th March 2011 with 9.0
magnitude earthquake near the east coast of Honshu, Japan which was 373 km NE of
Tokyo1.
First, we considered how quickly Twitter users responded to the earthquake. It is
unknown when the first public report about the earthquake was in Tokyo but the first
tweet on the topic originating in Tokyo occurred at 05:48:08 UTC, 1 minute and 25
second right after the earthquake happened at the epicenter. It is unsurprising but
noteworthy that the first tweet was in Japanese.
1
http://earthquake.usgs.gov/earthquakes/recenteqsww/Quakes/usc0001xgp.php
Within our corpus the first English tweets on the earthquake are given below, with the
first two tweets sent from an iPhone:
11-03-2011T05:48:54 Huge earthquake in TK we are affected!
11-03-2011T05:49:01 BIG EARTHQUAKE!!!
11-03-2011T05:50:00 Massive quake in Tokyo
The first Japanese tweets on the earthquake are as follows.
地震！"
" [Earthquake!]
地震だ〜縦揺れ！" [Earthquake ~ vertical shake!]
11-03-2011T05:48:08
11-03-2011T05:48:08 "
11-03-2011T05:48:14 地震！！！！" [Earthquake!!!!]
"
We can easily see that first Twitter users responded very quickly, with the first
English and Japanese tweets occurring about two minutes right after the earthquake
happened. Japanese tweets preceded the English tweets by about 47 seconds. This
might be because the numbers of Japanese language users are far greater than English
language users in Tokyo. We also note that when the earthquake occurred, because of
network outage there was no contact by cell phones but people could still access the
Internet through 3G services with smartphones such as the iPhone.
We note that the first tweet from a Tokyo resident about a tsunami in Tohoku was
a re-tweet at 06:02:35 UTC, 12 minutes after the first tsunami was reported. The first
tweet about a tsunami was an eye witness tweet at 2011-03-11T 05:52:23 UTC, 6
minutes after the earthquake occurred at its epicentre.
11-03-2011T 05:52:23 " オレ、津波の様子見てくるわ！！！！ " [I can see the
tsunami coming!!!!]
Fig. 2. Keyword frequencies for the earthquake event over time for English (left) and Japanese
tweets (right)
Let us consider the details of the aftershocks which were described in Wikipedia2.
“Japan experienced over 900 aftershocks since the earthquake, with about 60
registering over magnitude 6.0 Mw and at least three over 7.0 Mw. A magnitude
7.7 Mw and a 7.9 Mw quake occurred on March 11 and the third one struck offshore
on 7 April with a disputed magnitude”
2
http://en.wikipedia.org/wiki/2011_Tōhoku_earthquake_and_tsunami
“Four days later on April 11, another strong magnitude 6.6 Mw aftershock struck
Fukushima, causing additional damage and killing a total of three people”
As reported above, there are two other significant earthquake events: 7th April and
th
11 April. Both English and Japanese tweets in Figure 1 show how significant they
are since peaks occur on both of those date with frequencies of 0.12 (English) and
0.07 (Japanese) at 7th April and 0.18 (English) and 0.09 (Japanese) at 11th April,
respectively. It seems from our observations that Japanese language speakers were
more concerned on 11th April than 7th April.
The below tweets were the first to show concern about nuclear plants right after the
earthquake.
11-03-2011T05:57:53 " 原発大丈夫かな？
" [Is the nuclear power plant
詳しく福島第原発が穏やかじゃない

okay?]
11-03-2011-03-11T08:43:36 " RT @u_tips: 2
ですね。 " [In detail RT @u_tips: Fukushima Dai-ni power plant is on
福島原発ヤバい状況らしい。。。政府が国民を欺かないことを
alert.]
11-03-2011T09:50:49 "
願います " [The Fukushima plant is in a really bad situation… I hope that
the government won’t deceive the public.]
From drill down analysis we noticed that many people reported the situation
happening in Tokyo from their own personal experiences such as a lack of food in
convenience store on 11th and 12th March.
11-03-2011T11:27:03 People r suggested to prepare an "emergency
kit" consist of blanket, water, canned food, flashlight, aid kit,
clothes #bigearthquakeinjapan
12-03-2011T01:00:18 Wow. I've never seen a convenience store
depleted of food before, even during the Great Handgun quake. At least I
got toilet paper.
It is easy to see that such information could be automatically harvested for timely
planning in future disasters. From Figure 2, we can see that both English and
Japanese tweets correlate closely, reflecting the fact that public concern in both
English and Japanese are the same during the earthquake events.
Event 2: Radiation event

As reported in many public newswires, radiation was one of the main concerns after
the earthquake and tsunami severely damaged the Fukushima Daiichi nuclear plant,
causing three of its reactors to experience a meltdown although this fact was not
confirmed until several months later. The radiation keyword frequencies for both
English and Japanese tweets are depicted in Figure 3.
Following the timeline of the Fukushima Daiichi nuclear disaster 3 , there were
significant events about explosions of reactors as follows:
12th March, 15:36JST: Massive hydrogen explosion on the outer structure of the unit 1.
14th March, 11:01JST: The unit 3 reactor building explodes, injuring six workers.
15th March, 20:00JST: A major part of the fuel in reactor 2 drops to the bottom of the reactor
pressure vessel. Radiation levels at the plant rise significantly but subsequently fall back.
3
http://en.wikipedia.org/wiki/Timeline_of_the_Fukushima_I_nuclear_accidents
Radiation equivalent dose rates of 400 millisieverts per hour (400 mSv/h) are observed at one
location in the vicinity of unit 3.
Figure 3 shows that after the March 11th earthquake, Japanese tweets showed further
peaks on the 12th and 15th March whilst English tweets reached peaks one or two days
later later on the 13th and 17th March, respectively. Although the cause is not clear this
indicates that Japanese people in Tokyo were concerned about radiation earlier than
foreign residents in Tokyo. Once again the results indicate the important role that
aggregated tweets in the native languages play in early warnings.
When the earthquake hit the Fukushima nuclear plant on 11th April, both English and
Japanese tweets reached their peak a day later, on 12th April. This indicates that the
event is of major concern to both Japanese people and foreigner residents in Tokyo.
Fig. 3. Keyword frequencies for the radiation event over time for English (left) and Japanese
tweets (right)
Below are some examples of early English tweets at the peak on 11th and 12th April.
11-04-2011T23:21:09 Earthquakes, tsunamis, radiation what else you
got? Exploding Fuji? Let me get out my folding umbrella... #fb
#quaketrashtalk #backchannel
11-04-2011T23:51:09 This is the "New Normal" life after 3/11.
Always fearing how the situation of nuclear plant is #prayforjapan
12-04-2011T00:08:01 Nuclear Agency Japan has increased the level
of nuclear disaster to level 7 as the worst, which equal to Chernobyl
#Fukushimasradiation
12-04-2011T02:41:40 @<username> 4 of 6 reactors in meltdown. Spent
fuel rods melting. Two explosions, containment breached. Already a 7.
#fukushima #niisa
Topically, several tweets focus on concerns about radiation in relation to tap water in
Tokyo since 13th March.
13-03-2011T07:06:37 People are asked to close window, door; not
to use AC; use mask & not to drink tap water #Fukushimasradiation
23-03-2011T06:18:50 210 becquerel iodine (normal 100 becquerel)
discovered in Tokyo tap water. Infants are urged to avoid drinking it
#Fukushimasradiation
Event 3: Anxiety event

Using keywords with Twitter data can track the public mood such as in elections or
oil price changes [10]. It is interesting to see if they can be used to follow the anxiety
of Tokyo residents during the earthquake. Figure 4 shows anxiety frequencies for
English and Japanese tweets. Since the frequency of English tweets are relatively low
since we had few tweets about anxiety, we included it here as a reference.
Figure 4 shows that the highest frequencies of the anxiety event were at the starting
point -11th March and then decreased over a two week period and kept stable after
that. There was a slight rise in concern on 11th April but the upward change was not
so high when comparing to the earthquake and nuclear events in Japanese. This
perhaps underlines the well reported response about how Japanese people kept calm
during the disasters. Despite the relatively small number of English language tweets,
aggregated message counts from foreign residents show many peaks. However, when
comparing the peaks we can see similar trends with Japanese tweets, especially from
20th April to 11th May.
Fig. 4. Frequencies of the anxiety event by dates of Japanese tweets
Below are some examples for anxiety of people in Tokyo in English tweets,
11-03-2011T06:58:49 nerves frayed out on the streets. everyone
emptied out of buildings, on their phones, worried. some inspecting
damage to buildings
11-03-2011T07:09:56 @<username> I'm okay, thx. Worried about
others...
11-03-2011T07:14:52 @<username> thanks ! Now Me and my family are
ok. But worrying about my men...
4 Conclusions
In this paper, we provided an empirical investigation that showed high correlations

between aggregated tweets and disasters during earthquake events, i.e., earthquake and
tsunami, radiation. It appears that there is strong to potential for tracking public anxiety in
resident populations affected by the disaster. The results reveal that tweets in native
language play an important role in early warning in terms of their volume and timeliness.
Strong correlation between Twitter and public health events leads us to believe that
Twitter data can be a useful resource in an early warning surveillance systems as well as
a tool for analyzing public anxiety and needs during times of disaster.
In the future, we plan to extend our work on analysis to other aspects of the
earthquake using publicly available metrics for evaluation. Automated methods to
find relevant terms for tracking during disasters will also be investigated.
Acknowledgements. Authors would like to thank to Twitter Inc. for providing API
functions to access the Twitter data.
References
1. BBC (2001), http://www.bbc.co.uk/news/business-12889048
(retrieved March 28, 2011)
2. Bollen, J., Mao, H., Zeng, X.: Twitter mood predicts the stock market. Journal of
Computational Science 2(1), 1–8 (2011)
3. CBS News. New USGS number puts Japan quake at 4th largest, March 14. Associated
Press (2011), Archived from the original on (April 5, 2011),
http://www.webcitation.org/5xgjFTgf4 (retrieved March 15, 2011)
4. CNN. Japanese PM: ’Toughest’ crisis since World War II (March 13, 2011), Archived
from the original on (April 12, 2011),
http://edition.cnn.com/2011/WORLD/asiapcf/03/13/japan.quake/
index.html?iref=NS1 (retrieved March 13, 2011)
5. Collier, N., Doan, S., Kawazoe, A., Goodwin, R., Conway, M., Tateno, Y., Ngo, Q.-H.,
Dien, D., Kawtrakul, A., Takeuchi, K., Shigematsu, M., Taniguchi, K.: BioCaster:
detecting public health rumors with a Web-based text mining system. Bioinformatics
(2008), doi:10.1093/bioinformatics/btn534
6. Collier, N., Doan, S.: Syndromic Classification of Twitter Messages. In: Kostkova, P.,
Szomszor, M., Fowler, D. (eds.) eHealth 2011. LNICST, vol. 91, pp. 186–195. Springer,
Heidelberg (2012)
7. Eysenbach, G.: Infodemiology: Tracking Flu-Related Searches on the Web for Syndromic
Surveillance. In: AMIA Annu. Symp. Proc. 2006, pp. 244–248 (2006)
8. Ginsberg, J., Mohebbi, M., Patel, R., Brammer, L., Smolinski, M., Brilliant, L.: Detecting
influenza epidemics using search engine query data. Nature 457(7232), 2012–2014 (2009)
9. Lampos, V., Cristianini, N.: Tracking the flu pandemic by monitoring the social web. In:
Proc. of the 2nd International Workshop on Cognitive Information Processing (CIP), pp.
411–416 (2010)
10. O’Connor, B., Balasubramanyan, R., Routledge, B.R., Smith, N.A.: From tweets to polls:
Linking text sentiment to public opinion time series. In: Cohen, W.W., Gosling, S. (eds.)
Proceedings of the 4th International AAAI Conference on Weblogs and Social Media, pp.
122–129. AAAI Press (2010)
11. Polgreen, P.M., Chen, Y., Pennock, D.M., Nelson, Weinstein, R.: Using internet searches
for influenza surveillance. Clinical Infectious Diseases 47(11), 1443–1448 (2008)
12. Sakaki, T., Okazaki, M., Matsuo, Y.: Earthquake shakes twitter users: realtime event
detection by social sensors. In: Proc. of the 19th International World Wide Web
Conference, Raleigh, NC, USA, pp. 851–860 (2010)
Web-Based Stress Management System Goes Mobile:
Considerations of the Design of a Mobile Phone App
Åsa Smedberg1 and Hélène Sandmark2

1
Department of Computer and Systems Sciences, Stockholm University, Forum 100,
164 40 Kista, Sweden
asasmed@dsv.su.se
2
School of Health, Care and Social Welfare, Division of Public Health Sciences,
Mälardalen University, 721 23 Västerås, Sweden
helene.sandmark@mdh.se
Abstract. High levels of stress at work, great demands and lack of balance
between work and family; these are examples of components in the daily lives
of many people in modern society. Interventions of different kinds are needed.
However, changing patterns of thinking and behaviors is not an easy task, and it
demands for continuous support and learning efforts. This paper presents a
web-based stress management system and the way it was transferred to a
mobile phone app. The solution and considerations are presented. The design
principles of the web-based system were used to examine the use of the mobile
phone app. The results showed that the app to a large extent meets the design
principles, and some of the principles seemed even more applicable in the
mobile phone setting due to its mobility.
Keywords: Mobile phone app, stress management, web-based, design

principles, functions, stress areas.
1 Introduction
Today many people are on sick leave as a result of high levels of stress [1]-[2].
Studies show that high or unclear demands, low control and low social support at
work increase the risk of "job stress". If this state of high levels of stress is prolonged,
the risk of impaired health gets high. There may also be a disturbance in the balance
between work and family when the demands of one sphere do not comply with the
requirements of the other, or when the individual's efforts to fulfill its role in the work
are affected negatively by the demands from family [3]-[4]. Social support at work or
in private life has a mediating effect of perceived stress, and can to some extent,
prevent negative stress.
Prevention and health promotion measures aim at creating attitude and behavioral
changes in people who suffer from stress symptoms. Empowerment is a central
concept for increasing control in life. An important question is how to build on and
reinforce authentic participation, a sense of value and mastery in decision making.
68 Å. Smedberg and H. Sandmark
As for individual efforts in stress management to deal with negative stress

exposure, Jacobson was first to design a program of progressive muscle relaxation to
guide people to achieve relaxation [5]. This can prevent stress reactions. It has been
concluded that the effectiveness of interventions varies according to health-outcome
measures being used [6]. However, to combine different techniques for muscle
relaxation and cognitive-behavioral skills have seemed to be more effective across a
variety of health-outcome measures than using a single technique [6].
When dealing with stress-related health problems the Internet can play an
important role, with its facilities for exchanging information and support, as well as
opportunity to have continuous interactions. Through web-based systems, people with
high levels of stress can more easily come into contact with health professionals and
peers for support and get support on a continuous basis. Self-help groups on the
Internet have evolved in different health areas; stress and coping with stress
constitutes one. There are also various types of web-based intervention systems that
use different types of models for dealing with people's stress. However, available
web-based systems today tend to focus on either information or communication, and
to separate guidance from health experts from social support among peers.
In an attempt to close this gap, we have previously developed a web-based system
that takes into account the benefits of having different actors and types of support
coming together [7]-[9]. The web-based system has now also been transferred to a
mobile phone setting. Since a mobile phone has certain characteristics that differ from
a computer, new conditions follow. The advantage of being available to the user in
most social situations need to be exploited, but also poorer properties such as a
smaller screen and less performance need to be taken into account [10]. This paper
presents considerations as well as preliminary results of the mobile phone app
development.
In next sections, the web-based stress management system and its design principles
are introduced. Thereafter, the mobile phone app is presented. The purpose was to let
the mobile phone app be an extension of the web-based stress management system,
designed in accordance with the principles used for the web-based system (presented
in [7], e.g.). Therefore, we also compare functionality of the mobile phone app and
the web-based system, and examine the mobile phone app based on the design
principles used for the web-based system. The paper ends with some concluding
remarks.
2 The Web-Based Stress Management System
To manage stressful life situations, to develop and maintain a healthy lifestyle and
inner balance, can be difficult. E-health communities can contribute to this process
through its ability to foster continuous interaction between members. Previous studies
have shown that knowledge from health experts and experiences of peers can create a
good synergy; the use of web-based ask-the-expert functions, between citizens and
experts, and conversations among peers in e-health communities has shown to offer
different and complementary support [11]-[12].
Web-Based Stress Management System Goes Mobile 69
2.1 Design Principles

The web-based stress management system is based on eleven design principles
generated through research. This section provides an introduction to these design
principles.
1. Social and Learning Activities. Well-functioning web-based communities foster
social and empathical communication [13]-[14]. At the same time, they need to
support the learning processes regarding stress and how to manage stress. Both social
and learning activities must therefore coexist.
2. Integration of Knowledge and Actors. Health experts and peers possess different
kinds of knowledge and experiences and contribute differently to the system. If
diversity is allowed in the system instead of having actors and knowledge kept
separate, new understanding is more likely produced. To have direct communication
with peers and also with health professionals can create a turning point for people
trying to manage a stressful life situation.
3. Extended Relations. New materials and new ideas can be introduced when
different health experts are invited to online conversations. Also, people with previous
symptoms of stress, who have managed to recover from difficult situations in life, can
contribute with their insights. By this, both strong-tie and weak-tie relationships are
made possible [15].
4. Learning Conversations on Stress. Through questions and conversations about
practical situations, experienced and anticipated difficulties and issues of motivation,
users can get help in their learning processes and become more aware of them [12].
5. Tagged Conversational Topics and Acts. In order to allow users to navigate
easily among conversations, topics and answers could be tagged by content and type
of response. This could also help users to reflect on stress-related issues and to gain
increased awareness of learning processes.
6. Variety Management in Stressful Situation. The system should help the users to
boost their resources and strategies in order to better manage and prevent stress. This
can be done in various ways, through a battery of actions, such as advice on how to
approach a specific stressful situation, and how to filter the environmental signals and
experienced demands, as well as guidance on how to perform physical exercises to
relieve tension.
7. System Diversity and Variety. Individual differences in terms of personal
stressors and reactions [16]-[17] must be handled by the system. Various stress-
related areas can enable people with different needs to navigate between perceived
stress situations, information sources, exercises and discussions.
8. Continuous Learning. Continuous efforts are needed for people who want to
change unhealthy behaviors and patterns of thinking. Therefore, it is important that
the system has mechanisms for ongoing support in everyday situations.
9. Viable Platform. The system in itself should be learning based on the users and
user patterns. By identifying repeated conversation topics and requests from the users,
feedback on what goes on in the users' lives is available, and the system can be
adjusted when necessary. For example, exercises can be adjusted to better fit the
stress symptoms among the users.
10. Multimedia for Virtual and Real Life Integration. Communication between the
system's medical staff and the ones with stress symptoms can be supported by
multimedia applications. Breathing exercises can, for example, be easily
demonstrated through online video clips, and also other physical exercises and
instructions can benefit from using multimedia. The recorded exercises can be
performed by the users whenever they choose. Other types of tools can be used to
support the dialogue and make them richer.
11. Embedded Feedback Mechanisms. The system should include stress
measurements such as physiological markers and self-reporting since they help
measure the users' learning progress. Results from these tests could also be useful to
identify additional needs that the system should meet.
2.2 The Web-Based System Design

The design principles presented in the previous section constitute the basis of our
web-based stress management system. The system is a self-help system for stress
management outgoing from the idea that both stress experts and peers should
contribute with their different skills and experiences. The system provides the user
with information in the form of research and real life stories, practical exercises (both
text and image-based) and opportunities to have conversations with different groups
of participants, with both stress experts (through ask-the-expert function and group
counseling) and peers (through forums and group counseling). As the different active
participants and the various types of support are combined in one system, we call the
system a multiple-help system for stress management [7]-[9]. The stress management
system is also divided into four different stress areas: Sleep, Work and study, Balance
in life (work-study-life balance) and Physical well-being, based on previous research
[3]-[4], [18], [19]. Variety among the users with stress symptoms has to be met by
variety of the system.
3 Transfer to a Mobile Phone App

We now move on to describe the design of the recently developed prototype of a
mobile phone app for stress management. The work involved how to use the basic
ideas and functions in a mobile setting. The technical platform used is Android.
3.1 The Mobile Phone App Functions

When the mobile phone app is started, it displays the five different types of help or
functions of the system together with explanatory text. There is also a button in the
lower corner that provides brief information of the app and the research project. This
home page is seen on next page (Fig. 1). The five types of help, or functions,
available in the system are (from the top): Book counseling session, Research and real
life stories, Ask-the-expert, Exercises and Forum.
Fig. 1. Home page of the mobile phone app for stress management (in Swedish)
The functions work as follows.

Book Counseling Session. The function displays the group counseling sessions
related to the stress area selected via a spin component (android technique). If the user
clicks on one of the counseling sessions, a dialog with information of the chosen
counseling session is shown, and also a button to be used for signing up and another
one for cancelling a session. Another dialog box will then appear in which the user
can save the time and date of the session in the calendar on the phone. The user
cannot see registered counseling sessions in the app but only in the phone calendar.
The user is not able to attend the group sessions via the app; instead, he or she needs
to be online in the web-based system to participate in the registered sessions.
Research and Real Life Stories. The user can choose to either read or listen to the
documented research results and stories told that are related to the stress area selected.
Depending on what options are selected in the tab sliding drawer, data from the web-
based system are filtered and presented accordingly. The app is prepared to present
information as text or played as audio sound. If sound is the option, the audio file is
streamed from the web-based system.
Ask-the-Expert. The user can see the frequently asked questions and answers for
each stress area and also post own questions. To submit questions, the user clicks on a
button for "Ask question" presented in a sliding drawer at the bottom of the page. In a
questionnaire, the user can specify the stress area in question, add a title to the
question and type the question itself. At the moment, the app solution does not allow
the expert’s answer to be presented in the app; instead, the answer is handled through
the web-based system and sent as a personal e-mail to the sender of the question.
Eventually, it is also presented among the frequently asked questions and answers.
Exercises. The user can read, see or listen to the exercises related to the stress area
being selected. As in "Research and real life stories", the app is prepared to support
the formats selected by the user via buttons located in a sliding drawer. If the user
clicks on an item in the list, text is displayed or video and/or sound streamed from the
web-based system.
Forum. The user can actively participate in conversations with peers. As the web-
based system uses a pre-programmed standard forum system, the phone app adds a
link to the forum (via WebView) instead of retrieving the separate messages. Since
the standard forum system, VanillaForums, supports mobile phone apps, the interface
is automatically adjusted. The way it is displayed can be seen below (in Fig.2).
3.2 The Functions in Relation to Stress Areas

When one of the functions has been activated (“Forum”, e.g.), the system will display
a new page with actions available, and also with all the five functions visible through
a tab system (as presented in Fig.2). The five functions have been given the following
tab names: Råd (=Advice): Booking of counseling sessions, FoB (=R&S): Research
and Stories, FAQ: Ask-the-expert, Övn (=Exc): Exercises and practical tools, Forum:
Forum for conversations.
The figure below shows also how the four different stress areas implemented in the
web-based system are displayed. The stress areas are (from the top): Work and study,
Physical well-being, Balance in life and Sleep. By clicking on one of the areas,
conversations on topics related to it can be read and taken part in.
Fig. 2. Page showing the five different functions at the top. “Forum” is activated and forum
discussions in the four stress areas are displayed via WebView (in Swedish).
The forum function of the system is treated differently since it uses a pre-
programmed standardized forum system. The other four functions, however, have
been able to be programmed from scratch and adjusted to the overall phone app
interface. On next page is an illustration of the tab for booking counseling sessions
(Fig.3). The chosen stress area is balance in life. In the text box, information about the
scheduled counseling session is given (title, date, time and maximum participants).
Fig. 3. Page for booking counseling sessions. It illustrates how a session is displayed in the area
dealing with issues related to balance in life (in Swedish).
3.3 Comparison of Functionality

In order to find out about similarities and differences in functionality between the two
systems – the web-based system and the mobile phone app - the main five functions
of the systems were compared (see Table 1).
Table 1. Comparison between the web-based system and the mobile phone app regarding the
main five functions
Function Web-based system Mobile phone app
Research & real life Presentation of stories and Ability to choose between reading
stories research as both text and sound. or listening to selected stories and
research (via ‘sliding drawer’).
Practical tools, Presentation as text, sound and Presentation as for Research &
exercises video clips. real life stories. Video clips are
also available.
Ask-the-expert Frequently questions and answers Frequently questions and answers
are presented. The user can post are presented. The user can post
new questions (and receive new questions (and read personal
personal answers through e-mail). answers through e-mail).
Group counseling Information on upcoming Information on upcoming
counseling sessions are displayed, counseling sessions are displayed,
and the user can sign up for one or and the user can sign up for one or
more sessions. Information about more sessions. The user can also
registered sessions are displayed. add the session to his/her phone
The session takes place in a calendar.
certain chat area in the system.
Forum The forum system offers As the web-based system uses a
participation and displays questions pre-programmed standard forum
and answers between peers. system, the phone app offers links
to the forum system (via
WebView).
3.4 Evaluation Based on the Design Principles

In addition to comparing functionality between the two systems, the design principles
of the web-based system were used to conduct a minor evaluation of the mobile
phone app. This is a summary of the evaluation results.
Social and learning activities take place through the app. Linking the app to the
web-based forums and ask-the-expert function sessions allows for communication
among peers, and also with experts. To allow for integration of knowledge and
actors, the app communicates and displays information, exercises, forum discussions,
etc., to and from the web-based system. Thereby, the two systems together let both
health experts and peers contribute in different ways. Expert answers to own posted
questions still need to be worked on technically in order to have them displayed in the
mobile phone. Extended relations are possible in both environments. Through the
app, learning conversations on stress among peers (on practical situations,
difficulties, motivation, etc.) are possible to follow and participate in. The question of
tagged conversational topics and acts is to be dealt with by the web-based system
(standard forum system). A battery of actions are available that allows the user to be
better equipped for dealing with stressful situations, and thereby increasing his or her
variety management in stressful situations. This is also related to empowerment
among the users. Also, individual differences are handled by the system through its
stress areas (the same as in the web-based system) and its different types of help,
allowing for system diversity and variety. Furthermore, continuous learning is
something even more emphasized and evident in the mobile phone setting, where
ongoing support in everyday situations is possible. The app is strongly related to the
web-based system and should be adjusted to changes in the web-based setting,
thereby allowing for a viable platform. The app is designed to be prepared for
addition of stress areas, e.g. The app uses multimedia for virtual and real life
integration when demonstrating physical exercises, such as breathing techniques
through online video clips. The mobile phone setting allows the user to view exercises
independent of both time and place. So far, embedded feedback mechanisms, in the
form of stress measurements, are to be handled by the web-based system and will
eventually be available in the phone app.
4 Final Remarks
In this paper, we have presented a prototype of a mobile phone app for stress
management associated with a web-based system. So far, the mobile phone app has
shown promising results when evaluated by criteria from research in the area of stress
management and online support. This preliminary work will be followed by user-
based evaluations to identify needs to be addressed in the next iteration of design.
References
1. Henderson, M., Glozier, N., Holland, E.K.: Long Term Sickness Absence is Caused by
Common Conditions and Needs Managing. BMJ. 330, 802–803 (2005)
2. Dekkers-Sánchez, P.M., Hoving, J.L., Sluiter, J.K., Frings-Dresen, M.H.: Factors
Associated with Long-Term Sick Leave in Sick-Listed Employees: A Systematic Review.
Occup. Environ. Med. 65, 153–157 (2008)
3. Sandmark, H.: Work and Family: Associations with Long Term Sick-Listing in Swedish
Women. BMC Public Health 7, 287 (2007)
4. Sandmark, H.: Job Mismatching, Unequal Opportunities and Long-Term Sickness
Absence in Female White Collar Workers in Sweden. Scand. J. Public Health. 37, 43–49
(2009)
5. Jacobson, E.: Progressive Relaxation, 2nd edn. University of Chicago, Chicago (1938)
6. Murphy, L.R.: Stress Management in Work Settings: A Critical Review of the Health
Effects. American Journal of Health Promotion 11, 112–135 (1996)
7. Smedberg, Å., Sandmark, H.: Multiple Help Online: An Integrated E-Health System for
Stress Management. In: Proceedings of IADIS International Conference E-Health,
Freiburg, Germany, July 29-31, pp. 151–158 (2010) ISBN: 978-972-8939-16-8
8. Smedberg, Å., Sandmark, H.: Stress Intervention Online - Designing for Self-Help through
Multiple Help. In: Proceedings of the Third International Conference on eHealth,
Telemedicine, and Social Medicine, Guadeloupe, France, February 23-28, pp. 120–125
(2011) ISBN: 978-1-61208-003-1
9. Smedberg, Å., Sandmark, H.: Dynamic Stress Management: Self-Help through Holistic
System Design. In: Smedberg, Å. (ed.) E-Health Communities and Online Self-Help
Groups: Applications and Usage, pp. 136–154. IGI Global (2012)
10. Jones, M., Marsden, G.: Mobile Interaction Design, 1st edn. Wiley-VCH, Chichester
(2005)
11. Smedberg, Å.: To Design Holistic Health Service Systems on the Internet. Proceedings of
World Academy of Science, Engineering and Technology, 311–317 (November 2007)
12. Smedberg, Å.: Online Communities and Learning for Health - The Use of Online Health
Communities and Online Expertise for People with Established Bad Habits. Doctoral
Thesis. Department of Computer and Systems Sciences, Stockholm University (2008)
ISBN: 978-91-7155-689-9
13. Preece, J.: Empathic Communities: Balancing Emotional and Factual Communication.
Interacting with Computers 12, 63–77 (1999)
14. Preece, J.: Online Communities – Designing Usability, Supporting Sociability. Wiley &
Sons (2000)
15. Haythornthwaite, C.: Social Networks and Online Community. In: Joinson, A., McKenna,
K., Postmes, T., Reips, U.-D. (eds.) The Oxford Handbook of Internet Psychology, pp.
121–137. Oxford University Press, United Kingdom (2007)
16. McEwen, B.S.: The Neurobiology of Stress: From Serendipity to Clinical Relevance.
Brain Res. 886, 172–189 (2000)
17. Levi, L.: Working Life and Mental Health - A Challenge to Psychiatry? World Psych. 4(1),
53–57 (2005)
18. Siegrist, J., Marmot, M.: Health inequalities and the psychosocial environment—two
scientific challenges. Soc. Sci. Med. 58, 1463–1473 (2004)
19. Preckel, D., von Känel, R., Kudielka, B.M., Fischer, J.E.: Over Commitment to Work is
Associated with Vital Exhaustion. Int. Arch. Occup. Environ. Health 78, 117–122 (2005)
Use of SMS for Tsunami Early Warnings
at a Table Top Exercise*
Vincenzo Lagani, Vasilis Kontogiannis,

Panagiotis Argyropaidas, and Catherine Chronaki
Foundation for Research and Technology – Hellas (FORTH)

N. Plastira 100 Vassilika Vouton, GR-700 13 Heraklion, Crete, Greece
{vlagani,vasilik,parg,chronaki}@ics.forth.gr
Abstract. Tsunamis are one of the most dangerous and destructive natural
disasters: countries that experience a tsunami event are likely to undergo, after
the immediate destruction of the regions nearby the coast, several secondary
effects, for example epidemic outbreaks. Thus, early warning systems able to
timely advise the authorities and the population of an imminent tsunami are
valuable tools that should be implemented in any coastal region with high risk
of seismic events. In this paper we present the experience that we gained during
a Table Top exercise aimed at testing the effectiveness of an SMS – based
Tsunami Early Warning System (TEWS). SMS showed to be a valuable
additional support channel for propagating the alarm, even though some
drawbacks were identified e.g. lack of reliability, formal procedures and follow-
through training that must be carefully taken in account. Experience gained will
be used in an operational exercise scheduled October 2011 and potentially pave
the way for including SMS early warning in disaster management.
Keywords: Early Warning System, SMS, Tsunami, Disaster Management,

Emergency Response.
1 Introduction
Recently, two extraordinary Tsunami events (Indian Ocean, 2004 and Pacific Coast of
Japan, 2011) remind the entire world of the potential destructiveness of natural
disasters. Tsunami consequences do not run out in the immediate term; on the
contrary, long term effects of Tsunamis are even more devastating than the brute
destruction spread by the giant waves. In particular, Tsunamis often create unsanitary
health conditions that are likely to breed serious diseases and epidemics. Currently,
the only countermeasures for contrasting Tsunami events consist in (a) developing
emergency programs for population evacuation and (b) creating Tsunami Early
Warning Systems (TEWS) that are able to timely detect the arrival of a Tsunami for
*
This work was supported by the European Commission, Humanitarian Aid and Civil
Protection Directorate under contract 070401/2009/534360/SUB/A3: “POSEIDON:
Earthquake followed by Tsunami in the Mediterranean Sea”.
Use of SMS for Tsunami Early Warnings at a Table Top Exercise 77
advising both the authorities and the civil population to take action in view of the
imminent risk.
Real world, operating TEWS have been deployed in several countries after 2004,
for example in India [1] and Indonesia [2]. Interestingly, most recent TEWS started to
employ SMS services for spreading alert messages; however, using SMS for
broadcasting disaster warnings is still a controversial issue. Thanks to the wide
diffusion of mobile phones, SMS can easily reach a large number of recipients.
Recent studies have noted the possibilities of using SMS and MMS to provide
personalized, media-rich, context-aware advice [3,4] and enhance the
understandability of the alert for visual/hearing impaired citizens [5]. However, SMS
technology is intrinsically unreliable, and sending large numbers of SMS immediately
after a disaster may contribute to overloading the communication network [6].
In this framework, our research group along with National Civil Protection
Authorities, the Red Cross, and the regional Government, recently engaged in a new
initiative, the POSEIDON project [7], with the aim of conducting a European civil
protection exercise to train in the European Civil Protection Mechanism that includes
use of Information and Communication technology in population evacuation in case
of Tsunami. The operational field exercise will be held in October 2011, in the island
of Crete. For our simulation we plan to employ SMS as supplementary
communication channel for advising both competent authorities and the general
population about the imminent Tsunami. One of the preliminary activities of the
project consisted in performing a Table Top exercise (May 2011), with the objectives
of both strengthening the connections among the actors involved and refining the
organization of the simulation. We sent, through SMS, a Tsunami alert to all the
participants of the Table Top exercise that accepted to be involved in this first
experimentation. After sending the SMS, we collected the impression and the
comments of the recipients with a questionnaire prepared in advance. Our scope was
to exploit this first experience in planning the SMS alert system for the field exercise.
2 Methodology
Setting of the TEWS Simulation during the Table Top Exercise: The tabletop
exercise was hosted in the main building of the Decentralized Administration of
Crete, on May 30, 2011. Close to one hundred (100) Representatives of Health
Authorities, Civil Protection, Firefighters, Police, Port Authorities and Cretan
Municipalities were present. Among the Table Top participants, eighty–four (84)
accepted to participate in the simulation: the participants’ role consisted in receiving
two alert SMS on their mobile phones and subsequently answering a questionnaire.
The first message contained a Tsunami alert for the Municipality of Heraklion, while
the second one for the Municipality of Chania. Both messages contained a location –
dependent hyperlink pointing at a map of the coastal area at risk. Messages were sent
either in English or Greek, depending by the nationality of the recipient. The
prototype .NET application for sending the messages, consisting in a Graphical User
Interface connected to the Skype software via an application programming interface
that allowed maintaining the list of contacts, their role, and language preferences.
78 V. Lagani et al.
Evaluating the Results from the Table Top Exercise: We measured the
performance of the SMS service under two different aspects: (a) technical
effectiveness, consisting in evaluating whether the SMS service succeeded in timely
delivering all planned messages, and quantified as the percentage of sent SMS over
the number of planned messages (%SENT) and the percentage of participants that
actually received the SMS over the number of intended recipients (%DEL); (b)
informativeness, much more complex to measure, since it involves the interaction
between the reader and the message itself. For evaluating SMS informativeness we
devised an ad-hoc questionnaire subdivided in the following sections (dimensions):
understandability, credibility, usability and usefulness. Question are structured as five
points items, ordered from “totally negative” (score: 1) to “totally positive” (score: 5).
The questionnaire is available online at: http://kwiksurveys.com/?u=poseidon
3 Results
Out of one–hundred sixty eight (168) planned messages (eighty–four (84) recipients,
two messages each), we managed to send one–hundred nine SMS (%SENT = 64.9%)
to sixty two (62) distinct recipients. At the end of the table-top exercise, we collected
twenty eight (28) filled and four (4) blank questionnaires (thirty two (32) in total).
We estimated the percentage of delivered messages as %DEL = 28/32 = 87.5%. After
the questionnaires collection, we evaluated the results as follows: for each dimension
D we calculated a normalized score as the average score of the questions associated to
D. We then calculate the mean of the normalized scores across all questionnaires.
This procedure produced a single, mean normalized score for each dimension, ranging
from 1 to 5, where “1” indicates that the participants’ average evaluation of the
message was completely negative, while “5 “corresponds to a totally positive
assessment. In particular, the score for each dimension are the following:
Understandability, 3.66; Credibility, 3.58; Usability, 3.67; Usefulness, 2.26.
4 Discussion and Conclusion

The scope of our first experience in using SMS for TEWS during a Table-Top
exercise consisted in detecting possible technical and organizational problems to be
solved before the wider and more complex POSEIDON field exercise. From this
perspective, our simulation was a success: the outcomes of the exercise pointed out a
number of unexpected problems that otherwise would have never been identified. It
was confirmed that exercises are the best way to prepare for the unexpected.
The first difficulty that we experienced was a malfunctioning of the Skype
software; due to this blatant technical problem, our simulation plan was substantially
subverted. This inconvenient taught us that SMS technology in general is not reliable,
even for sending a few hundreds of messages, especially if an external application
(Skype) is used as SMS gateway. However, SMS unreliability does not only reside
into the application employed for sending the messages: any part of the technical
infrastructure used for transmitting the SMS can fail. Despite SMS low reliability, the
results of our simulation also underlined that SMS can still be a useful media for
diffusing Tsunami alerts. The participants generally rank as “sufficient” to “good” the
Use of SMS for Tsunami Early Warnings at a Table Top Exercise 79
Understandability, Credibility and Usefulness of the message. This result suggests

that, once the message is received, it can be an effective tool for informing the
recipient about the imminent risk of Tsunami. The only dimension that did not pass
the threshold of sufficiency was “Usefulness”, the questionnaire dimension measuring
the helpfulness of the accessory information provided by the hyperlinks; we
hypothesize that the difficulties in accessing the hyperlinks may be due to a lack of
training in processing and acting upon the information in the SMS message. Thus, the
results of the Table Top exercise suggested us the following actions:
− the reliability of the SMS service must be improved, at least as far as concern the
SMS gateway;
− the content of the message must be improved, especially regarding the
accessibility of the multimedia information;
− participants (i.e. civil workers) and the public (should the service be expanded)
need to be prepared and trained to receive and process SMS information;
− Alternative means of communicating the information in the SMS message need
to be explored in tandem. SMS messages alone are not adequate for public
awareness and informed response to disasters.
In synthesis, this experience confirmed that SMS messages seem to be a valuable tool
for diffusing Tsunami alerts. However, the low reliability of SMS technology
indicates that SMS can only be employed as a complementary medium along with
other communication and awareness methods. Civil protection workers and the public
need practical training on action upon SMS receipt through customized profiles that
are included in preparedness exercises, to cultivate trust and improve effectiveness of
response and follow-through based on the individual role and skills.
References
1. Kumar, T.S., Kumar, C.P., Nayak, S.: Performance of the Indian Tsunami Early Warning
System. In: International Archives of the Photogrammetry, Remote Sensing and Spatial
Information Science, Kyoto, Japan, vol. XXXVIII, part 8 (2010)
2. Lauterjung, J.: Installation of a tsunami early warning system in the Indian Ocean. In:
Proceedings of the 13th Annual ACM International Workshop on Geographic Information
Systems (GIS 2005), p. 1. ACM, New York (2005)
3. Meissen, U., Voisard, A.: Increasing the Effectiveness of Early Warning via Context-
aware Alerting. In: Proceedings of ISCRAM Conference, Washington, DC, USA (May
2008)
4. Malizia, A., et al.: CAP-ONES: An Emergency Notification System for all. In: Proceedings of
the 6th International ISCRAM Conference, Gothenburg, Sweden (May 2009)
5. Mitchell, H., Johnson, J., La Force, S.: Wireless Emergency Alerts: An Accessibility
Study. In: Proceedings of the 7th International ISCRAM Conference, Seattle, USA (May
2010)
6. Pau, L.F., Simonsen, P.: Emergency messaging to general public via public wireless
networks. In: Proceedings of ISCRAM Conference, Washington, DC, USA (2008)
7. Evangelia, T., et al.: Risk Assessment Study based of the 365AD Earthquake to Drive a
Civil Protection Exercise. ERCIM News (April 2010), http://ercim-
news.ercim.eu/en81/special/risk-assessment-study (visited on July
11, 2011)
Seamless Evaluation of Interactive Digital Storytelling
Games: Edugames4All
Patty Kostkova
City eHealth Research Centre

City University London
patty@soi.city.ac.uk
Abstract. Interactive Digital Storytelling (IDS) is a growing hybrid discipline

bringing together computer games and cinematic storytelling enabling
entertaining education. CeRC developed IDS game “Edugame4All” to improve
children understanding of hygiene and responsible antibiotic use. This short
paper presents an evaluation results assessing knowledge gain against the
project prescribed learning objectives and proposes a novel framework called
“seamless evaluation” inserting fine-grain questions into the story narrative and
introducing a debrief session to evaluate without decreasing players’
immersion.
Keywords: educational web games, game evaluation, interactive storytelling.
1 Introduction
Serious games are increasingly becoming established as a new method of education
[1]. IDS is a growing hybrid discipline bringing together computer games and
cinematic storytelling enabling a creation of unique learning applications [2]. The
positive use of games to aid learning has been established [3-4]. The Edugames4All
project (www.edugames4all.org) based on the DG SANCO-funded e-Bug project
(www.e-bug.eu) developed a story-telling game to teach children basic principles of
hygiene and antibiotic resistance [5]. As a traditional ‘role-playing game’ it relies
heavily on narrative [2]. There were 3 missions implemented, each teaching a set of
learning objectives (LOs) prescrived by the project. In this paper we discuss the
evaluation results and propose a seamless evaluation utilizing the storytelling nature
of the game and engaging the player in a dialogue with the characters allowing to
fine-tune the understanding of the player while not disturbing the immersion.
2 Interactive Story Telling and Edugames4All

Unlike linear digital storytelling, an interactive storytelling makes the player “feel”
he/she is changing the course of the game. As defined by O’Hagan, the diagram in
Figure 1 illustrates the 4 levels of the game.
2.1 Edugames4All: Case Study

The IDS mechanics was chosen based on focus groups conducted with 13-15 years
old children in the UK [6]. There are three puzzles to solve, each testing several LOs.
Seamless Evaluation of Interactive Digital Storytelling Games: Edugames4All 81
Here we describe the first puzzle based around hand hygiene related LOs (the 2nd and
3rd teach appropriate use of antibiotics and the issue of antibiotic resistance).
Referring to standard IDS levels, defined in [1], illustrated on Figure 1, in the
“animation base”, the player is presented with a scene, chooses narrative to talk to
characters, collects evidence, investigates evidence in a laboratory and presents an
answer to “puzzle” – the cause of infection or reason for an illness – to the boss of e-
Bug (“Big C” character, Figure 2). At the second “dialog base”, there is the story. For
example, the story in the first puzzle is as follows: at a BBQ party, a famous actor
gets by infection transmitted as a result of poor hand hygiene at a bathroom but
poisoning food and insufficient hygiene in the kitchen must be eliminated by
investigating a Chef and by collecting and testing evidence samples. Investigate
dialogue in the kitchen scene is illustrated on Figure 3.
Fig. 1. 4 levels of interactive story-telling (taken from [1])
Fig. 2. Big C Fig. 3. Chef investigation narrative Fig. 4. “Micro-vision”
The game mechanics “rules” require the player to test the samples and prove the
hypothesis. This includes so called “micro-vision” (illustrated on Figure 4) turning the
room into a “microbe-vision light” allowing users to see microbes on the scene and
collect samples for testing in a microbiological laboratory. The evidence is tested and
hypothesis about the cause of the infection either proved or disproved. Finally, on the
4th level - “plot outline” - the story “Big C at the Investigators HQ assigns player a
task to investigate the reason of poisoning and approves the correct investigation
outcome: the Actor’s sickness was caused by a contamination of the salad at the BBQ
by microbes transmitted due to poor hand hygiene at a toilet.
3 Evaluation Results
The games were evaluated during the development phase in schools in the UK in
terms of knowledge change and usability. Final game was evaluated in summer 2010
82
Table 1. Summary of player responses for questionnaires for each Mission, including the specific questions provided at each level corresponding
to e-Bug pack learning outcomes. The pre and post responses are written via symbols between the ‘&’ where C=correct, W=wrong, ?=other
(either answer not provided or both True and False selected – wrong in either case). For example, ‘C&C’ = correct pre and correct post answers;
‘C&(W|?)’ = correct pre-answer and either wrong post-answer or other response selected. ‘Remaining’ refers to the tallied player responses for
the following cases: ‘?&?’ , ‘?&W’ , and ‘W&?.’
Mission Question # Kids Started # Kids Finished C&C W&W (W|?)&C C&(W|?) Remaining McNemar chi-square p value
16
1. Viruses and bacteria are two types of micro-organisms. 22 10 1 4 1 27 1.25 0.26
P. Kostkova
18
2. If you cannot see a microbe in plain daylight, it is not there. 22 10 0 3 1 27 0.56 0.45
17
3. Fungi are microbes. 22 10 2 3 0 27 2.08 0.15
18
4. Virus microbes can be spread by sneezing or coughing. 22 10 0 2 2 27 0.06 0.80
19
5. Virus microbes can be spread by sneezing or coughing. 22 10 0 0 3 27 4.08 0.04
Mission 1
20
6. Virus microbes can spread if people don’t wash their hands after using the loo. 22 10 0 1 1 27 0.13 0.72
19
7. Washing vegetables and cooking meat removes bad microbes from our food. 22 10 1 1 1 27 0.13 0.72
21
8. People should wash their hands before eating. 22 10 0 0 1 27 2.25 0.13
20
9. Before making a meal, people don’t need to wash their hands. 22 10 1 0 1 27 2.25 0.13
19
10. Microbes are found on most surfaces like your skin, or food that you eat. 22 10 0 2 1 27 0.08 0.77
10
1. Antibiotics kill bacteria. 14 10 1 1 2 35 0.75 0.39
2. Microbes cannot be spread by sneezing or coughing. 14 10 6 0 2 6 35 2.53 0.11

11
3. Vaccines help protect people against some microbes. 14 10 0 0 3 35 4.08 0.04
11
4. Bacteria and viruses are the same. 14 10 0 0 1 37 2.25 0.13
3
5. Antibiotics kill viruses. 14 10 4 5 1 36 2.04 0.15
Mission 2 9
6. Most coughs and colds get better without antibiotics. 14 10 1 3 1 35 0.56 0.45
13
7. If antibiotics are overused, bacteria can mutate to become resistant to the drugs. 14 10 1 0 0 35 N/A N/A
12
8. You should only use antibiotics with your doctor’s permission. 14 10 2 0 0 35 N/A N/A !
13
9. Overuse and abuse of antibiotics leads to antibiotic resistance. 14 10 1 0 0 35 N/A N/A
11
10. Antibiotics are needed to treat bacterial illnesses and to make you feel better. 14 10 1 2 0 35 1.13 0.29
17
1. Microbes are contagious. 22 13 0 3 2 27 0.05 0.82
20
2. If you cannot see a microbe in plain daylight, it is not there. 22 13 0 1 1 27 0.13 0.72
18
3. Bad microbes can spread when you touch something or someone that is sick. 22 13 1 3 0 27 2.08 0.15
19
4. If people wash their hands, they are less likely to get ill. 22 13 0 3 0 27 2.08 0.15
19
5. You should only use antibiotics with your doctor’s permission. 22 13 0 0 3 27 4.08 0.04
Mission 3
0.27
6. You can stop taking antibiotics when you are feeling better. 22 13 7 5 7 3 27 1.23
0.40
7. Finishing your course of antibiotics ensures that all bacteria is out of your system. 22 13 7 6 6 3 27 0.69
N/A
8. You should never give your antibiotics to a friend to use. 22 13 22 0 0 0 27 N/A
9. People should take antibiotics to prevent them from getting sick (like a vaccine). 22 13 12 3 3 4 27 0.32 0.57
0.27
10. Antibiotics kill our good bacteria in our gut. 22 13 10 2 7 3 27 1.23
Seamless Evaluation of Interactive Digital Storytelling Games: Edugames4All 83
by 49 students attending “Technology Camp,” a summer school held at Elstree

School, near Reading, Berkshire. Their age range was 10-13 year olds including both
girls and boys. The study took two 1.5 hour timeslots provided for each of the two
sets of children (divided at the convenience of the summer school counselors). The
results of the evaluation using pre and post questionnaire, administered on paper
before and after children played the game, were analyzed using t-pair test and
McNemar test, as illustrated in Table 1. We observed a large dropout rate (which is
not unusual and was observed at the junior game evaluation [6]). Mission 1 – 22
started, 10 finished Mission 2 – 14 started, 10 finished, Mission 3 – 22 started, 13
finished. Further Several students played more than one mission – hence the higher
starting numbers for Missions 1 and 3. On average (49 students divided into thirds),
approximately 16 students should have played each Mission. The prescribed Learning
Outcomes were mapped to individual Missions as requested by the project, however,
as they were the least common denominator across European science curricula these
proved to be too simple and did not provide a sufficient level of granularity. Table 1
illustrates most kids at this science camp knew correct answers beforehand, thus, we
could measure little significant knowledge change. A negative result was achieved -
P=0.04 for three LOs; however, the knowledge change was from correct to incorrect.
The problem was that the majority of students knew the answers before game playing
as the level of biology and science knowledge at the camp was well above average.
Secondly, the LO set to allow unification of the project aims cross Europe proved too
simple to test a deeper understanding but provided a baseline for a novel finer-grained
seamless evalution.
4 Work in Progress: Seamless Evaluation

In addition to designing the LOs at the right level of depth to better evaluate the
educational impact of games, the administration of tests needs to be "seamless" by
inserting questions in the game mechanics in order to decrease an exposure suspicion
bias while maintaining user immersion and enjoyment. The need for innovative non-
intrusive approach to evaluation has been highlighted [8], however, to actually design
a seamless evaluation framework an appropriate design methods matching the
mechanics of evaluating game are required. In our previous research, a seamless
evaluation of the Edugames4All platform game was achieved by inserting a “How to
be a Millionaire” scenes (player was asked questions in a competitive quiz format)
before and after playing a bacteria fighting session in human body [5]. IDS games are
providing a nature format for seamless evaluation. Further, to enhance the leaving
effect a “debriefing” phase is to be inserted providing reflections on one’s decisions
by observing the game from critical distance [3]. Alex, the “investigator” characters
in Edugames4All, has to return to the Mission HQ and report each decision back to
Big C. These two implementations of seamless evaluation are currently in tested on a
Global Handwashing Day Game [7] and preliminary results will be presented at the
conference.
84 P. Kostkova
5 Conclusion
Computer games have been established as new media for education. We described the
design of the IDS game Edugames4All teaching hygiene and antibiotic resistance and
the results of an evaluation using pre and post questionnaires methodology. A work in
progress method, the seamless evaluation, integrating fine-grained questions into the
game dialogue and introducing a “debrief” session evaluates knowledge gain without
decreasing players’ immersion.
Acknowledgements. Thanks to DG SANCO, David Farrell, L Lazareck, D

Weerasinghe, e-Bug team lead by C McNulty, J Weinberg, D Lecky; and artists N
Lai, S Beveridge.
References
[1] O’Hagan, M.: Video games as a new domain for translation research, Revista
Tradumàtica – Traducció I Tecnologìes de la Informaciŏ I la Comunicaciŏ 05:
Localitzaciŏ de videojocs, ISSN 1578-7559,
http://www.fti.uab.cal/tradumatica/revista
[2] Spieerling, U.: Interactive Digital Storytelling: Towards a Hybrid Conceptual Approach.
In: Proceedings of DiGRA 2005 Conference: Changing Views – Worlds in Play (2005)
[3] Klabbers, J.: The gaming landscape: A taxonomy for classifying games and simulations.
In: Copier, Raessens (eds.) Level up: Digital Games Research Conference, Utrecht Uni.
(2003)
[4] Robertson, J., Collins, T., Kelly, A.: Tales of Adventure Author: Learning through Making
Games. Narrative and Interactive Learning Environments, Edinburgh (2008)
[5] Farrell, D., Kostkova, P., Lazareck, L., Weerasinghe, D.: Developing web games to teach
microbiology. The Journal of Antimicrobial Chemotherapy 66(suppl. 5), v33–v38
[6] Farrell, D., Kostkova, P., Weinberg, J., Lazareck, L., Weerasinghe, D., McNulty, C.A.M.:
Computer games to teach hygiene: An evaluation of the e-Bug junior game. The Journal
of Antimicrobial Chemotherapy 66(suppl. 5), v39–v44
[7] http://www.edugames4all.org/IntegratedCRD.nsf/a2e0dec55e90ea
ac80257914005ad890/c7b7205052d61c578025791400661778?OpenDocu
ment
[8] Vavoula, G., Meek, J., Sharples, M., Lonsdale, P., Rudman, P.: A Lifecycle Approach to
Evaluating MyArtSpace. In: Proc. 4th Int. Workshop of WMUTE 2006, pp. 18–22 (2006)
Patient-Centered Care as a Learning Process
Jan Aidemark and Linda Askenäs
Linneaus University
351 95 VÄXJÖ
Sweden
{Jan.aidemark,Linda.askenas}@Lnu.se
Abstract. The focus of this paper consists of investigations into a strategic

planning framework for information systems in support of patient-centered care
(PCC) processes. The planning perspective that underpins the research includes
learning theories, organizational learning and knowledge management in
general. A brief review of current PCC goals and perspectives is used as a
starting point for an investigation of PCC activities and support system. The
examples of PCC activities are organized as a learning process and presented in
a framework.
Keywords: Patient learning, Information systems, Planning framework,

E-health.
1 Introduction
Patient-centered care (PCC) is an emerging approach to health care, on which a great

deal of hope has been pinned as it indicates lower costs and better care quality (Laine
and Davidoff, 1996). PCC is a break in a traditional disease-oriented model of care
(Epstein, 2007). However, a coherent definition of PCC is still to appear, one recent
effort being given by Robinson et al. (2008, p. 600) according to which “fundamental
characteristics of PCC” include ”a) patient involvement in care and b) the
individualization of patient care”. In the center of this definition lies the interaction
between patient and caregiver in the hope of a better quality of care, increased patient
satisfaction and higher patient adherence (Robinson, et al. 2008, p. 601). This makes
PCC an approach for how to design and perform health care which contains
implications for every aspect of the health care process. In this paper we focus on how
information systems and information technology (IS/IT) could be used to support a
PCC strategy towards health care. Our approach is to build on a learning tradition
within IS/IT and to apply this to the task of planning and designing IS/IT support for
PCC activities and processes. The contribution of this paper is a literature review that
focuses on the connection between pressing PCC issues and current theoretical
frameworks, concept and models. The study evolves into an outline of knowledge-
and learning-oriented planning and design framework for PCC support.
86 J. Aidemark and L. Askenäs
2 Background: Patient-Centered Care (PCC)

A number of possible aspects of the concept of patient-centered care (PCC) have been
identified. A fairly early expression of patient-centered care was made by Sherer et al.
(1993), who defined it as a “[d]esign of patient care wherein institutional resources
and personnel are organized around patients rather than around specialized
departments”. This can be seen as an organizational approach to achieving patient-
centered care. From economic aspects a number of benefits would be accrue from a
patient-centered approach. Charmel and Frampton (2008) argue that a there is a solid
business case for PCC, with great possibilities for lower costs as well as for an
increased market share. Robinson et al. (2008) adopt four PCC perspectives on: 1)
The public policy perspective, focusing on the partnership among practitioners,
patients and their relatives, and on the care decisions that are made, so that these
decisions follow the needs and wants of the patient and enable the patient to take part
in these decisions in an informed and knowledgeable way. 2) The economic
perspective, which focuses on patients as consumers of health care and on their
competence to make informed decisions with respect to various quality characteristics
of the care. Robinson et al. 2008 mention as one example the possibility of direct-to-
consumer marketing of pharmaceuticals, and how this connects to the patient as
decision-maker involved in care decisions. 3) A clinical practice perspective, focusing
on the view of the caregiver and on the changes in how, for example, practitioners
understand, interact with and treat their patients. The PCC approach affects how the
caregiver understands patient needs and tailors the care to be given as well as
generally increasing the humaneness of the care. 4) The patient’s perspective, finally,
focuses on patient priorities, which might include factors like respect, courtesy,
competence, efficiency, decision-making, time for care, availability, accessibility,
wants/needs of care and communication/information. To handle the complexity of
PCC models have been developed of what could be included in a PCC care process.
Steward et al. (1995) propose a model of six interacting components: 1) exploration
of the illness/disease and the patient’s experience of it, 2) understanding the patient as
a whole person, 3) planning care, goals of treatment and areas of responsibilities, 4)
work with prevention and health promotion, 5) creating in the patient-doctor relation a
working therapeutic relationship including power sharing, and 6) a realistic and
effective use of time and money. Building on this model, Aita et al. (2005) suggest a
somewhat generalized and more concise one for PCC. This model contains four
components: 1) the physician perspective, including style, value or medicine
philosophy, 2) the patient perspective, including priorities, values and the philosophy
of health, 3) practice organization, including priorities and culture, and 4) the
community aspect, including culture and priorities.
3 Investigations of a Patient-Centered Care Model
3.1 The Condition for Understanding - What Is Knowledge

The focus in support of care processes is often on fact, information and messages that
are directed towards the patient. The move towards learning/knowledge puts the focus
Patient-Centered Care as a Learning Process 87
on what happens with the facts when they have reached the patient, and on what it
takes for facts to become actions. In the core of this we find the concept of knowledge
and the processes that surround it, like creation, storage, transfer, use and so on, in
short, what a patient understands from the facts presented. This is a process going on
in all the aspects, including measures, planning, choices, actions and evaluations.
Forbat et al. (2009) discuss the nature of PCC and its key feature of patient
involvement in the care process, which means that the patient is part of planning,
decision-making, delivery and evaluation. However, this involvement is not enough,
but the patient must really be engaged in the care process. This requires that the
patient gains a personal understanding of the sickness, and this does not mean just an
information-seeking process. Forbat et al. (2009) argue that patient and care personnel
must be coached into being able to be engaged. The method which Forbat et al.
(2009) suggest builds on study groups of patients and staff that work according to a
Plan Do Study Act (PDSA) method as a support for learning and building personal
understanding.
3.2 Measuring – Gathering Facts
Gathering facts by taking measurements on the patient is central for making plans and
decisions about the continued care process. In PCC this has been translated into a
need of personalized measurements that capture the individual perspective, or, how
the patient perceives his or her situation. The outcome of a measurement might be
different whether a patient or a physician makes the assessment (for example,
Stephens, 1997). In the care of chronic illness special measures for understanding the
long term progress of the individual have been developed. The typical example of this
is the “quality of life” (QoL) concept. Davidson et al. (2004), researchers on heart
failure, argue for the task of assessing patient needs as being central to tailoring the
care for the individual patient. The approach builds on structured measures to capture
this information, including for example patient satisfaction, quality of life and utility.
But the key is how to individualize these measures, by involving the patient and
making him or her take active part in the measurements, and turning it into a long-
term process and not a one-time happening. In their paper the authors argue for a
general framework and for developing specific and quantitative needs assessment
questionnaires within each area. They also point towards a number of problems with
the QoL measure. First, this concept is not properly defined, which makes it
somewhat less reliable even though it often includes factors like physical functions,
psychological processes, social and economic concerns, and spiritual/existential
aspects.
3.3 Planning – Exploring Options for Care

A great deal of what precedes the actual decision and execution of individual care
could of course be seen as care planning. From a decision theory aspect the planning
phase is about generating options which are to be decided on later. An example of this
could be found in a study of a goal-setting support for diabetes patients (Langford et
al., 2007). In this case a framework for a self-management goal cycle has been set as
support for the patients. The central activities include: 1) visits, both personal and
group meetings, also including scheduled or ad hoc telephone calls; 2) goal lists, i.e.
lists of things the patient should achieve; 3) measurements, including one measure for
the likelihood of reaching a goal and one for measuring the ability of the caregiver to
solve problems associated with the goal; 4) a checklist for the caregiver as support
when gathering facts about the situation; and 5) problem-solving activities when a
problem arises about the goal.
3.4 Decision-Making – Making Care Choices
The decision on what care a patient needs is the centerpiece of the care process. A
patient who is well informed and is thus empowered to participate actively in the
decision plays an important part in a patient-centered care model. The preceding
activities build up to this moment, the patient-doctor meeting, facts and measures, the
planning and the learning processes all supporting an empowered patient. Great
advantages are expected from a patient-centered decision including a focus on issues
that are important to the patient, decisions made in line with patient values, the
improvement of patient compliance with and commitment to the care. In the long run
this would also mean a lower total cost for care. Shared decision-making is a concept
advocated in a brief case study (Walker, 2008) as part of a patient-centered care
model. Walker (2008) defines shared decision-making as “[t]he collaboration between
patients and caregivers to choose treatment options in line with patient health plans”.
The patient in the case becomes informed by support like counseling for good
information gathering and understanding, videos with treatment choices and facts in
form of a treatment handbook. Gustavsson et al. (1999) present a study of a support
system that was provided to a group of patients for home use. The patients were
provided with support like information, decision support, and connections to experts
and other patients. The impact of the system was measured with self-reports of their
quality of life and of the frequency and duration of their use of medical services. The
benefits included, for instance, patients spending less time during care visits,
communicating with health providers by phone and experiencing fewer and shorter
hospitalizations.
3.5 Action and Learning: Connecting the Cause and Effect of Self Care
Taking the action that is the actual care is of course the goal of the care process. The
actions can be taken by patients more or less on their own, or be performed by
someone else, maybe a caregiver from the professional organization, or by someone
close to the patient. The focus of PCC is to empower the patient to be able to perform
self-care. Robinson et al. (2008) study PCC from an adherence perspective, defining
the concept as the patient’s efforts to follow health care advice. Adherence is related
to the concept of compliance, defined in Robinson et al. (2008) as a when a person’s
behavior coincides with clinical advice. To use the term adherence reflects more of
PCC, i.e. individual care and patient involvement. Self-management is one approach
to creating patient-centered care. Coleman and Newton (2005) discuss self-

management in the context of chronic care, including a number of aspects: 1) Patient
education and application in real time situations. 2) Providing the patient with
problem-solving skills. 3) Helping the patient to act with efficacy, using the most
appropriate alternatives. 4) Training the patient in recognizing barriers to action and
how to remove these. 5) Performing motivational interviews to support patients to
commit themselves to the care. However, self-management does not work on its own;
it requires that the other aspects of the care process are in line. The patient must be
able to understand the sickness with its social as well as physical consequences, do
measurements, obtain help with goal-setting and perform follow-up and evaluation.
The care process must be integrated with a focus on the patient’s ability to enact the
care. Instructions like drug warning labels form one possible type of support for self-
care, but making instructions into effective tools for learning and acting is not simple.
Webb et al. (2008) give an example of a warning text for prescription drugs. The
findings from a discussion group show that a majority of the participants found that
the texts contained difficult language or were confusing. The participants in the study
requested more actionable texts in the most simple and concise manner possible.
Webb et al. (2008) argue that the level of misunderstandings among patients with
regard to warning texts is high, and that such texts should be developed towards
adopting a patient perspective.
3.6 Keeping Records: Documentation and Evaluation
Evaluation is needed to ensure that PPC approaches promote better health for patients.
Building the databases needed as sources of facts is closely related to record-keeping.
Both evaluation and records are ongoing activities that are performed in each phase of
the care process, as discussed throughout this section.
Evaluations are necessary for making a good case for PCC. For example, Stewart
et al. (2000) made a study of the differences in outcome between cases using and
those not using a patient-centered approach. The outcome of this study showed that
patient-centered communication influences patients’ health because they perceived
that their visit was patient-centered. Stewart et al. (2000) therefore drew the
conclusion that patient-centered practice improved health status and increased the
efficiency of care by reducing diagnostic rests and referrals. For the patient to find
common ground with the physician was especially important. Cassivi et al. (2008)
discuss the problems with measuring by giving an example of a framework in the
thoracic surgical area. The main problem is identified as what is called a void in
quality measures. The problem is one of a duality in the use of measures. The
measures are both used for quality improvement or for economic aspects of care and,
together with these, as means for supporting the right care choices. This is also
reflected in the type of measures, whether they are outcome-oriented or process-
oriented. To fill this void in measures, frameworks for measuring and measures
should be developed on the basis of a patient-centered view. Cassivi et al. (2008), also
point toward the need of building databases of records from the measures and the
importance of these records for future quality improvements. The financial outcome is
also connected with the evaluation and measurement of PCC and its importance for
care institutions. Charmel and Frampton (2008) point to number of economic reasons
for PCC, including reduced length of stay, a lower cost per case, fewer adverse
events, higher employee retention rates, a decrease in malpractice claims and
increased market shares.
3.7 Interactions: The Patient-Doctor Meeting

In the communication processes the exchange between patient and doctor is the key
piece, which has received a fair amount of attention from several perspectives on the
PCC area. This activity is not entirely comparable to the other six above. The patient-
doctor contact is a meeting that in practice is the time and place where many of the
other activities take place. However, what we here focus on is the nature and
condition of the interactions between patient and doctor as such. The nature of this
interaction is a key to patient learning in the sense that it is in the presence and
interaction with another person (i.e. the doctor) that a fact becomes trustworthy and
real to the patient. In this sense it is a central aspect of the process of patient learning.
Culture-competent communication has been discussed as one aspect of the patient-
doctor meeting. One example of a frame for these processes is the cultural awareness
model, a model suggested by Teal and Street (2009) consisting of four elements:
communication repertoire, situational awareness, adaptability, and core cultural
issues. These four elements indicate a set of skills that the caregiver should possess,
and which could form an area for learning and KM activities. It would probably be
even more to the point to have support systems for the caregivers in the current
situation and as support for learning over time. Another example comes from cancer
care, where Epstein and Street (2007) present a model for patient-centered
communication. This model consists of six interconnected functions: fostering healing
relationships, exchanging information, responding to emotions, making decisions,
managing uncertainty, and enabling patient self-management. The book argues for the
connection between good communication and the outcome of the care given.
4 Conclusion: PCC as Learning Process
The aim of this paper has been to investigate how to understand and model a health
care process as a patient learning process for the use as a IS/IT planning approach.
Starting with a tentative model of PCC, we have investigated current practices of PCC
as reported in the literature, and found the model a useful way of understanding the
PCC process. All the seven types of activity are in use today, albeit not in a coherent
or systemic way. The focus of most of the activities is not on the learning aspect as
such, however important part it is. Most of the practices can be seen to deal with more
than one type, but they usually have a focus on one of them. There are clear needs of
development in all these learning processes, and the way they interact over time, all
directed towards an integrated and systemic understanding of this problem area. For
the care organization learning process there has been progress in directions like care
improvement research, but more generally it would be expected that lessons could be
learnt from organizational learning areas. When it comes to community learning,
which takes place on a person-to-person level theories within the social learning area
could be expected to contribute. The picture of the patient learning process seems to
be a very information-oriented and rational one. This is the picture that emerges both
in the practice as we have studied it and in the different situations that we have
reviewed in this paper. This sits well in the planning perspective and a rational world
view of information systems in general. However, this may not be the best way of
understanding how a patient really experiences the process.
Acknowledgement. The project Bridging the Gap 2, Futurum - the academy for
healthcare, Jönköping county council, Sweden.
References
1. Aita, V., McIlvain, H., Backer, E., McVea, K., Crabtree, B.: Patient-centered care and
communication in primary care practice: what is involved? Patient Education and
Counseling 58, 296–304 (2005)
2. Cassivi, S.D., Allen, M.S., Vanderwaerdt, G.D., Ewoldt, L.L., Cordes, M.E., Wigle, D.A.,
Nichols, F.C., Pairolero, P.C., Deschamps, C.: Patient-centered quality indicators for
pulmonary resection. Ann. Thorac Surg. 86, 927–933 (2008)
3. Charmel, P.A., Frampton, S.B.: Building the business case for patient-centered care.
Health Care Financial Management (March 2008)
4. Coleman, M.T., Newton, K.S.: Supporting self-management in patients with chronic
illness. American Family Physician 72(8), 1503–1510 (2005)
5. Davidson, P., Cockburn, J., Daly, J., Sanson Fisher, R.: Patient-centered needs assessment
– Rationale for a psychometric measure for assessing needs in heart failure. Journal of
Cardiovascular Nursing 19(3), 164–171 (2004)
6. Dubé, L.: What’s missing from patient-centered care? Marketing Health Services 23(1),
30–36 (2003)
7. Epstein, R.M., Street Jr., R.L.: Patient-centered communication in cancer care: promoting
healing and reducing suffering. National Cancer Institute, NIH Publication No. 07-6225,
Bethesda, MD (2007)
8. Forbat, L., Cayless, S., Knighting, K., Cornwell, J., Kearney, N.: Engaging patients in
health care: An empirical study of the role of engagement on attitudes and action. Patient
Education and Counseling 74, 84–90 (2009)
9. Gustafsson, D.H., Hawkins, R., Boberg, E., Pingree, S., Serlin, R.E., Graziano, F., Chan,
C.L.: Impact of a patient-centered, computer-based health information/support system -
reliability and validity in a patient population. American Journal of Preventive
Medicine (16,1), 1–9 (1999)
10. Heisig, P., Iske, P.: European knowledge management framework. In: European Guide to
Good Practice in Knowledge Management (Work in in progress, Draft Version 3.0).
CEN/ISSS, Berlin, Amsterdam (2003)
11. Laine, C., Davidoff, F.: Patient-centered medicine. A professional evolution.
JAMA 275(2), 152–156 (1996)
12. Langford, A.T., Sawyer, D.R., Giomio, S., Brownson, C.A., O’Toole, M.L.: Patient-
centered goal setting as a tool to improve diabetes self-management. The Diabetes
Educator 33(suppl. 6), 139s–144s (2007)
13. Robinson, J.H., Callister, L.C., Berry, J.A., Dering, K.A.: Patient-centered care and
adherence: Definitions and applications to improve outcomes. Journal of the American
Academy of Nurse Practitioners 20, 600–607 (2008)
14. Sherer, J.L., Anderson, H.J., Lumsdon, K.: Putting patients first. Hospitals work to define
patient-centered care. Hospital 67(3), 18–23 (1993)
15. Stephens, R.J., Hopwood, P., Girling, D.J., Machin, D.: Randomized trials of quality of
life endpoints: are doctors ratings of patient symptoms interchangeable with patients self-
ratings?! Qual. Life Res. 6, 225–236 (1997)
16. Steward, M., Brown, J.B., Weston, W.W., McWhinney, I.R., McWilliam, C.L., Freman,
T.R.: Patient-Centered medicine: transforming the clinical method, p. 25. Sage, Thousand
Oaks (1995)
17. Stewart, M., Brown, J.B., Donner, A., McWhinney, I.R., Oates, J., Weston, W.W., Jordan,
J.: The Impact of Patient-Centered Care on Outcomes. Journal of Family Practice 49(9),
796–804 (2000)
18. Teal, C.R., Street, R.L.: Critical elements of culturally competent communication in the
medical encounter: a review and a model. Social Science and Medicine 68, 533–543
(2009)
19. Walker, T.: Shared decision making gains recognition as patient-centric care model.
Managed Health Care Executive (April 2008)
20. Webb, J., Davis, T.C., Bernadella, P., Clayman, M.L., Parker, R.M., Adler, D., Wolf,
M.S.: Patient-centered approach for improving, prescription drug warning labels. Patient
Education and Counseling 72, 443–449 (2008)
Wireless Computer Games
and Applications in the Medical Education Curriculum:
Adventures in Pedagogy
Juanita Fernando and Nicolette Peters
Medicine, Nursing & Health Sciences, Monash University

Wellington Road, Monash, Victoria 3800, Australia
{juanita.fernando,nicolette.peters}@monash.edu
Abstract. This manuscript reports work on the first of several related projects
in a series entitled “Adventures in Pedagogy”. Serious Computer Games and
Applications on wireless devices connected to the Internet are commonly
embedded into everyday clinical practice. Yet a review of the literature
indicates a scarcity of such curricula in undergraduate medical education.
Graduates are evidently not well prepared for wireless e-health practice during
University study. Consequently, we have introduced a selective for first year
medical students called “Computer Games and Applications for Health and
Wellbeing”. The selective is designed to support the clinical application of the
wireless tools in an ethical and practical manner while embedding fundamental
IT concepts to help prepare graduates for new practice horizons.
Keywords: medical education, health informatics education, professional

education, wireless technology, telemedicine, telehealth.
1 Introduction
Most clinicians agree that mobile access to Serious Computer Games and
Applications (SCG&A) using wireless devices over the Internet has opened new
practice horizons for health. The SCG&A are designed to facilitate improved health
and wellness outcomes, epidemic intelligence and public health event detection [1].
Many SCG&A tools are already widely used in Australian, if not international,
clinical practice.
Designers of medical education curricula have largely overlooked the SCG&A
health practice milieu [2]. Anecdotal evidence suggests the enormous amount of
information students must learn over their five year undergraduate qualification
militates against explicitly introducing informatics into the medical curriculum [3].
Educators would be well advised to develop a dialectic curriculum that resolves the
current mismatch between ubiquitous SCG&A tools for health and busy medical
education models. We have begun to address the mismatch by introducing a
Selectives unit, “Computer Games and Applications for Health and Wellbeing”, into
the first year undergraduate medical education.
94 J. Fernando and N. Peters
2 Study Design
We rely on a qualitative grounded theory study design to facilitate a research

approach encouraging reflection on the continuous interplay between data collection
and analysis.
2.1 Aim
The aim of this work, the first of several related projects in a series entitled
“Adventures in Pedagogy”, reflects on the process of embedding health informatics
into MBBS curriculum to support graduates for future wireless practice.
2.2 Methods
We use an action research method. Action research is well suited to exploring and
sustaining change processes to established curriculum.
2.3 The Site and Milieu

The Selectives program at Monash University (Victoria, Australia) offers first year
Bachelor of Medicine Bachelor of Surgery (MBBS) students’ non-clinical units to
study during their second semester. The program provides students with opportunities
to develop or acquire new or existing skills outside traditional areas of medical
education. The Selectives program ran for eleven weeks for two hours between July
and October 2011.
2.4 Participants
We are the only participants at this stage of the research program, which has received
University human ethics authorization.
2.5 Evaluation
Students will evaluate the unit in October 2011 using a tool we have developed and
the Student Evaluation of Teaching and Units tool administered by Monash
University (4). Both evaluations will be triangulated with data collected from our
reflections. These data will be analysed interpretively so we may analyze the full
complexity of Selective outcomes.
3 Learning Objectives
Unit learning objectives were informed by several meetings with MBBS candidates of
varying experience over many years, daily interactions with colleagues, familiarity
Wireless Computer Games and Applications in the Medical Education Curriculum 95
with the relevant educational and health informatics literature and professional
expertise [5].
3.1 Course Content

The course content was designed so that the medical students might “dip their toes”
into mobile e-health tools for education, epidemic intelligence and everyday practice.
It incorporates technology skills assessment and training as required, followed by
classes supporting students to devise a suitable evaluation tool. The tool will guide the
students’ reviews and analyses of serious computer games, telehealth/telemedicine
portals, smart phone and tablet applications, social media and 3D applications. The
cohort will be organized into groups divided according to free applications using
tablets or smartphones connected to the Internet from the classroom. At the Selective
conclusion, each group will present their reviews. We hope to publish the student
work collaboratively in the future.
4 Support and Logistics
Multidisciplinary University support for the unit included access to devices and other
resources drawn from Faculty colleagues and those from IT Support and E-Learning
divisions. Expertise from the entire Monash community underpinned much of the
syllabus design.
The multinational Australasian College of Health Informatics (ACHI) membership
made several suggestions for meaningful syllabus inclusions too. ACHI members
often used their own professional networks to support the selective [6]. The authors
drew on a generous range of international expertise to inform the classes on topics
such as the application of telehealth and telemedicine (public health) to African and
transient communities, 3D worlds hosted locally, in New Zealand and the United
Kingdom. This pedagogical adventure would not have been able to proceed without
such direct and collegiate support.
Finally, the Selectives’ focus on personal and professional development provided a
ready-made vehicle for us to offer a new topic, simplifying logistics management.
The unit did not require integration into an already busy curriculum [3]. The
challenge of embedding health and medical informatics syllabus into medical
education more generally is among the subjects of our current research interests.
4.1 Marketing the Selective to Medical Students
The published literature and anecdotal evidence based on research experience from
the biomedical and health informatics arena informed the Selectives title. While many
clinical students may be technologically savvy this does not necessarily translate to
comfort using wireless devices for health in real life [2]. Clinical students often
believe new informatics pathways in medical education are too hard to learn as part of
their degree program (7). Thus the friendly, if a little misleading, term “computer
96 J. Fernando and N. Peters
games”, a Selective component, is embedded in the unit title. To ameliorate this

concern the entire syllabus overview was published to students on the University’s
internal e-learning system.
4.2 Syllabus Overview

Key learning goals for each class were framed to ensure Selective learning objectives
were largely achieved. However we hoped detailed content would largely be driven
by student enrolees. Delivering the Selective was likely to call on the complete depth
and breadth of our informatics and research experience.
5 Conclusion
The introduction of medical education curriculum methodically examining mobile

access to innovative wireless SCG&A tools on the Internet seems a logical response
to new practice horizons for health. We are gratified to be part of an international
vanguard of medical informatics educators involved in enriching health professional
education.
References
1. Health Informatics Society of Australia: Providing Leadership in E-Health (2011),
http://www.hisa.org.au/
2. Otto, A., Kushniruk, A.: Incorporation of Medical Informatics and Information
Technology as Core Components of Undergraduate Medical Education – Time for
Change! Stud. Health Technol. Inform. 143, 62–67 (2009)
3. Confederation of Postgraduate Medical Councils (CPMC): Australian Curriculum
Framework for Junior Doctors (2011), http://curriculum.cpmec.org.au/
4. Monash University. Student Evaluation of Teaching and Units (2011),
http://opq.monash.edu.au/us/surveys/setu/index.html
5. Fernando, J.: Personal web page (2011),
http://users.monash.edu.au/~juanitaf/
6. Australasian College of Health Informatics (ACHI) (2011),
http://www.achi.org.au
7. Berglunda, M., Nilssona, C., Révaya, P., Petersson, G., Nilsson, G.: Nurses’ and Nurse
Students’ Demands of Functions and Usability in a PDA. Int. J. Med. Inform. 76(7),
530–537 (2007)
The Process of Policy Authoring
of Patient-Controlled Privacy Preferences
Thomas Trojer1 , Basel Katt1 , Thomas Schabetsberger2,

Richard Mair2 , and Ruth Breu1
1
Research Group Quality Engineering, University of Innsbruck, Austria
{thomas.trojer,basel.katt,ruth.breu}@uibk.ac.at
2
ITH-icoserve GmbH, Innsbruck, Austria
{thomas.schabetsberger,richard.mair}@ith-icoserve.com
Abstract. Discussions about appropriate security controls to protect

medical records led to the understanding that the patient her-/himself
plays a crucial role in networked electronic health-care. Patients have in-
dividual privacy concerns and may want to execute their personal right
of self-determination on access and usage of their medical records. The
ability for patients to have control over their personal medical data is
the essence of patient-centric networked electronic health-care, but poses
challenges regarding its tool support. Since patients can be generally
treated as non-security experts as well as non-health-care domain ex-
perts, usability-supporting factors of authoring tools for privacy pref-
erences have to receive major attention by implementers. Additionally,
domain characteristics have to influence the design of such authoring
applications. Finally expressed privacy preferences have to be analysed
to inform the patient-author and guide her/him in the policy authoring
process. In this paper we discuss the process of authorization policy au-
thoring for shared electronic health records which we use to implement
patient-controlled access control authoring tools. Further a use-case in
the context of a specific health-care infrastructure is presented.
Keywords: Privacy, Patient privacy policy, Access control, Authoring

tools, Information self-determination, Integrating the Healthcare Enter-
prises (IHE).
1 Introduction
Discussions (e.g., in [9]) about appropriate security controls to protect personal

medical records led to the understanding that the patient her-/himself plays a
crucial role in networked electronic health-care. This is due to the fact that a
patient constitutes the identified individual within a health-record and therefore
processing of such medical data is bound to common data privacy regulations of

This work was partially supported by the Austrian Federal Ministry of Economy as
part of the Laura-Bassi – Living Models for Open Systems – project FFG 822740/QE
LaB, see http://lab.q-e.at/

98 T. Trojer et al.
a country (see e.g., [1]). Further patients have individual privacy concerns and
may want to execute their personal right of self-determination on access and
usage of their medical records. Legal regulations on data privacy and therein
especially information self-determination represent the underpinning motive to
let patients express personal privacy concerns. To meet these regulations, cus-
tomized applications and IT-infrastructure have to be built to make electronic
health-records controllable.
Therein we see a major requirement being usable authoring tools supporting
and guiding patient-authors of privacy preferences during the authoring process.
An important usability-supporting factor during this authoring process is the
analysis of privacy policies. Analysis results are used to inform the patient-
author about quality and effects her/his privacy settings imply on the overall
functionality of the health-care information system. Besides policy analysis, the
integration of health-care domain characteristics and domain knowledge also
support usability. Such integrations try to answer questions like, ”who are the
typical stakeholders within the health-care domain?”, ”which data is involved
and how sensitive is it?” or ”what data is required by those stakeholders?”. In
this paper we present the process of authorization policy authoring for shared
electronic health records and discuss usability-supporting factors involved. The
importance of developing highly usable authoring tools comes from the fact that
patients are not considered security experts and are not necessarily familiar
with working processes of the health-care domain. Therefore patients have to be
supported when trying to express their individual conception of privacy towards
corresponding enforceable privacy policies.
2 Problem Statement
Privacy can be ensured when a consent or agreement on a purpose of use is stated
by the identified individual which gets enforced by the security infrastructure.
Further access restrictions limit usage of data in order to prevent potential dam-
age and misuse. Here we want to specifically emphasize on the necessity that the
explicit source for setting privacy preferences is the corresponding patient.
A problem gets visible when changing the perspective, asking how privacy
policies can be declared in a way so that they match patients’ individual con-
ception of privacy. Since it is not feasible to put in place a trusted party who
manages policies for each patient, a patient by her-/himself should be allowed to
act as the author of privacy settings. An initial requirement to successfully em-
power patients to do so, is to consider usability-supporting factors of the policy
authoring tools. These factors leads to a change in the traditional authoring pro-
cess, which allows only security experts to define security artifacts. Furthermore,
as privacy policy authoring requires health-care domain information, aspects of
integration to an established health-care infrastructure have to be covered at the
same time.
The Process of Policy Authoring of Patient-Controlled Privacy Preferences 99
3 Related Work
Our ongoing work related to patient-controlled access control is based on pro-
posals published by ELGA1 , which is the working group driving the Austrian
e-health initiative [2].
There has been general work published in the field of usable security, e.g., [7].
The authors in this work state that when employing usable applications, guiding
the user in a privacy policy authoring process will lower the risk of inappropriate
use of personal information. In their work they conducted an empirical study to
evaluate the use of tools guiding and not guiding users through the authoring
process. Significant advantages of the employment of guided tools are shown
in their study. This also justifies our effort on implementing patient-controlled
access control policy authoring. Still, our work differs by the use-case within the
health-care domain together with the domain-aware analysis of patient privacy
policies to support the user. Further the authors in [13] evaluated the SPARKLE
policy workbench, an enterprise privacy policy authoring application in order to
gain information on usability challenges. We are able to develop our usability
requirements based on parts of their work, although related to characteristics of
a networked health-care landscape.
A core part of the process of policy authoring we propose in this work is
authorization policy analysis. An analysis component therefore analyses patient
privacy settings and provides feedback to the patient-author. Policy analysis,
similar to what we implemented is covered in [10]. Still, in our work we dynam-
ically retrieve health-care domain characteristics, required to enable domain-
aware analysis.
Katt et al. [8] propose an architecture for enforcing access control in Integrat-
ing the Healthcare Enterprises (IHE) based systems. IHE is also the basis for
our work regarding the retrieval of domain characteristics. Their work can be
used to implement the actual enforcement of patient privacy policies.
4 Authoring of Privacy Preferences

In this section we define the authoring process and discuss usability-supporting
factors of authoring of privacy preferences performed by non-security experts. In
order to allow those privacy preferences to be machine-interpretable and enforce-
able by a security infrastructure we translate them to access control policies.
Fig. 1 describes the activities and artifacts involved in our policy authoring
process. Based on the type or expertise of a user, a set of templates for declaring
privacy settings is gathered and made accessible to the user. After setting pri-
vacy options based on domain information a policy analysis component decides
whether the privacy policy is enforceable or needs to undergo further editing.
From this process we extract factors which play a part in supporting the us-
ability of policy authoring tools. These usability-supporting factors are shown in
1
electronic health-record (German, ”Elektronische Gesundheitsakte”)
100 T. Trojer et al.
Authorization
User interface domain policy model
design
patient
!
privacy policy
Stakeholder view
Stakeholder
view Template
select view select template author policy Policy analysis
policy valid
Domain
characteristics
no yes
Fig. 1. Authorization policy authoring process represented as activity diagram
Fig. 2 as part of our authorization policy authoring framework. A central build-

ing block of this framework is the authorization policy authoring model which
provides entities required to be presented by policy authoring user interfaces
as well as to be translated into patient privacy policies. Entities of this model
correspond to the health-care domain and define authorization aspects. These
authorization aspects are introduced to the model via extending the authoriza-
tion domain policy model (cf. our previous work [14]). The authorization domain
policy model defines e.g., an access target describing the requesting subject and
an action to be executed on an information system resource. Further permis-
sion or restriction entities cover those access targets and further limit access by
additionally providing conditions or obligations to be fulfilled. Based on these
definitions enforceable access control policies can be generated. Fig. 2 indicates
this via the vertical connections between the policy authoring user interfaces and
the patient privacy policy through the authorization policy authoring model.
Below we discuss the usability-supporting factors related to authorization
policy authoring. User interface design and stakeholder views are only concep-
tionally mentioned here, since the actual design of authoring tools as well as
studies on user groups and detailed use-case scenario analysis, respectively, are
out of scope and considered future work.
Authorization
Stakeholder domain policy model
views
Domain
User interface <<extends>>
characteristics
design
Patient
privacy policy
!
Authorization policy
authoring model
Policy authoring Policy

Templates
user interfaces
analysis
Fig. 2. Usability-supporting factors of access control policy authoring

Stakeholder Views. A stakeholder view defines the authoring functionality

which is available for a specific type of user, e.g., a patient or medical professional.
By analysing different use-case scenarios, where access to a shared electronic
health record of a patient is involved, all required stakeholder views can be
identified. A typical use-case scenario regarding a patient is the viewing of her/his
own health record or the definition of trusted medical staff gaining extended
access rights to health records. On the other hand, medical practitioners want
to access the health record of a patient during or after a medical treatment
or want to have the option to protect very critical health records from patient
access (e.g. if laboratory markers suggest to diagnose a severe disease).
Having different views allows us to define multiple customized applications
related to one authorization policy authoring model, expressing the needs of dif-
ferent stakeholders. Further, when developing views according to this model, we
can guarantee that the user interface reflects domain and access control aspects
appropriately.
User Interface (UI) Design. When defining graphical user interfaces human
cognition as well as user behavior during task execution has to be considered. A
usable UI is designed in a way e.g., to show interface elements in a well-placed
(grouped) manner, describes necessary steps to reach a certain goal, gives a user
a history of previous actions taken, lets a user abort (and maybe continue) at
any time and keeps a user informed about the application state. In our case the
state of the application links back to policy analysis as it will be described later
in this section.
Domain Characteristics. Policies employed by a security infrastructure may

be diverse and complex. A patient’s ability to define privacy policies may be in-
fluenced by multiple information sources from within the electronic health-care
network. Auxiliary information have to be used by the policy authoring appli-
cation in order to enable a domain-aware policy authoring process. We identify
the following types of domain-related information and associated attributes to
be integrated:
– Patient, i.e. unique patient identifiers, corresponding health records, related

medical practitioners, etc.
– Medical data, i.e. record identifiers, record types, related stakeholders (e.g.,
creator or identified patient)
– Health-care provider, i.e. working roles, unique identifiers
– Health-care work processes, i.e. record type – working role mappings, types
and purposes of data processing, patient – practitioner relationships (via
medical treatments, referrals or maintained health records), practitioner –
resource needs-to-know [4] relationships, etc.
Domain characteristics represent the core factor to actually implement usable

policy authoring tools. Integration of domain characteristics in the context of an
IHE-based health-care infrastructure is discussed in Section 5.
Templates. Templates are partial instances of the authorization domain policy

model. A template covers a common concept of a health-care working environ-
ment which implies an authorization rule supporting it. Templates are defined via
a label which associates entities of the authorization domain policy model (used
as placeholders) and the actual domain data specified by domain characteristics.
The total amount of available templates defines the instantiation possibilities of
the underlying authorization domain policy model. Therefore templates have to
be based on evaluated use-case scenarios in order to be meaningful and to provide
guidance and an overview of functionality to the user. Finally for each template
different stakeholder views and user-interface designs can be considered.
E.g., the template with the label family practitioner associates an authoriza-
tion domain policy stating that the selected medical practitioner (i.e. a place-
holder for domain data) is permitted (i.e. an instance of the permission entity) to
access all health records (i.e. the actual domain data of type health record). As
another example, a template labeled referral associates a policy which allows a
selected medical practitioner (i.e. the target of the referral) to access (i.e. a stated
permission) a basic set of health records (e.g., defined via record types or via the
(derived) record sensitivity) of a patient. Such templates, as they correspond
to the describing authorization domain policy model can be easily transformed
to enforceable access control policies, e.g., expressed by the Extensible Access
Control Markup Language (XACML) [12].
Policy Analysis. Policy analysis is in general considered a recurring task. By

analyzing policy artifacts, feedback can be provided to inexperienced or non-
expert users. In our context, a patient expressing privacy preferences can be
informed e.g., if certain access rules are in conflict. Further a patient has to be
warned if her/his settings would lead to privacy at risk or interfere with working
processes in the health-care domain, e.g., by restricting access to medical data
where access is required.
By assuming syntactically valid authorization policies, we see two different
types of analysis to be performed:
– Conflict or redundancy detection between policies (see e.g., [11]), i.e. access
control is undecidable or one policy dominates another policy, respectively
– Constraint checking regarding health-care work processes, i.e. evaluating if
e.g., privacy is potentially at risk as no inter-personal relationship between
a patient and a practitioner can be derived
Policy analysis is bridging the gap between patients and security and domain
experts. On one side patients request privacy protection which conforms to their
conception of privacy and on the other side security experts are able to actually
express such preferences in a machine-readable way. Further domain experts can
contribute knowledge about properties for a functioning health-care domain. A
policy analysis component functions herein as a kind of advisory system to a
patient. By reporting issues regarding domain characteristics together with the
authorization policies themselves, policy analysis positively contributes to the
usability of an authoring application.
5 Use-Case: Authoring Tools for IHE IT Infrastructure

In order to establish authorization policy authoring tools in the context of a
health-care network infrastructure, we have to identify the required interfaces
for retrieving domain characteristics and build upon a specific access control
framework. The Austrian ELGA proposes the use of IHE-based systems together
with authorization enforcement via an XACML infrastructure. Fig. 3 shows the
IHE-profiles which the authoring applicaton integrates, as well as XACML secu-
rity components it relates to. Detailed descriptions of the security components,
their interaction and access control mechanisms are found in [8,12,6] and out of
scope of this specific work.
IHE ITI-TF 1
IHE ITI Supplements Policy authoring tool
PWP HPD PIX / PDQ
IHE XDS
Patient identity source
Document Document
repository registry
PEP PDP
!
Patient privacy policy
PAP
XACML actors
ISO/IEC 10181-3
Fig. 3. Integration of policy authoring tools, their relation to the XACML actors and
their dependencies to an IHE-based infrastructure
In the following we identify and list different IHE-profiles the authoring ap-
plication needs to incorporate:
– Cross-enterprise document sharing (XDS) [3] document registry, in order

to support retrieval of medical data and health-care work processes domain
characteristics (cf. Section 4 – Domain characteristics)
– XDS patient identity source, in order to get local patient information (e.g.,
from within a specific hospital) as part of the patient domain characteristics
– Patient identifier cross-referencing (PIX) and patient data query (PDQ) [3],
to support a unified patient identifier and the retrieval of extensional patient
information as part of patient domain characteristics, respectively
– Healthcare provider directory (HPD) [5] to fetch provider meta-data like work
role or credentials, as part of health-care provider domain characteristics
– Personnel white pages (PWP) [3], to gain extensional provider information
as part of health-care provider domain characteristics
Employing these profiles within a health-care network maintaining electronic

health records enables highly domain-aware policy authoring. The Austrian
ELGA and our business partner ITH-icoserve correspond to IHE profiles, al-
lowing our work to be put in a practical context.
6 Conclusion and Future Work

In this paper we described usability-supporting factors for developing effective
authorization policy authoring tools. The authoring application enables patient-
controlled privacy preferences which get transformed to enforceable access con-
trol policies. These policies protect access to shared medical data within a net-
worked electronic health-care system. We elucidated factors related to the design
of such application which are related to the authorization policy model, domain
characteristics influencing the specification of policies and the analysis of poli-
cies to provide feedback to the patient-author of policies. Finally we described
a use-case of an IHE-based health-care infrastructure and a set of IHE profiles
to be integrated within an authorization policy authoring tool. These profiles
define the required interfaces to retrieve domain characteristics.
Future work discusses three major work packages: (i) Usability studies guiding
the design of user-interfaces and stakeholder views, (ii) a more formal discussion
of policy analysis regarding policy conflicts and domain-related constraints and
(iii) experiences from our efforts on integrating our authoring prototype to the
national health-care network in conformance to all national regulations.
References
1. European Commision. Directive 95/46/EC, Data Protection Directive (1995)
2. IBM Austria. Feasibility Study for implementing the electronic health record
(ELGA) in the Austrian health system (2006)
3. IHE. IT Infrastructure (ITI) Technical Framework. Integration Profiles, vol. 1
4. IHE. IT Infrastructure Access Control (White Paper) (2009)
5. IHE. IT Infrastructure (ITI) Technical Framework, Supplement, Healthcare
provider directory (HPD) (2010)
6. ISO. ISO/IEC 10181-3:1996 Information technology – Open Systems Interconnec-
tion – Security frameworks for open systems: Access control framework (1996)
7. Karat, C., Karat, J., Brodie, C., Feng, J.: Evaluating interfaces for privacy policy
rule authoring. In: CHI 2006. ACM (2006)
8. Katt, B., Breu, R., Hafner, M., Schabetsberger, T., Mair, R., Wozak, F.: Privacy
and Access Control for IHE-Based Systems. In: Weerasinghe, D. (ed.) eHealth
2008. LNCSIT, vol. 1, pp. 145–153. Springer, Heidelberg (2009)
9. Kotschy, W.: STRING ELGA Datenschutzrechtliche Analyse (German, Electronic
health record – Data privacy aspects). Austrian Federal Ministry of Health (2005)
10. LeMay, M., Fatemieh, O., Gunter, C.A.: PolicyMorph: Interactive Policy Transfor-
mations for a Logical Attribute-Based Access Control Framework. In: SACMAT
2007. ACM (2007)
11. Moffett, J.D., Sloman, M.S.: Policy conflict analysis in distributed system manage-
ment (1993)
12. OASIS. eXtensible Access Control Markup Language (XACML) v2.0 (2005)
13. Reeder, R.W., Karat, C.-M., Karat, J., Brodie, C.: Usability Challenges in Security
and Privacy Policy-Authoring Interfaces. In: Baranauskas, C., Abascal, J., Barbosa,
S.D.J. (eds.) INTERACT 2007. LNCS, vol. 4663, pp. 141–155. Springer, Heidelberg
(2007)
14. Trojer, T., Katt, B., Wozak, F., Schabetsberger, T.: An Authoring Framework for
Security Policies: A Use-Case within the Healthcare Domain. In: Szomszor, M.,
Kostkova, P. (eds.) e-Health. LNICT, vol. 69, pp. 1–9. Springer, Heidelberg (2011)
Modelling a User Authorisation and Data Access
Framework for Multi-specialty Research Systems
in Secondary Health Care
Ire Ogunsina1, Sarah N. Lim Choi Keung1, Lei Zhao1, Gavin Langford 3,
Edward Tyler1, and Theodoros N. Arvanitis2
1
Department of Primary Care Clinical Sciences
2
School of Electronic, Electrical and Computer Engineering
University of Birmingham, Edgbaston, Birmingham B15 2TT, United Kingdom
3
Birmingham and Black Country Comprehensive Local Research Network
{i.ogunsina,s.n.limchoikeung, l.zhao, e.tyler,
t.arvanitis}@bham.ac.uk, Gavin.Langford@uhb.nhs.uk
Abstract. Patient identification and consequent recruitment in clinical trials is

normally preceded with searches on electronic health record (EHR) systems.
Query results may be collated across multiple health organisations and
specialties. In such scenarios, a prime concern is the possibility of systems and
their users inadvertently or otherwise impinging on the privacy of patients.
Access to patient data is crucial for research purposes, but the degree of access
must be controlled in such a way that it conforms to agreed legal, organisational
and ethical policies. In this paper, we present a proposed model for managing a
dynamic matrix of roles and data access privileges within the context of
research systems in secondary care.
Keywords: Role Based Access Control (RBAC), secondary care research,

electronic healthcare records.
1 Introduction
Patient data is arguably the most essential resource in health care systems. The
recommendations of the Caldicott guardian stipulate that access to patient data must
strictly be on a “need-to-know” basis [1]. Caldicott-compliant systems need to be
effective in granting and restricting access to patient data and resources according to
system and user specifications. This work is part of a larger project involved with the
design and development of a research system to be used for clinical studies across
multiple specialties in secondary care. System users (subsequently referred to as the
subjects) will typically be health practitioners with proficiency in at least one
specialty and affiliated to one or more health organisations, typically NHS Trusts.
Apart from the patient data resource, license protected resources also exist for the
computation of clinical data. Access to all resources need to be verified along several
lines to ascertain rights and permissions. The paper continues with an overview of the
106 I. Ogunsina et al.
classic RBAC, some known limitations and possible solutions. Sections 3 and 4 focus
on the context of our work, the reasoning and rationale behind the proposed model.
2 Overview of the Traditional RBAC

Originally proposed by Sandhu [2], RBAC is a standard of access control whose
popularity is proof of its effectiveness. RBAC is based on a grouping mechanism,
known as roles. A subject’s role, therefore, determines what resources they have
access to. This approach breaks the tight coupling between the subject and the
permissions to resources. From a data modelling perspective, the inclusion of the role
entity resolves the many-to-many relationships between subject and rights or access to
resources [3]. Within the health enterprise, the effectiveness and appropriateness of
the RBAC approach as a privileges management infrastructure has been successfully
demonstrated by Slevin and Macfie [4].
2.1 Problems
Although RBAC has proven its effectiveness and good utilisation record across a
wide industrial spectrum, the approach is not to be considered as a panacea for all
access control issues [2]. The traditional RBAC has been shown to have limitations,
making it unsuitable for scenarios requiring complex access requirements [5]. Some
of the main problems, often associated with RBAC, are related to potential policy
conflicts and inconsistencies with authorisation of subjects with multiple roles. The
objective of our work is to model an effective framework for granting or denying
access requests to patient data and resources. This framework must take into
consideration the combination of the roles, specialties, Trusts, resources and actions
requested. The scope of the proposed model does not include more challenging
scenarios, such as emergency access requests.
3 Making Access Decisions

3.1 Policies
Multiple role assignments to a subject can potentially contribute to the problem of

separation of duty and role precedence [5]. Our approach lends from the PERMIS
authorization infrastructure project [6] where the subject, for example, is an object
which could have an attribute-key, role with the attribute-value, researcher-only, or
an attribute-key first_name, where the attribute-value may be Fred. The proposed
model is also a policy driven model where access criteria are contained in separate
policies. Trust policies for instance, would exist to specify the different relevant
directives that may exist for the subject within the context of their affiliation with the
health organisation. Similarly, specialty policies exist to provide directives relating to
subject specialties that could influence the access decision. By default, each attribute
has its policy. Our model contains an access decision framework, which provides a
decision based on the aggregation of the different policies.
Modelling a User Authorisation and Data Access Framework 107
3.2 Resources
Apart from patient data, another requirement of the model is the management of
system resources and tools. An example is licensed quality of life questionnaires
which may exist in paper or electronic format. A typical case is the Hospital Anxiety
and Depression Scale (HADS) [7] score in the COPD specialty. These licenses are
usually bought by the Trust and used in calculating patient data.
The RBAC model of the now defunct NHS National Programme for IT (NPfIT)
project [8] includes additional concepts for suitability within the health enterprise.
Within each policy entity, directives would specify the concept status such as
Legitimate Relationships (LR) - ensuring that patient identifiable data is only
accessible if the subject is involved in the patient’s care. Non patient identifiable data
may be accessed as allowed within the Sealed Envelopes (SE) segments of policy
documents. In most access request scenarios, it would be impossible to be granted
access without Patient Consent (PC) status being true within policy documents.
As shown in Figure 1, the scalable model generates a decision outcome as a
product of aggregated policies. This approach gleans from Blobel’s more detailed
model [9] as well as HL7’s security policy information model [10].
Policy
Composite Resource
Subj ect Policy
Trust Role Specialty Action
Fig. 1. Policy-based model for resource access decisions
4 Conclusion and Related Work

4.1 Related Work
Farzad and Yu [11] extended Crook’s RBAC model [12] which modelled the
concepts of responsibility, operation and context in addition to the role concept as
criteria for object access and permission, bears good resemblance with our approach
work described with particular focus on including the model during the knowledge
108 I. Ogunsina et al.
engineering phase. Although the work by Slevin and Macfie [4] involves a single
specialty and Trust in a clinical environment, it highlights challenges common to
access control mechanisms within healthcare systems and possible solutions.
4.2 Conclusions and Future Work

Traditional RBAC may be extended to include policies and constraints that will
augment its suitability in complex scenarios. Our proposed model is a policy-oriented
framework which handles access decisions in a flexible way within the context of
multi-specialty research systems in secondary health care. It would be beneficial to
see how integrating Cassandra, [1] a role based, trust management system into the
permissions management infrastructure would facilitate our model. Furthermore, it
would be interesting to research on aligning our model fully with the HL7’s Privacy,
Access and Security Services (PASS) Access Control Services model [10].
Acknowledgements. The work was supported by the England’s National Institute for
Health Research (NIHR) and the Birmingham and Black Country Comprehensive
Local Research Network (BBC CLRN).
References
[1] Becker, M.Y., Sewell, P.: Cassandra: Flexible Trust Management, Applied to Electronic
Health Records. In: Computer Security Foundations Workshop, pp. 139—154 (2004)
[2] Sandu, R.S., Coyne, E.J., Feinstein, H.L., Youman, C.E.: Role-Based Access Control
Models. IEEE Computer 29, 38–47 (1996)
[3] Ferraiolo, D.F., Kuhn, D.R.: Role Based Access Controls. In: 15th National Computer
Security Conference, pp. 554–563 (1992)
[4] Slevin, L.A., Macfie A.: Role Based Access Control for a Medical Database. In:
IASTED-Software Engineering and Applications Conference, pp. 19–21 (2007)
[5] Covington, M.J., Moyer, M.J., Ahamad, M.: Generalized Role-Based Access Control for
Securing Future Applications. Technical Report GIT-CC-00-02. Georgia Institute of
Technology (2000)
[6] PERMIS. FAQ (2011), http://sec.cs.kent.ac.uk/permis/documents/FAQ.shtml
[7] Snaith, R.P.: The Hospital Anxiety and Depression Scale. Health Qual. Life Outcomes 1,
29 (2003)
[8] National programme for IT (NPfIT), http://www.gpchoice.org/npfit.aspx
[9] Blobel, B.: Authorisation and Access Control for Electronic Health Record Systems.
International Journal of Medical Informatics 73, 251–257 (2004)
[10] HL7. Privacy, Access and Security Services (PASS) Access Control Services Conceptual
Model. Release 1 (2010),
http://hssp-security.wikispaces.com/
PASS+HL7+Balloted+Documents
[11] Farzad, F., Yu, E., Hung, P.C.K.: Role Based Access Control Requirements Model with
Purpose Extension. In: Workshop on Requirements Engineering, pp. 207–216 (2007)
[12] Crook, R., Ince, D., Nuseibeh, B.: Modelling Access Policies Using Roles in
Requirements Engineering. Information and Software Technology 45(14), 979–991
(2003)
Software Engineering-Inspired Approach
to Prevention Healthcare
Victor Rentea, Andrei Vasilateanu, Radu Ioanitescu, and Luca Dan Serbanati
Faculty of Engineering Taught in Foreign Languages, Politehnica University of Bucharest,

Bucharest, Romania
victorrentea@gmail.com, andrei.vasilateanu@upb.ro,
ioanitescuradu@yahoo.com, luca@serbanati.com
Abstract. The paper extracts proven principles from software engineering to be

applied in the field of prevention and wellness preserving. For instance, the con-
cern-oriented partitioning principle could support the analysis and design of health-
care information systems. Results of such an approach are multifaceted systems
that integrate all stakeholders’ concerns. For such a system, the paper proposes a
multi-agent system, where a patient avatar negotiates with expert systems assisting
the user to monitor his health state and supply proven prevention alternatives. The
aggregated patient-centered virtual organization also aims at stimulating the patient
empowerment through personalized information based on their conditions.
Keywords: patient empowerment, prevention, virtual health assistant, patient-

centric virtual organization, multi-agent systems.
1 Introduction
In the past two decades, much of the growth in health expenditures has been attributa-
ble to chronic conditions in the context of global population ageing. Future health
reform should rely on stimulating patient empowerment [1], i.e., people to take re-
sponsibility of their health status and act proactively upon maintaining it [2]. One of
the most promising applications of patient empowerment is in the field of prevention,
aiming to avoid even the debut of chronic conditions, with an alarming increasing
prevalence in young people [3]. However, the existing healthcare information systems
are not suitable for this purpose because they focus on acute disease care and favor
diagnosis and treatment based mainly on the current symptoms [4]. The paper ap-
proaches the healthcare area from the original and seminal perspective of software
analysis and design, applying proven methods for tackling partial knowledge prob-
lems in the prevention field. Using a concern-oriented approach to health monitoring
and prevention, a multi-agent system is designed around the Personal Health Record
(PHR) to provide continuous assistance throughout the whole lifespan of the user.
Section 2 presents a brief state of the art in the fields this paper relies upon. To ad-
dress some of the identified problems, the conceptual design of a multi-agent system
is presented in Section 4, building upon a theoretical basis described in Section 3.
Conclusions and future work are presented in section 5.
110 V. Rentea et al.
2 State of the Art

The use of agents in healthcare has a wide range of applications, including decision sup-
port, medical data management, resource planning and pervasive care [5]. Numerous
prototype agent systems were proposed which demonstrate clear strengths compared to
classic programming, such as support for expert systems or personalization.
Significant improvements in clinical practice have been demonstrated through the
use of decision support sytems [6]. Such systems also present new opportunities for
patient empowerment through feedback and recommendations.
3 Theoretical Background
The goal of our research is to compare the software development process with the
management of the patient’s health state in the attempt to apply proven methods from
the former to innovate in the prevention field. Although apparently very different in
nature, the two domains have many similarities: both produce unique, unrepeatable
work, both make critical decisions relying on some abstract model built through itera-
tive analysis and both emphasize the continuous nature of their work, i.e. treatment
and maintenance, respectively. Furthermore, both medical assistance and software
development requests are triggered by a change in an unpredictable universe, health
status and business environment, respectively.
The IEEE defines the concerns for a system as “those interests which pertain to the
system’s development, its operation or any other aspects that are critical or otherwise
important to one or more stakeholders” [7]. Concern-Oriented Software Development
[8][9] is an approach to partitioning a software system in a concern-driven way, with
the aim of supporting the specification of stakeholders’ concerns and their composi-
tion into a working system. The concerns about a system are represented as facets,
usually interrelated and overlapping. A key supposition is that the sum of all the fa-
cets identified is the system to be developed, because the composite representation
subsumes all the knowledge about the system available at that moment [8]. Thus,
the system analyst has to consider all the stakeholders’ concerns and design the soft-
ware system to harmonize all the different facets. Similarly, the medical specialties
have different, sometimes overlapping or even conflicting views and concerns about
the health state of a patient that should be harmonized in the final medical advice.
In software development several requirements often recur in similar forms, e.g., the
requirement for security, but to implement it correctly, worldwide-agreed standards
and proven solutions are reused. In medical practice, the proven knowledge is built on
clinical trials [10] that quantify the benefic or detrimental effect of habits and lifestyle
on the health state in the form of a Relative Risk (RR%), and embedded in clinical
guidelines and protocols that indicate the best therapeutic attitude.
After delivery, a software system enters the maintenance phase throughout which its
various problems and limitations emerge in the ever-changing business environment.
Similarly, in healthcare, while treating a patient multiple adverse effects together with
unexpected medical conditions may begin to manifest. However, regular monitoring
could predict acute and possibly dangerous manifestations.
Software Engineering-Inspired Approach to Prevention Healthcare 111
4 The Patient-Centric Virtual Organization
The system represents a Patient-Centric Virtual Organization (PCVO) and consists of

three parts: the PHR, the Virtual Health Assistants (VHA) and the patient avatar.
Conceptually, the PHR provides views on the patient’s current health state and clini-
cal history, including both self-gathered data and authoritative information retrieved from
the Electronic Health Record (EHR). VHAs are collaborating expert systems whose role
in the organization is to predict and prevent health deterioration based on the PHR data.
The system is designed using a concern-oriented approach, with each VHA dedicated to
a medical specialty. This organization enables distributed development and responsibility
of the prevention Knowledge Bases (KB) containing proven advices. In the PCVO, the
avatar is the leading agent that coordinates the activity and the only entity that directly
communicates with the user. One of its main tasks is to capture the user profile and to
adapt to it in order to increase its influence on her/his life style. On the other hand, in
terms of virtual/real distinction, it is a boundary agent that maps the real into the virtual
[4] and provides the required information to the VHAs. It is a design objective for our
system to be low-intrusive, essentially remaining a reactive system and using proactive
communication channels only for critical health conditions. A typical usage scenario is
presented below (Fig. 1):
• The user contacts the avatar which in turn asks the VHAs for information requests.
• Each VHA responds taking into consideration the known state of the respective
anatomic system, predispositions and interactions with other VHAs.
• The avatar aggregates and provides the user a personalized view on these requests.
• The user responds based on current symptomatology and/or by executing several
quick self-tests (e.g. glucose level) and her/his responses are stored in the PHR.
• To build a clearer image, a VHA can request further pacient information.
Fig. 1. Patient-Centric Virtual Organization

112 V. Rentea et al.
• The different recommendations are negotiated between VHAs, e.g., potassium salt
is good for patients with high blood pressure, but questionable in case of arrythmia.
• The final recommendations are presented to the user and integrated in the PHR.
5 Conclusions
Software development was compared to healthcare provisioning and proven methods

were applied in the field of medical prevention. The paper presents the conceptual
design of a concern-oriented multi-agent system for monitoring the health state and
stimulating the patient empowerment. Unlike most existing agent systems [5], the
control loop does not include any medical staff, thus allowing for a quicker adoption.
What differentiates our approach from other medical expert systems [6] is the empha-
sis on prevention targeted at persons with no previous symptomatology. Incipient
stages of an otherwise unsuspected illness are detected and help is provided to prevent
its debut. Future work will address the behavior of the avatar and the negotiation
among the VHAs that determines the questions and recommendations for the patient.
The user of the system benefits from a continuous and unobtrusive monitoring
achieved through simple and personalized questions. From the user’s perspective, the
system is in continuous evolution following his current health state and providing
proper advices. On a social dimension, an important expected impact is on stimulating
the patient empowerment and disseminating basic medical information, through perti-
nent advices and continuous care provided by the system. However, the greatest po-
tential of the system relies in the opportunity to avoid the very development of a
chronic condition, one of the today’s major health concerns.
References
1. Thorpe, K., Ogden, L., Galactionova, K.: Chronic conditions account for rise in medicare
spending from 1987 to 2006. Health Aff. (February 2010) doi:10.1377/hlthaff.2009.0474
2. Anderson, R.M., Funnell, M.M.: Patient empowerment: reflections on the challenge of fos-
tering the adoption of a new paradigm. Pat. Ed. and Counsel. 57, 153–157 (2005)
3. Schleiffenbauma, B.E., et al.: Unexpected high prevalence of metabolic disorders and
chronic disease among young male draftees – the Swiss Army XXI experience. Swiss
Med. Wkly 136, 175–184 (2006)
4. Serbanati, L.D., Ricci, L.R., Mercurio, G., Vasilateanu, A.: Steps towards a digital health
ecosystem. Journal of Biomedical Informatics (in press) doi: 10.1016/j.jbi.2011.02.011
5. Isern, D., Sánchez, D., Moreno, A.: Agents applied in health care: A review. International
Journal of Medical Informatics 75, 145–166 (2010)
6. Kawamoto, K., Houlihan, C.A., Balas, E.A., Lobach, D.F.: Improving clinical practice us-
ing clinical decision support systems: a systematic review of trials to identify features crit-
ical to success. BMJ. 330, 756 (2005)
7. IEEE Recommended Practice for Architectural Description of Software-Intensive Systems.
IEEE Std. 1471 (2000)
Software Engineering-Inspired Approach to Prevention Healthcare 113
8. Serbanati, L.D.: Integrating Tools for Software Development. Prentice Hall (1992)
9. Bogdan, C., Serbanati L.D.: Concern-oriented and Ontology Based Analysis of Informa-
tion Systems. In: Proc. ITAIS Conference 2008, in Information Systems: People, Organi-
zations, Institutions, and Technologies, pp. 245–253. Springer, Heidelberg (2010)
10. Timmermans, S., Mauck, A.: The promises and pitfalls of evidence-based medicine.
Health Aff. 24, 18–28 (2005)
11. Arrow, K.J.: Uncertainty and the welfare economics of medical care. Am. Econ. Rev. 53,
941–973 (1963)
Evaluation of a Web-Based Patient Portal
for Chronic Disease Management
Stacey Guy, Alexandria Ratzki-Leewing, and Femida Gwadry-Sridhar
Lawson Health Research Institute,

Commissioners Rd E. 801, N6C 5J1 London, Canada
stacey.guy@sjhc.london.on.ca
alexandria.ratzkileewing@lawsonresearch.com
femida.gwadry-sridhar@lhsc.on.ca
Abstract. Chronic disease directly affects more than 9 million Canadians.

Efficient strategies are needed to cope with the demand on health care services and
to increase patient adherence to treatment. Emerging web 2.0 technologies present
viable options for patient engagement in health care. We undertook a pilot project
to assess the feasibility of two chronic disease management patient portals. A total
of 35 patients participated in the assessment. Portals were evaluated for participant
expectations, motivations, usability, and recommendations for future iterations.
Findings suggest the features of this portal were useful. Important issues to
participants include access to their medical record, communication with health care
professionals and other participants regarding topics of interest, keeping track of
biometrics, and keeping up with the latest clinical studies.
Keywords: Diabetes, Prostate Cancer, Web 2.0, Chronic Disease, Health

Portal.
1 Introduction
Currently, more than 9 million people are suffering from chronic disease in Canada
[1]. Accounting for nearly 87% of all disability in the country and consuming over
67% of all healthcare costs, chronic disease poses an incredible burden on the
Canadian healthcare system. By 2015, the World Health Organization predicts that
chronic disease will account for 89% of all Canadian deaths [2].
In Ontario, Canada, the impact is just as severe. Approximately 1 in 3 people suffer
from at least one chronic disease in the province, costing the healthcare system a total
of 80 billion dollars annually [3]. The impact of adverse effects of chronic disease are
especially salient in Southwestern Ontario where rates of chronic disease, particularly
prostate cancer and type II diabetes, are disproportionately higher than in other
regions of the province [4]. A 2011 report by the Canadian Cancer Society found
prostate cancer to be the most frequently diagnosed cancer in Ontario followed by
breast cancer and colorectal cancer [5].
Canada, like most developed countries have a cohort of aging baby-boomers. In
Southwestern Ontario the prevalence of diabetes and cancers continues to grow at an
Evaluation of a Web-Based Patient Portal for Chronic Disease Management 115
alarming rate. However notwithstanding current strategies to mitigate the prevalence

and severity of the diseases, Ontario’s healthcare system is underperforming with
regards to chronic disease management and treatment [3]. Increasing chronic disease
patient participatory care needs requires innovative use of communication technology,
including interactive, tailored programs with feedback and social support through
networks, not to mention access to care.
1.1 Study Objective

In conjunction with a web portal provider - we developed, implemented and evaluated
a web-based chronic disease management system for patients suffering from diabetes
(My Diabetes Wellness PortalTM) and prostate cancer (ProPortalTM), respectively
(MedManager Interactive Corp., Waterloo, ON). The goal of this pilot project is to
test the effectiveness and usability of an interactive web-based patient portal in
providing prostate cancer, and diabetes patients with the knowledge guidance and
education they need to help them understand their disease.
Fig. 1. My Diabetes Wellness PortalTM interface. Features of this portal can be seen in the side
bar located on the left-hand side.
2 Methods
We designed and implemented a pilot study where patients from each disease cohort
were allocated to one group per portal and followed for 6 months. All patients were
given access to usual care (print material, advice from their physician and other health
providers) and new sources of education, via the web-based portal. Patients had
access to self-monitoring tools, and the ability to track disease-related metrics. Our
ethics committee (University of Western Ontario) found this approach acceptable
(REB #16100E).
Two disease cohorts were targeted through convenience sampling. Patients
diagnosed with prostate cancer, including first and second line, of any age, were
approached to participate. Patients with type II diabetes were invited to participate.
116 S. Guy, A. Ratzki-Leewing, and F. Gwadry-Sridhar
This cohort was chosen as the disease requires many variables to be managed to
achieve optimal care. In both populations patients needed to have access to a
computer and the internet. Patients with comorbidities were not excluded. This
allowed us to have a fully representative population. We were not concerned with the
age range between cohorts (prostate cancer patients are generally an older cohort) as
we were not comparing cohorts. Rather this range allowed us to examine possible
barriers to use of web 2.0.
Patients were recruited by the physicians and educators at the London Health
Sciences Centre. One clinic was located in the Cancer Program, the other in
Endocrinology Care. These clinics are ideal as a large number of patients from
practices in South-western Ontario are referred for treatment. In addition to face-to-
face recruitment, mail outs were sent to patients. We aimed to recruit a total of 50
patients for each condition.
The patient letter of information and consent clearly explained the nature of what
data was going to be stored and how this was going to be stored. The data stored is
encrypted, and the portal company did not have access to any data. Data is stored
behind a secure hospital firewall and backed up nightly. The principal investigator,
physician, research assistant and healthcare provider on the team (a social worker, and
a pharmacist) had access to the portal. Patients were given randomly generated logins
which they changed in order to ensure confidentiality. Disease specific data for the
site is provided through manual entry (information entered by the patient) and HL7
data transfer.
An interactive, guided help video is available on the portal. Upon signing in, the
help video would pop up. A toll-free helpline was established for participants
experiencing difficulties. This was manned by the portal provider. Participants were
also able to call the research assistant if an issue arose.
We initiated the study in September 2009. Each portal was monitored by a
healthcare provider. The portal enables participants to track their disease-related
metrics (e.g., diabetes patients could download readings from their blood glucose
monitor) and visualize the data via graphs. External notifications based on goals set
by patients are automatically sent. Evidence-based educational material, chosen by
healthcare providers on the team, was accessible through the portal. Patient-to-patient
and patient-to-provider interaction was available through a community forum and
short messaging service.
The feasibility of this portal is assessed by a telephonic survey and a focus group.
A 10 minute telephonic survey was administered to patients from both portals. This
survey consists of 28 likert scale items and 4 open-ended questions. It was created to
assess patient experience with the portals over the course of 3 months. Questions were
derived from a questionnaire developed by Evangelista et al., (2006) [6] as well as an
expert panel of software developers, and healthcare providers (specifically, team
members). Items evaluated: motivation to use the portal, expectations, usability,
aesthetics, specific features, support team service, and benefit to health. Responses to
the survey were analyzed according to frequency and were subsequently grouped into
the themes.
Focus groups were held, at study closure, with participants from each portal, to
provide in-depth understanding of portal experience. A focus group is a qualitative
research data collection method. Focus groups are particularly useful for exploring
opinions, preferences and experiences of a study [7, 8]. Focus groups "have an
advantage for researchers in the field of health and medicine...they can encourage
participation from people reluctant to be interviewed on their own or who feel they
have nothing to say" p.299[9]. Focus group validity is recognized by considering the
participants' responses as "an accurate representation of the perceptions of reality for
the group members and therefore valid" p.489 [10]. According to Calder [7] enough
focus group sessions have been held when it is possible for the moderator to
anticipate what will be said next. A semi-structured interview schedule was
developed, aimed at developing iterations of the portal that would provide optimal
support for managing a disease. The themes explored within the focus groups
included: motivations and expectations of the project, usability of the portal, reasons
for usage or non-usage of the community forum, and suggestions for the next portal
prototype. Two facilitators were present at each focus group (SG, and FG-S). These
sessions were audio recorded.
3 Analysis
Data from the telephonic questionnaire was entered into SPSS (Chicago, Il.) where
the frequency of items was computed. Open-ended questions were grouped into
themes initially laid out by the areas delineated in the survey. In addition, usage
statistics were collected through the backend of the portal. This provided data on
number of logins, time spent on each task etc.
The two focus group sessions (one for each cohort) were audio recorded and the
raw data was transcribed by a moderator (S.G). This data was analyzed according to
thematic analysis. Qualitative thematic analysis provides a rigorous method of
analysis across which a gathered data set will be searched and organized in to pre-
empted and emergent themes (repeated units of meaning or patterns) [11; 12] This
analysis requires that initial codes be generated after transcription, searching for
patterns across the data set, reviewing the themes, defining and naming themes, as
well as reporting issues considered relevant to the research question.
4 Results
Thirty-four out of 64 participants completed the survey (at the time of the survey, 64
participants were enrolled in the study). Ten male and five female My Diabetes
Wellness PortalTM (MyDWP) participants completed the survey – the majority of
which (5 participants) were between the ages of 40 to50 years. The majority of
ProPortalTM (ProP) participants (n=19) who completed the survey were between the
ages of 73-83 years (8 participants). At study closure, when the focus groups were
conducted, a total of 99 participants (46 MyDWP participants, and 53 ProP
participants) had consented to the study. Five ProP, and 2 MyDWP participants took
part in the focus groups. Findings from both the survey and focus groups are reported
within themes in the table below.
Table 1. Survey and focus groups findings across both cohorts
Themes Findings
Expand knowledge base and receive Canadian content
Sense of community and social network.
Motivations
Help others.
Find out how to improve overall health.
Involvement of primary physician.
Complete medical record history pertaining to disease to be
Expectations
available via portal.
System to be available to hospital personnel.
Easy to navigate, well-organized, clear, caught-on quickly.
Usability Feelings of frustration at missing medical information led to
discontinued use.
Time spent on MyDWP: average 15.60 minutes with average
participant login of 1.64 per day.
Time spent on ProP: 17.58 minutes with average participant
Portal Usage
login of 0.66 times per day.
Frequency of login related to checking for new information and
postings.
Privacy &
Felt personal information secure and protected.
Security
Features
Lack of participation by participants and healthcare team.
Community Wanted to see more activity.
Forum Recommendations: Discussion led by healthcare provider on
topic chosen by participants.
Tracking Well-liked and used.
Tools Need to combine items of likeness.
External
‘No new messages’ led to turning off feature.
Notification
MyDWP: liked ability to upload glucometer readings. However,
Personal some glucometers were not compatible with the system.
Health Delay between results and appearance on portal.
Record Results uploaded were close to unreadable.
ProP: Wanted portal pre-populated with biometric data.
Table 1. (continued)
Future Portal Recommendations

Ability to design guest accounts.
Visual identifiers for healthcare team.
One main portal with functionality to choose co-morbidity.
Design Receive notification of portal updates and reminders.
Relational graphing of biometrics.
Glossary of acronyms.
Intuitive system.
Video clips of talks/programs/recent news.
Pedagogy “Frequently Asked Questions”
Information on new study findings and products.
Medical Clinicians to make recommendations based on info entered by
Records participants.
Portal of worth and would pay a fee for that.
Economic Costs
Build into insurance of chronic disease.
5 Discussion
With 1 in 20 diabetes patients in Ontario experiencing major complications within a

year, [13] & 1 in 7 Canadian men at risk for prostate cancer [14], it is essential we
explore the feasibility of tailored, interactive web-based portals to encourage patient
self-monitoring. Regardless of age (majority between 61-83 years old) 58 participants
logged in and used the portal. The majority of participants derived benefit from the
tools provided, however they wanted a more interactive social medium.
Future forays into web-based self-management programs need to engage
participants in evaluation to ensure sustainability. Participant recommendations
include emphasis on the social aspects of the portal, as well as integration of pre-
populated medical records for tracking. Self-management opportunities will support
future actions based on engaging patients in evaluation and improvements to these
portals. Their suggestions include, specifically increasing forum activity- for chats
and provider–patient interactions, posted clinician recommendations and pre-
populated medical records for tracking.
5.1 Study Limitations

Despite the initial sample size of 64 participants (size of the sample at the telephonic
interview time period), only 33 patients completed the survey and only 7 participated
in the focus group (an effort was made to contact all participants). This concern
summons question bias. Did non-compliant users not respond to the survey? Would
their answers have provided different information than what was collected?
Additionally, questions regarding the motives of focus group participants and what
implications these motivations may have had on study results must also be
considered. Repeating the trial with a larger sample size will improve the validity and
generalizability of the study; in painting a more accurate picture of the MyDWP,
researchers can accurately assess the portal’s effectiveness. In the future, it may be
beneficial to apply the technology adoption model (TAM) to determine user
acceptance of the portal and to make study outcomes more generalizable to the
research population.
6 Conclusion
This pilot study marks an important journey into e-based chronic disease management
in Canada. As the role patients with chronic disease play in their ‘healthfulness’
(as opposed to illness) becomes larger, cost-effective avenues to explore self-
management become crucial to the survival of our healthcare system. By learning
from the findings discussed in this paper, researchers will be able to deploy future
iterations of portals that encompass more of what patients want to see. Issues of
importance to participants include access to their medical record, communication with
health care professionals and other participants regarding topics of interest, keeping
track of biometrics, and keeping up with the latest clinical studies.
Acknowledgments. This work was funded by the Ministry of Research and

Innovation, The Health Technology Exchange. We would like to thank the following
investigators on the project: Dr. Glenn Bauman, Dr. Merrill Edmonds, and
MedManager Interactive Corp.
References
1. Canadian Academy of Health Sciences.: Health System Transformation to Meet the
Burden of Chronic Disease: The Challenge (2008),
http://www.cahs-acss.ca/e/pdfs/CDM_Challenge.pdf
2. World Health Organization.: Preventing Chronic Diseases: A Vital Investment. World
Health Organization Press, Geneva (2005)
3. Ontario Health Quality Council.: 2008 Report on Ontario’s Health System. Ontario Health
Quality Council, Toronto (2008)
4. The University of Western Ontario.: Facing Facts (2011),
http://www.robarts.ca/facing-facts
5. Canadian Cancer Society.: Ontario cancer Statistics (2011),
http://www.cancer.ca/ontario/about%20cancer/
cancer%20statistics/ontario%20cancer%20statistics.aspx
6. Evangelista, A., Stromberg, A., Westlake, C., Ter Galstanyan, A., Anderson, N., Dracup,
K.: Developing A Web-based Education and Counseling Program for Heart Failure
Patients. Prog. Cardiovasc. Nurs. 21, 196–201 (2006)
7. Asbury, J.E.: Overview of Focus Group Research. Qual. Health Res. 5, 414–420 (1995)
8. Barbour, R.S.: Making Sense of Focus Groups. Med. Educ. 39, 742–750 (2005)
9. Kitzinger, J.: Qualitative Research – Introducing Focus Groups. Br. Med. J. 311,
299–302 (1995)
10. Carey, M.A.: Concerns in the Analysis of Focus Group Data. Qual. Health Res. 5,
487–495 (1995)
11. Braun, V., Clarke, V.: Using Thematic Analysis in Psychology. Qual. Res. Psych. 743,
77–101
12. Huberman, A.M., Miles, M.B.: The Qualitative Researcher’s Companion. Sage, Thousand
Oaks (2002)
13. Ontario Health Quality Council.: 2010 Report on Ontario’s Health System. Ontario Health
Quality Council, Toronto (2010)
14. Canadian Cancer Society.: Prostate Cancer Statistics at a Glance (2011),
http://www.cancer.ca/Canada-
wide/About%20cancer/Cancer%20statistics/Stats%20at%20a%20gla
nce/Prostate%20cancer.aspx?sc_lang=en
Cardio Online Reader/COR: A Web
2.0-Based Tool Aimed at Clinical Decision-Making Support
in Cardiology
Vendula Papíková and Miroslav Zvolský
Centre of Biomedical Infomatics, Institute of Computer Science,

Academy of Sciences of the Czech Republic,
Pod Vodárenskou věží 2, 182 07 Prague, Czech Republic
{papikova,zvolsky}@euromise.cz
Abstract. There is a wide acceptance of the fact, that processes of clinical

decision-making has to be based on sound scientific evidence. But in the
context of a rapidly growing amount of new information, it is increasingly
difficult task. That is why specialized tools and resources enabling to quickly
and efficiently search and disseminate relevant evidence are needed. Web 2.0 is
a platform that provides an armamentarium with great potential to contribute to
solving this task. This paper presents a newly developed information resource
aimed at support of clinical decision making in cardiology called Cardio Online
Reader/COR. The COR provides the best available evidence from database
MEDLINE/PubMed by means of an online application equipped with tools and
services typical for Web 2.0. A beta-version of the application is freely
accessible at http://neo.euromise.cz/cor.
Keywords: EBM, Web 2.0, medical information sources, clinical decision-

making support.
1 Introduction
There is a wide acceptance of the fact, that the processes of clinical decision-making
has to be based on reliable scientific evidence, nowadays derived from clinical
research trials [1], [2], [3], [4]. A general agreement exists also around the statement,
that medical doctors have been challenging a huge increase of information, which is
still unstoppably growing. They do not know about important advances, and feel
overwhelmed by new scientific information [5], [6].
1.1 Information Overload

According to the study of Richard Smith medical knowledge will increase four times
during a professional lifetime [6].Nowadays, a total of 20,000 randomized controlled
trials (RCTs) are published each year and approximately 50 new RCTs are published
every day. Therefore, to keep up to date with RCTs alone, a general practitioner (GP)
Cardio Online Reader/COR 123
would have to read one study report every half hour, day and night. In addition to
RCTs, about 1000 papers are also indexed daily on MEDLINE from a total of about
5000 journal articles published each day [7]. In addition to that, the quality of most of
the information is very poor: most published information is irrelevant and/or the
methods are not good [7]. Medical doctors have to select and appraise the sought
information, which requires specific knowledge, skills and experience. But previous
studies unveiled, they are not good at finding new information, and do not know how
to evaluate it when it is found [5], [6]. Information retrieval therefore is not a trivial
task for most of them.
Even when clinicians have time to read some of a new literature, it is difficult to
identify which information will be most useful in clinical practice and to recall the most
up-to-date findings when they need them [7]. Moreover, push technology to disseminate
information has magnified the problem to unwanted information [8]. Unsolicited
information received through the mail alone can amount to kilograms per month [7]. And
so busy clinicians are now caught in an information paradox - overwhelmed with
information but unable to find the knowledge they need when they need it [8].
Considering this reality, it is not surprising, that medical practitioners, particularly
GPs, are overloaded with information[7] and - at the same time - they face a serious
problem in keeping up to date [5], [6], [7].
1.2 Evidence-Based Clinical Practice

Evidence-based medicine (EBM) movement has been promoting the use of results of
clinical research in medical practice. But in the above described context, it is an
increasingly difficult task. It has been published a whole range of studies referring
about difficulties and obstacles of adequate utilization of the latest and best available
evidence in the clinical practice (e.g. [2], [9]).
One of the biggest obstacles is clinicians´ lack of time [10].A questionnaire study
of general practitioners was carried out in the former Wessex region, England.
Randomly selected sample of 25% of all general practitioners (452), of whom 302
replied. The study revealed that the major perceived barrier to evidence based practice
was lack of personal time [11]. Time limitations dictate that evidence based
practitioners also rely heavily on conclusions from pre-appraised resources, which
apply a methodological filter to original investigations and therefore ensure a minimal
standard of validity (e.g. Cochrane Library or ACP Journal Club) [12].
Another clinicians´ barrier to evidence-based practice is lack of specific skills[12].
They have difficulty finding, assessing, interpreting, and applying current best
evidence [13].
1.3 Clinical Information Sources
Doctors need to be linked to the medical research literature in a way that allows them to
routinely obtain up-to-date, outcomes-based information [7]. Most of the questions
generated by doctors can be answered, usually from electronic sources, but it is time
consuming and expensive to do so-and demands information skills that many doctors do
124 V. Papíková and M. Zvolský
not have [6].Thoughnew resources focused on clinical doctors´ information needs (e.g.
Clinical Evidence or Evidence Updates) have been created, new information tools still
are needed [4], [6]. They should respect the "3Rs" of evidence communication, which are
reliability, relevance, and readability [4]. They are likely to be electronic, portable, fast,
easy to use, connected to both a large valid database of medical knowledge and the
patient record, and a servant of patients as well as doctors [6].
Some of the features mentioned above can be accomplished by means of the
Internet. Doctors are increasingly proficient with the Internet. Searching with Google
came out as useful means to formulate a differential diagnosis in difficult diagnostic
cases [14]. Yet doctors are seeking new methods of information discovery because of
the limitations of search engines [15]. Also the use of MEDLINE/PubMed to answer
daily medical care questions is limited because it is challenging to retrieve a small set
of relevant articles and time is restricted [9].
The Web 2.0 proved a potent platform able to provide right tools for the above
mentioned tasks. Its applications and services are characterized by features enabling
collaboration, information sharing and aggregation, composition of independent
services and provision of rich user interaction [16]. Using an RSS doctors can fight
information overload. RSS feeds enable to them to organize new web content sent to
them by various medical websites in a single interface of an RSS reader [15]. All of
the facts described above motivated us to develop an information system aimed at
targeted dissemination of the best available evidence from the cardiology by means of
tools and services of the Web 2.0.
2 Cardio Online Reader/COR
The COR (Cardio Online Reader) application is based on domain focused records of
scientific publications, which are presented using Web 2.0 technologies. The application
functions as an online RSS reader and database of selected types of scientific articles. In
the process of selecting information we put the accent on their high reliability and clinical
relevance according to principles of evidence-based medicine. The articles have been
gained from the biomedical database MEDLINE/PubMed. The automated script
periodically loads selected records from free accessible interface of the PubMed and
stores them in the own fully searchable database of the COR. Afterwards the most recent
articles are displayed at top positions of the COR title web page. The user interface of the
COR was developed with an accent on ease of use and simplicity of control.
The goal of the COR is to simplify tracking and searching for methodologically
valid and clinically relevant publications to disseminate the latest piece of knowledge
from the clinical research to the clinical practice, and to create a space for discussion
about these findings and articles. These goals are reached by following attributes and
functions:
• The online reader presents articles selected from the MEDLINE/PubMed database.
• The articles have been chosen according to criteria of evidence-based medicine,
specifically methodological reliability and clinical relevance. For that reason
content of the application is created from five types of articles. They are Guidelines
(G), Practice Guidelines (PG), Systematic Reviews, (SR), Systematic Reviews

with Meta-Analysis, (SR&MA), and Randomized Controlled Trials (RCT).
Separate types of articles are marked by variously colored graphical elements.
• A default page of the application contains a chronological list (in the reverse order,
latest articles are on top of the list) of abbreviated records in the following format:
article title – article authors – the date of enlisting to the PubMed”. Besides that the
number of comments and user rating for each article are also displayed.
• A directly accessible detailed display of the record consists of bibliographical data,
generally the abstract of the article, link to the full text (placed at its original
location), and link to the original MEDLINE/PubMed record (which can be
accessed in a new browser window). Furthermore each article can be rated using a
five-point scale within the detailed display of the article record.
• A search form can be filled out by keywords, MeSH terms, author name, and type
of an article. Search results can be limited by a time filter including a graphical
calendar. The full text search works with Boolean terms. The interface of the COR
application provides also a form enabling a direct retrieval of PubMed using the
Entrez system.
• The articles are tagged by MeSH terms, which are displayed as a tag cloud
showing the relative frequencies of MeSH terms (“MeSH cloud”). A simple MeSH
cloud consists from about sixty the most frequent terms within the selected type of
articles (or within all articles, if no type of article is selected). The cloud is
displayed in the right column of every page. A complete MeSH cloud (or a MeSH
list in alphabetical order) can be accessed by provided link. These lists are built
from all MeSH terms contained in all articles enlisted in the COR database. An
user can simply add each MeSH term to the search form by clicking on it.
• There are implemented web services enabling viral propagation of the whole
application, as well as the individual articles. Users can use the Facebook social
network or the Twitter microblog as well as social news websites, social
bookmarking websites, and many blogging platforms. The application is connected
with web services providing social bookmarks including tools for archiving and
sharing expert publications as Connotea or CiteULike.
• Users can trace activities related to the COR through Facebook or Twitter,
YouTube, blog, and/or. They can also subscribe to website actualizations via
integrated RSS feeds.
3 Discussion
The presented web application uses freely accessible source of biomedical bibliographic
information and brings the added value of domain specific focus (cardiology). Its
development is concentrated on quality, simplicity, and usability.
We plane to develop a fully individualized interface, predefined filters, list of
favorite articles, authors or MeSH tags, and other advanced functions and tools for
registered users in next versions of the application. There is also a need for at least
basic registration process to ensure the chance to archive the authorship of comments
126 V. Papíková and M. Zvolský
and ratings. A big deal will be to adjust the application to future trends in information
sharing, as they develop spontaneously.
4 Summary
We introduce the Cardio Online Reader/COR, which is a medical information

resource focused on cardiology. It is intended to support clinical decision-making and
medical care based on evidence ("Evidence-Based Health Care"). The application is
therefore designed to facilitate both monitoring and searching for scientific evidence
to doctors, and to streamline the transfer of the latest findings of clinical research into
the clinical practice. The content has been acquired from one of the leading
biomedical databases, MEDLINE/PubMed, and has been equipped with tools and
services specific for Web 2.0. It also offers space for communication of users about
the articles by means of comments or via the website forum. So emerging content
generated by a community of users-professionals ("user-generated content") will be
an added value to complement the core of a purely scientific content of the database.
A beta-version of the COR is available at http://neo.euromise.cz/cor.
Acknowledgements. This work was partly supported by the project 1M06014 of the
Ministry of Education CR and the research plan AV0Z10300504.
References
1. Eddy, D.M.: Evidence-based medicine: a unified approach. Health Affairs (Project
Hope) 24(1), 9–17 (2005)
2. Moyer, V.: Evidence based medicine: is it practical? Arch. Dis. Child. 89(5), 399–400
(2004)
3. King, J.F.: A short history of evidence-based obstetric care. Best Practice & Research.
Clinical Obstetrics &Gynaecology 19(1), 3–14 (2005)
4. Straus, S., Haynes, R.B.: Managing evidence-based knowledge: the need for reliable,
relevant and readable resources. CMAJ 180(9), 942–945 (2009)
5. Williamson, J.W., German, P.S., Weiss, R., Skinner, E.A., Bowes, F.: Health science
information management and continuing education of physicians. A survey of U.S.
primary care practitioners and their opinion leaders. Annals of Internal Medicine 110(2),
151–160 (1989)
6. Smith, R.: What clinical information do doctors need? BMJ: British Medical
Journal 313(7064), 1062–1068 (1996)
7. Glasziou, P., Del Mar, C., Salisbury, J.: Evidence-Based Medicine Workbook, 1st edn.
BMJ Publishing Group, London (2003)
8. Gray, J.A.M.: Where’s the chief knowledge officer? To manage the most precious resource
of all. BMJ: British Medical Journal 317(7162), 832–840 (1998)
9. Hoogendam, A., Stalenhoef, A., Robbe, P., Overbeke, A.J.: Analysis of queries sent to
PubMed at the point of care: Observation of search behaviour in a medical teaching
hospital. BMC Medical Informatics and Decision Making 8(1), 42 (2008)
10. Oxman, A.D., Guyatt, G.H.: The science of reviewing research. Ann. N Y Acad. Sci. 703,
125–134 (1993)
11. McColl, A., Smith, H., White, P., Field, J.: General practitioners perceptions of the route to
evidence based medicine: a questionnaire survey. BMJ 316(7128), 361–365 (1998)
12. Guyatt, G.H., Meade, M.O., Jaeschke, R.Z., Cook, D.J., Haynes, R.B.: Practitioners of
evidence based care: Not all clinicians need to appraise evidence from scratch but all need
some skills. BMJ: British Medical Journal 320(7240), 954–955 (2000)
13. Haynes, R.B., Haines, A.: Barriers and bridges to evidence based clinical practice. BMJ:
British Medical Journal 317, 273–276 (1998)
14. Tang, H., Ng, J.H.K.: Googling for a diagnosis–use of Google as a diagnostic aid: internet
based study. BMJ 333(7579), 1143–1145 (2006)
15. Giustini, D.: How Web 2.0 is changing medicine. BMJ: British Medical
Journal 333(7582), 1283 (2006)
16. Karkalis, G.I., Koutsouris, D.D.: E-health and the Web 2.0. In: itab 2006, Congress Center
Du Lac, Ioannina, Greece (2006)
An Agent Based Pervasive Healthcare System:
A First Scalability Study
Johannes Krampf1 , Stefano Bromuri1 , Michael Schumacher1 , and Juan Ruiz2

1
Institute of Business Information Systems,
University of Applied Sciences Western Switzerland
{johannes.krampf,stefano.bromuri,michael.schumacher}@hevs.ch,
http://aislab.hevs.ch/
2
Department of Endocrinology, Lausanne University Hospital, 1011 Lausanne
juan.ruiz@chuv.ch
Abstract. Gestational Diabetes Mellitus (GDM) occurs during preg-

nancy due to an increased resistance to insulin caused by the growth of
the baby. It appears after the 24th week of pregnancy and it is treated
with diet counselling and insulin treatment. In this paper we present
the complete implementation of a Pervasive Healthcare System (PHS)
based on intelligent agents to support continuous monitoring of pregnant
women affected by GDM. Our infrastructure is composed of a mobile in-
terface connecting to a distributed multi-agent system which in turns is
connected to a patient management system. This stores the data pro-
duced during the monitoring phase and present them to the doctors in
charge of the patient. Our system’s scalability is then evaluated to show
the strong and weak points of our approach.
Keywords: Personal Health Systems, Agents, Pervasive Healthcare.
1 Introduction
Gestational diabetes mellitus (GDM)[10] affects 3%–10% of all pregnant women
with no history of diabetes before pregnancy and manifests itself in high blood
sugar levels during pregnancy. Current treatment guidelines [13] consist in diet
adjustment and in anti-diabetic medicines such as insulin and metformin. In
particular, the patient starts the treatment by simply monitoring the levels of
glucose 4 times per day, with one preprandial observation and one postprandial
observation in the morning, and two postprandial observations after the lunch
and after the dinner. Such values are then written in a notebook that is handed
to the doctors twice weekly. According to the behaviour of the physiological
values the doctors may introduce further checks at lunch and dinner, and, if the
glucose values are outside the boundaries, start the treatment with metformin
or insulin. If not treated, GDM may have severe risks for the mother, who may
develop high blood pressure and protenuria (preeclampsia) [14], and for the baby,

This work has been partially funded by the Hasler Stiftung and by the Nano-Tera
Consortium.

An Agent Based Pervasive Healthcare System: A First Scalability Study 129
who may become large for the gestational age (macrosomia), with complications
at delivery and later in life.
Rather than checking the patient once or twice weekly, a better monitoring
may allow doctors to assess the situation of the patient and propose the cor-
rect treatment. One approach to continuous and intelligent patient monitoring
is represented by pervasive healthcare [12]. The goal of a pervasive healthcare
system (PHS) is to break the boundaries of hospital care, allowing patients to
be monitored while living their day-to-day life and to keep in touch with health-
care professionals. Due to its distributed nature, a PHS is faced with three main
challenges: scalability, accuracy and security. Scalability is important for PHSs
as these systems must be able to serve many patients at the same time, without
experiencing disruption due to high loads. Secondly, an accurate PHS should
be able to filter information efficiently in order to save time to the healthcare
professionals and produce alerts only when needed, with a good trade-off be-
tween false positives and false negatives. Finally, security is also an important
dimension for a PHS as it deals with medical data, which is sensitive data.
In this paper we present a PHS to monitor patients affected by GDM.
A previous version of this system was presented in [2], where we modelled a
distributed agent-based PHS. We have chosen agents as a modelling abstraction
for our PHS as they are understood to be autonomous software entities, that act
proactively and pursue a set of goals [15] in an intelligent way, by applying AI
reasoning techniques. Using multi-agent systems (MAS) abstractions to model
PHSs is beneficial as this programming paradigm is well suited for distributed
systems, due to the autonomy property of the agents, and thanks to distributed
event based approach that these systems take into consideration to model the
interactions between the agents and the other available resources [3].
In [2] we have already provided a first validation of the accuracy of the noti-
fications provided to the health professionals by our intelligent agents. On one
hand, in this paper we present the full implementation of our PHS for GDM,
evaluating the scalability of our system and illustrating how healthcare profes-
sionals can utilise the functionalities of our tool. On the other hand, the security
of our PHS will be evaluated in future publications as the system is currently
being audited for security at the Lausanne University Hospital, although in this
paper we also present how we secured the interfaces of our PHS. The remainder
of this paper is structured as follows: Section 2 discusses the functionalities of
the components of our system; Section 3 discusses an evaluation of our PHS in
terms of its scalability; Section 4 puts our work in comparison with relevant re-
lated works; finally Section 5 concludes this paper and draws the lines for future
work.
2 A Personal Health System for GDM

Fig. 1 shows that our system is composed of three main components, which are
the Mobile Infrastructure (MI), the Agent Environment (AE) and the Patient
Management System (PMS). Furthermore, these components are interfaced be-
tween each others by means of a mediator component, realised as a Web service
130 J. Krampf et al.
:TrustedServer
:ApplicationServer
Service Layer Data Access Layer

External Services Internal Services
Patient
4
1 2 1 LDAP
Mobile phone Doctor's interface Authentication
Service
:WebServer
1 SSL +
Reverse Proxy
3
Web browser Data access
Mobile connector Database
service
4 1
Personalised Patient Monitoring Component

Doctor Agent Environment & Multiagent System
Legende
1 Mutually authenticated HTTPS 3 Unix sockets (local)
2 AJP local 4 Mutually Authenticated TLS
Fig. 1. The Pervasive Health System Logic Architecture and the Security Interfaces
Data Gateway connector that accepts HTTPS requests. The MI component col-
lects the physiological data of the patient and delivers such data to the AE
component and to the PMS component.
The AE component utilises logic programming to model intelligent agents that
filter the data submitted to the PMS and provide alerts in case of significant
events, such as a possibility of preeclampsia in the patient or a high level of blood
sugar that requires a treatment adjustment. The AE system is subdivided in cells
associated to an area of a real city where the patients connect with their mobile
phones to produce their physiological data, that are then evaluated by intelligent
agents. The patients are represented in the AE as avatars that can communicate
to a personal intelligent agent, embodied in the AE. This representation of the
patient is convenient as we can reuse the AE communication and notification
facilities to interact with the intelligent agents situated in it. To every patient
avatar we associate an intelligent agent whose cognitive architecture will be
explained later in this Section. Finally the PMS allows the doctors to visualise
the patient’s data, to modify its treatment and to visualise the alerts produced
by the AE.
The three tier logic architecture shown on the top of Fig. 1 translate then to a
four tier architecture as shown on the bottom of Fig. 1. In particular, the mobile
phone and the Web browser represent the presentation layer, the reverse proxy
and the Web server represent the Web application layer, the agent environment
represents the business logic layer while the database represents the data layer.
The Web application layer accepts outside secure connections only on the
HTTPS port. It connects business logic and data layers. Caretakers and patients
use client authenticated HTTPS to connect to the system. A second authentica-
tion factor is provided by the combination of user name and password. Internal
components use local in-memory or mutually authenticated TLS connections to
communicate with each other. The data base partition is encrypted to protect
against physical access to the hard disk. User access to resources is restricted
by membership in one of the three groups users, caretakers and administrators.
Access to patient data is further restricted by an access control list which only
allows caretakers who treat a patient to access this patient’s data. All actions
are logged including IP address, user name, resource and success of the action
to provide an audit trail.
2.1 The Mobile Infrastructure

The MI component is based on Android, and it is provided to the patient through
mobile phones and tablets to introduce their physiologic data and symptoms
associated to GDM: blood pressure, blood glucose, weight, pulse, blurred vision,
epigastric pain, oedema in the legs, dyspnoea, chest pain, head ache. Such an
interface allows the patients to see if their physiological values meet the targets
for the week, with a set of pie charts as shown in Fig. 2. At the same time the MI
is also built with a synchronisation approach to avoid data loss: whenever the
connection with the AE is impossible, the MI saves the data in a local database.
When the connection with the AE is possible again, the data stored in the mobile
phone is submitted for storage in the PMS.
Fig. 2. The Mobile Interface
2.2 The Agent Environment and the Patient Management System

Our PHS makes use of intelligent agents to analyse and filter the physiological
data produced by the patients. In particular, we decided to include the GOLEM
agent platform in our system as it currently implements the patterns of Dis-
tributed Event-Based Systems (DEBSs) as described in [2]. By means of these
patterns, the agents in the GOLEM agent platform can subscribe to events pro-
duced by the patients and the GOLEM platform will take care of notifying such
events when they take place.
Fig. 3. Caretaker Agent Mind Architecture
In the particular case of our PHS, the patients are represented in the AE as
avatars that can communicate with a personal caretaker agent, whose architec-
ture is reported in Fig. 3. Every agent is deployed in GOLEM in a container.
A GOLEM container represents a portion of the distributed agent environment
which in this case is associated with a portion of the real environment, in order
to distribute the load of the requests of the patients. This topology was chosen
because we imagine that this system could work in synergy with the actual cel-
lular network. As described in [2], every caretaker agent has a cognitive model
with a deductive and an abductive part, whose specification is shown in Fig. 3.
The deductive rules are specified in Event Calculus [8], to describe the evo-
lution in time of the patient physiological values. Such rules specify how the
treatment of the patient should evolve. For example rule R6 specifies that, when
the patient had high glucose in the postprandial observations, then the agent
suggests to introduce further preprandial observations. Similarly, if the patient
is already in a 6 checks per day regime, then the agent suggests the doctors
to introduce a slow insulin in the morning to tackle the values that are out
range. The abductive rules take into consideration the symptoms of the patient
to provide alerts of macrosomia or preeclampsia to the doctors. In particular,
for preeclampsia, we also provide the probability of adverse outcome using the
fullPiers model [14], also reported in Fig. 3. To be able to provide this probabil-
ity, the agent connects to the PMS using the GOLEM middleware to download
the blood samples needed by the fullPiers model and introduced in the system
through the PMS. Further details about the agent cognitive model and its ac-
curacy are reported in [2] and we refer the interested reader to this publication.
The Patient Management System allows healthcare professionals to visualise
and analyse data as well as to introduce new data gathered during a patient’s
visit. The PMS is a hybrid application incorporating both elements of a clas-
sic server side Web application and a modern AJAX-powered client side Web
application.
Fig. 4. The Patient Overview Page
After logging in to the PMS, healthcare professionals can visualise a dynamic

patient page, shown in Fig. 4. The patient page is divided into a static side
bar on the left and a changing content area on the right side of the window.
When showing different information about the patient, an HTML fragment will
be loaded and replace the content area. Graphs are handled completely on the
client side. There is a maximum of about 1700 data points, assuming 6 daily
blood sugar measurements during 40 gestational weeks, for a graph. Processing
and displaying this number of data points is almost instant on modern browsers
and allows for better interactivity than server side processing would. Healthcare
professionals can interact with the graphs by showing a specific time period,
changing this period and changing the view on the data. There exist two further
views on the data beside the point view in Fig. 4, which are a percentage view,
showing the data points in a week which are below, inside and above the normal
range of a physiological value, and and a distribution view that shows the 9th,
25th, 50th, 75th and 91th percentile for each week as well as outliers. These views
were created to fulfil the different information needs of doctors at the Lausanne
University Hospital.
3 Evaluation
To evaluate our solution, we measured the performance for HTTPS requests
with different requirements on the application and database. Our goal in evalu-
ating our PHS was to understand if the system could support the traffic load of
the patients of an hospital of a medium sized city, such as the city of Lausanne
in Switzerland, where we plan to perform field tests. Also, another goal of this
evaluation is to understand what is the maximum amount of patients that we
can serve before having to introduce load balancing techniques in the PHS. We
therefore perform our evaluation on those components representing a bottleneck
of the current architecture. We do not perform an evaluation on the agent en-
vironment as this is based on the GOLEM platform, whose performances have
been previously evaluated in [3], showing that the system can scale up with the
number of GOLEM containers spawned for the application. For the tests, we
ran our PHS on a 3 GHz Intel Core 2 Duo processor, 4 GB RAM and Ubuntu
10.04.

!
"! !

# $ % & '(

!
$ ) !
*

Fig. 5. Requests/second results and 90th percentile response time results
The tests were performed using the ApacheBench [1] utility, which works by
performing a defined number of requests to a specified URL and measuring vari-
ous values such as response time or transferred bytes for each request. After fin-
ishing the benchmark, ApacheBench shows a statistical analysis of all requests,
showing transfer rate and mean time per request. To prevent the benchmark
utility from influencing the results, it was executed on a notebook with a 1.6
GHz Intel Core 2 Duo processor, 1 GB RAM running Linux with a direct cable
connection to the server, to minimise influences caused by network latency. We
tested different usage scenarios stressing different parts of the system. All re-
quests were executed at 1 to 100 concurrent requests to simulate different usage
load. Values were recorded during a 60 seconds stress testing period.
The curves on the left of Fig. 5 show the response time to different concurrent
calls performed in the system. In particular, we show the time to: retrieve the data
for glucose in the PMS; retrieve a patient history; insert glucose values from the
mobile phone; retrieve a patient summary page; access the login page and patient
page. We also assume that concurrent calls coming from different patients are dis-
tributed in different cells of the agent environment. The requests per second values
increase with higher concurrency levels until a plateau is reached. At this point
the server becomes overloaded and the response time increases. The graph of re-
quests to the login page shows the plateauing behaviour when reaching about 1400
requests per second. By looking at the detailed values we discovered that the re-
sponse time increase between 400% and 500% when comparing 20 and 100 concur-
rent connections while the processed requests per second are virtually unchanged.
The curves on top of Fig. 5 also show the plateauing behaviour of the requests with
database activity. The maximum value for requests per second is reached between
10 and 20 concurrent requests. This suggests a database related limit in concur-
rency. When increasing the concurrency from 10 to 15, the glucose insertion from
the mobile phones experiences a sharp fall from 130 to 20 requests per second.
The message queue in the Web service data gateway interface is the limiting fac-
tor here: messages are acknowledged in order to ensure message delivery and with
many concurrent requests messages cannot be acknowledged fast enough.
The 90th percentile response time charts on the right of Fig. 5 show the
expected maximum response time for 90% respectively of all requests. The re-
sponse time begins to be above 1 second and noticeable by a user at about 50
concurrent requests for requests with medium database activity and at about 15
concurrent requests for database heavy requests.
Each patient will transmit a number of values each day: Twice daily blood
pressure and four to six times blood sugar. Furthermore, the patient will trans-
mit one weight value per week and she will report symptoms when she experi-
ences them. We are interested in the maximum number of requests in a short
time period and will make the pessimistic assumptions that during one second
of usage of her mobile phone the patient transmits symptoms, blood pressure,
blood sugar and weight at the same time in the morning. The maximum num-
ber of requests/second for a user is therefore 4 requests/second. The worst case
scenario is all 10 patients of a planned pilot study making their 4 requests con-
currently in the same second, leading to 40 requests/second with 10 concurrent
connections. When producing Fig. 5 we found that the system is capable of 132
requests/second for a concurrency level of 10. As at the Lausanne University
Hospital, that serves the Canton Vaud in Switzerland, there are a maximum
of 5–6 patients with GDM at the same time, the system we defined is viable to
deal with the load experienced by a big sized university hospital. To estimate the
maximum number of users the system can serve, we modify our assumptions to
assume an uniform distribution of the 4 requests over the course of 30 minutes.
We will furthermore use 20 requests/second as the system’s performance due to
performance drop off at higher concurrency levels. This results in about 0.002
requests/second per user (Eq. 1) and 9000 patients (Eq. 2) with 20 concurrent
connections (Eq. 3). This allows us to consider usage for the whole canton Vaud.
The canton has a population of 700,000 [11] and 9.4 births per 1000 inhabitants
per year. This results in 6580 births per year (Eq. 4) which means that the
system can theoretically monitor all pregnant women in the Canton of Vaud.
4 request × user 1 request × user
= (1)
30 ∗ 60 second 450 second
request 1 request × user

20 / = 9000 user (2)
second 450 second
request request
(4 × 9000 user)/30 × 60 second = 20 (3)
user second
9.4 birth birth
700000 inhabitant × = 6580 (4)
1000 inhabitant × year year
4 Related Work
From the related work stand point, several attempts have been done in the past
to combine agent technology with the healthcare domain. The systems described
by Huang et al. in [7] and by Hammond and Sergot in [6] use symbolic reasoning
over clinical workflows to manage oncological patients within a healthcare in-
stitution and to simplify the management of clinical trials. Larson et al present
Guardian in [9], an early attempt to provide an agent-based system for medical
monitoring and diagnosis. Guardian uses a tuple space based approach where
cognitive agents with a properly programmed knowledge base, provide a diag-
nosis for situations such as liver failure and hypothermia. In [4] Ciampolini et
al present a distributed MAS to deal with distributed diagnosis performed by
heterogeneous distributed abductive agents. In Ciampolini’s approach the di-
agnosis is provided in term of probabilities, although they do not consider a
realistic model for their experiments. The ASPIC project [5] has developed an
architecture based on argumentation theory for an autonomous agent that single-
and multi-agent healthcare applications can use. Evaluation scenarios focus on
the management and treatment of people with heart disease. With respect to
the systems reported above, our contribution is twofold: first of all we devel-
oped a practical system that takes into consideration scalability and security
issues following the needs of medical doctors at Lausanne University Hospital;
secondly, for our intelligent reasoning agents, we also utilise clinical models like
the fullPiers [14], whereas the systems mentioned above lack this approach.
5 Conclusion and Future Works

In this paper we presented a fully implemented Personal Health System for
the monitoring of GDM and alerting of treatment adjustment suggestions and
continuous diagnosis of conditions related to GDM. This system is composed
of a mobile infrastructure used by patients, a distributed agent environment
and a patient management system for medical professionals in charge of the
patients. We evaluated the performances in terms of scalability of our system
demonstrating the feasibility of the approach in the case of GDM. Future work
implies the evaluation of security interfaces amongst the different layers of our
infrastructure and the produced data. Another approach to future work is to have
a different PHS to doctors notification system, where the notifications are not
just produced in the Patient Management System, but also imply the submission
of emails or SMS to the medical professionals.
References
1. ApacheBench, http://httpd.apache.org/docs/2.2/programs/ab.html
2. Bromuri, S., Schumacher, M., Stathis, K., Ruiz, J.: Monitoring gestational dia-
betes mellitus with cognitive agents and agent environments. In: Proceedings of
the 2011th IEEE/WIC/ACM International Conference on Intelligent Agent Tech-
nology (IAT 2011) (August 2011)
3. Bromuri, S., Stathis, K.: Distributed Agent Environments in the Ambient Event
Calculus. In: DEBS 2009: Proceedings of the Third International Conference on
Distributed Event-based Systems. ACM, New York (2009)
4. Ciampolini, A., Mello, P., Storari, S.: Distributed medical diagnosis with abductive
logic agents. In: ECAI 2002 Workshop on Agents in Healthcare, Lione (2002)
5. Fox, J., Glasspool, D., Modgil, S.: A canonical agent model for healthcare appli-
cations. IEEE Intelligent Systems 21, 21–28 (2006)
6. Hammond, P., Sergot, M.: Computer support for protocol-based treatment of can-
cer. The Journal of Logic Programming 26(2), 93–111 (1996)
7. Huang, J., Jennings, N.R., Fox, J.: Agent-based approach to health care manage-
ment. Applied Artificial Intelligence 9(4), 401–420 (1995)
8. Kowalski, R., Sergot, M.: A logic-based calculus of events. New Gen. Comput. 4(1),
67–95 (1986)
9. Larsson, J.E., Hayes-Roth, B.: Guardian: An intelligent autonomous agent for med-
ical monitoring and diagnosis. IEEE Intelligent Systems 13, 58–64 (1998)
10. Serlin, D.C., Lash, R.W.: Diagnosis and management of gestational diabetes mel-
litus. Am. Fam. Physician 80, 57–62 (2009)
11. Swiss Statistics. The population of Switzerland 2009 (2010),
http://www.bfs.admin.ch/bfs/portal/en/index/themen/01/22/
publ.Document.136821.pdf
12. Varshney, U.: Pervasive Healthcare Computing: EMR/EHR, Wireless and Health
Monitoring. Springer Publishing Company, Incorporated (2009)
13. Vogel, N., Burnand, B., Vial, Y., Ruiz, J., Paccaud, F., Hohlfeld, P.: Screening
for gestational diabetes: variation in guidelines. Eur. J. Obstet. Gynecol. Reprod.
Biol. 91, 29–36 (2000)
14. von Dadelszen, P., et al.: Prediction of adverse maternal outcomes in pre-eclampsia:
development and validation of the fullPIERS model. Lancet 377, 219–227 (2011)
15. Wooldridge, M.: MultiAgent Systems. John Wiley and Sons (2002)
An Agent-Based Approach to Real-Time Patient
Identification for Clinical Trials
Gareth Tyson1 , Adel Taweel1 , Simon Miles1 ,

Michael Luck1 , Tjeerd Van Staa2 , and Brendan Delaney1
1
King’s College London, UK
2
General Practice Research Database, UK
Abstract. Patient recruitment for clinical trials is expensive and has

been a significant challenge, with many trials not achieving their recruit-
ment goals. One method that shows promise for improving recruitment
is the use of interactive prompts that inform practitioners of patient el-
igibility for clinical trials during consultation. This paper presents the
ePCRN-IDEA recruitment system, which utilises an agent-based infras-
tructure to enable real-time recruitment of patients. In essence, whenever
patients enter a clinic, the system compares their details against eligi-
bility criteria, which define the requirements of active clinical trials. If
a patient is found to be eligible, a prompt is raised to notify the user.
In this way, it becomes possible for recruitment to take place quickly
in a cost effective manner, whilst maintaining patient trust through the
involvement of their own health care practitioner.
1 Introduction
Clinical trials are the gold standard by which medical research is evaluated.
They are used to study various aspects of medical science, as well as being a
vital stage in the deployment of new drug treatments. Currently, however, such
trials are frequently unsuccessful at recruiting sufficient patients. A review of the
UK Medical Research Council found that only 31% of trials actually recruited to
their planned target, with 30–40% of costs arising during the recruitment phase
alone [1]. This is because discovering and contacting eligible potential recruits is
both logistically and legally challenging. Consequently, many research projects
take far longer to complete than is desirable, resulting in an unnecessary burden
for those who could potentially benefit from the results.
The main challenge for patient recruitment lies in locating and contacting
patients in a sufficiently timely manner to allow them to participate. However,
the ease with which this can be done varies dramatically with the type of trial;
for instance, recruitment can be challenging for trials that have high recruit-
ment targets or complex eligibility criteria. Currently, recruitment is performed
in a highly laborious manner, which is ill-suited to the above situations. It often
involves a human recruitment agent visiting clinics in an attempt to locate suit-
able patients (e.g. asking practitioners or searching local medical records). This
creates significant overhead as it is both slow and costly, as well as non-scalable

An Agent-Based Approach to Real-Time Patient Identification 139
for most trials. For example, a trial investigating rare ailments might need an
extensive number of visits to locate sufficient patients.
In consequence, it is of paramount importance to address the recruitment
challenges within trials to ensure the future efficacy of medical research. It is
therefore necessary to find a scalable way by which eligible patients can be
discovered. To address this, we propose replacing human agents with software
agents that permanently reside within clinics, with the aim of searching for
patients who might satisfy trial eligibility criteria. Through this, the agent could
inspect patient information in real-time to ascertain eligibility before presenting
notifications to local practitioners. Importantly, by using a software agent this
could be done rapidly within a consultation before a patient has left, thereby
shortening the recruitment lifecycle (as well as enabling trials based on incidental
cases). The paper’s contributions are therefore as follows:
– A critique of existing clinical trial recruitment approaches, highlighting that
current techniques are both slow and expensive.
– An agent-based distributed architecture called ePCRN-IDEA that enables
real-time recruitment of patients, whilst avoiding the key limitations of ex-
isting approaches.
– A procedure by which software agents can guide the recruitment of patients
to their most appropriate trials.
The rest of the paper is structured as follows; first the background to the research
is discussed in Section 2 before detailing the ePCRN-IDEA recruitment system
in Section 3. Following this, a discussion is presented in Section 4, alongside
future work and conclusions.
2 Background and Related Work
2.1 Recruitment in Clinical Trials

Clinical trials are a challenging stage in the research of clinicians due to the
complexity of recruiting patients for participation. Many types of trials can suf-
fer from such difficulties; for instance, trials that have potential recruits who are
widely distributed over many clinics (e.g. primary care) are extremely difficult to
recruit for due to the intense resource requirements. Similarly, trials with certain
types of patient eligibility criteria can be difficult to recruit for; for example, a
trial dealing with incidental/acute conditions would need a practitioner to imme-
diately identify patients in real-time. This can be highly challenging, especially
when dealing with complex eligibility criteria or trials that require immediate
actions (e.g. a change of drug treatments). This has a significant impact on med-
ical research and stunts potential life-saving advances. Recruitment is currently
performed in a number of ways; examples include:
– Recruitment Visits: Using recruiters to visit or contact practices so that they
can search local medical records and/or ask local practitioners.
140 G. Tyson et al.
– Advertisements: Using posters, web sites, mailing lists or newspaper adverts

describing what is required for the trial.
– Practitioners: Using practitioners to suggest patients who might be eligi-
ble. Alternatively, practitioners may be asked to look for patients (during
consultations) who match the criteria in real-time.
Clearly, the above approaches are slow and often quite expensive. The predom-
inant approach of employing recruiters to contact and visit clinics is laborious
and often infeasible, especially when dealing with the above types of trials. For
instance, different clinics often use different database systems, thereby requiring
a high degree of training for recruiters; something that is exacerbated by the
limited in-house IT training in many clinics. Further, attempts to standardise
this database search procedure (e.g. MIQUEST) are often poorly understood
by researchers, whilst such standards can often be undermined by variations in
coding and medical dictionaries (e.g. if multiple databases use different seman-
tics). Consequently, there is little automated support for finding eligible patients,
often leaving recruitment very much as an ad-hoc process that falls outside of
recognised standards; Moreover, there is little infrastructure available that can
be exploited by clinicians to assist in the stage.
As an alternative to the recognised approaches, Embi et. al. [2] propose the
use of a Clinical Trial Alert (CTA) system, through which practitioners could be
notified while they are in consultation with a patient who is eligible for a trial.
Their work shows that significant increases in recruitment could be achieved.
However, their pilot study was only deployed in a single clinic with a single trial,
in an attempt to study the improvements in recruitment. Other similar tech-
niques have also seen only limited large-scale testing, e.g. [3]. Consequently, a
number of issues (e.g. scalability) have not been investigated, leaving the sole
focus on the recruitment outcomes. To address these concerns, this paper there-
fore focusses on more infrastructural aspects for enabling a large-scale trial alert
system, specifically through the use of intelligent agents.
2.2 Agents in Healthcare

Agents have emerged as a prominent technology for handling a range of real-
world problems [4]. An agent can be defined as “a computer system that is
situated in some environment, and that is capable of autonomous action in their
environment in order to meet its delegated objectives” [5]. Agents in healthcare
have seen widespread investigation; in general, their use attempts to address en-
demic issues such as (i) distributed information and expertise, (ii) unpredictable
dynamics, and (iii) uncertainty in reasoning and data.
For example, MAID [6] is an agent-based system for integrating heteroge-
neous data sources within a hospital environment. The hospital studied had
24 departments, each using their own information systems. To address this,
agents were constructed to interoperate with each system to monitor changes
and retrieve data for insertion into a central repository. In a subsequent work,
HealthAgents [7] went beyond MAID to also enable decision support, specifically
for diagnosing brain tumours. A range of agent-based systems have also been pro-
posed for handling distributed expertise. These includes using agents to enable
better communication between healthcare workers based on ambient informa-
tion, e.g. their role, location etc. [8], as well as using agents to remotely monitor
patients [9][10]. These systems also often involved data analysis; S(MA)2 D [10],
for instance, uses statistical analysis to cluster patients into similar groups. This
ability to scalably perform data analysis in real-time, clearly, also shows poten-
tial for enabling the type of eligible patient identification discussed previously.
Despite this, so far little work has been performed into using agents to improve
clinical trial recruitment. Consequently, the rest of this paper explores exploiting
the properties of agents to enable real-time recruitment of patients to trials.
3 ePCRN-IDEA Recruitment System

This section presents the ePCRN-IDEA recruitment system, which is designed
to enable real-time identification of eligible patients. We first present the over-
all architecture before describing the individual components, focussing on the
behaviour of the clinic-based agents.
3.1 Overview
The core goal of the ePCRN-IDEA recruitment system is to improve patient
recruitment. In order to do so, researchers must formally define the eligibility
criteria of participants, then distribute it to software agents that reside locally
on practitioners’ computers. These software agents listen to interactions between
the practitioner’s local Electronic Healthcare Record (EHR) database and the
user in an attempt to locate patients who are eligible for trials. Importantly,
this occurs in real-time during the consultation, thereby allowing a pop-up to be
generated, notifying the practitioner of the patient’s eligibility. In this way, the
patient can be instantly consulted regarding the trial and, if interested, recruited
via a web interface. The key architectural entities in the system are as follows:
– LEPIS: An agent that resides at primary care practices and investigates the
eligibility of any present patients, termed the Local Eligible Patient Identi-
fication Service.
– CCS: A point of storage and distribution that allows clinical researchers to
inject new trials into the system, termed the Central Control Service.
– CTMS: A website that handles the actual recruitment process once an el-
igible patients has been discovered, termed the Clinical Trial Management
System.
Prototype implementations of all these components have already been developed.
Fig. 1 provides an overview of these, as detailed in the rest of this section.
142 G. Tyson et al.
Fig. 1. Overview of ePCRN System
3.2 Central Control Service (CCS)

A trial store maintains a repository of active trials within the system, as entered
by any clinical researchers wishing to recruit patients. Trials are stored using
a standard model (the PCROM standard [11]), which defines the data format
used to represent the various aspects of the trials. This includes a description
of the trial as well as the eligibility criteria, which can be based on a variety of
aspects ranging from a pre-computed list of eligible patient identifiers to complex
diagnosis information. The trial store is realised within the ePCRN-IDEA system
through the Central Control Service (CCS), which is a service used to manage
all the trials. When a new trial is created, it is injected into the system via the
CCS and stored in a MySQL database back-end before being made accessible
to the appropriate (and authorised) parties — namely, the recruitment agents.
These are accessed securely using an encrypted SQL connection initiated by each
agent to the CCS.
3.3 Local Eligible Patient Identification Service (LEPIS)

The Local Eligible Patient Identification Service (LEPIS) is a Java-built soft-
ware agent that resides on every practitioner’s PC (who is authorised to recruit
patients). The purpose of this agent is to actively discover any eligible patients
who might be seen by the practitioner. It therefore operates as a replacement for
the existing human agents who visit clinics in an attempt to locate suitable pa-
tients. Every LEPIS agent is required to obtain two distinct sets of information.
First, it must acquire information about patients in real-time as they enter the
clinic. Second, LEPIS must also acquire the necessary trial information (from
the CCS) so that it can compare patients against each trial’s eligibility criteria;
ideally, eligibility should also be computed locally to avoid unnecessary delays
or bottlenecks.
Accessing Patient Information. To discover the eligibility of a patient for a

trial, it is first necessary for LEPIS to gain access to any pertinent information
related to the patient. Whenever a patient enters a clinic, the practitioner opens
his or her medical record using the Electronic Healthcare Record (EHR) database
on their desktop computer. This medical record contains a range of demographic
and medical information about the patient, allowing practitioners to retrieve
information about the patient during the consultation, as well as enter new
information. This offers an existing platform through which a recruitment agent
can access information about patients. We have modified a popular EHR system,
Vision [12], to interact with LEPIS. Whenever a patient record is opened or
modified, the information is passed to LEPIS (through a standard file using a
shared XML schema) so that it can attempt to locate trials for which the patient
may be eligible. Information is coded using standard Read Codes and Multilex
Drug Codes to allow LEPIS and the EHR to understand each other. LEPIS
is therefore given real-time access to information about any patients who are
currently in consultation. Although, evidently, eligibility criteria is limited to
those attributes provided by the EHR, which can vary based on both policy and
EHR implementation (many EHR vendors exist).
Accessing Trial Information. To allow a LEPIS agent to compute a patient’s
eligibility, it must first gain access to trial information. Ideally, this should be
stored locally to enable real-time eligibility checks within a consultation. How-
ever, evidently, this is largely infeasible with the huge number of active trials run-
ning; e.g. clinicaltrials.gov currently lists well over 100,000 trials. Consequently,
it is necessary for each agent to independently select the most appropriate trials
for its clinic and practitioner. Each agent thus maintains a set of trials T of size
n, as limited by the host’s local resources. An agent therefore selects n based
on the capabilities of its host, by performing eligibility checks on a random set
of trials repeatedly for one second; n is then set as the number of iterations. It
then keeps a persistent record of all patient identifiers, Read Codes and Multi-
lex Codes provided by the EHR to build up a profile of the clinic. Using this
information, n trials are retrieved from the CCS through the following process:
1. LEPIS attempts to retrieve a set of n trials from the CCS containing:
(a) p trials that includes a known patient registered within the clinic (p =<
n);
(b) if p < n, c trials that includes coded information previously encountered
within the clinic (c =< n − p); and
(c) if p + c < n, r randomly selected trials (r =< n − p − c).
2. Remove any trials that are fully recruited.
These two steps are repeated throughout an agent’s lifetime with a configurable
interval, which is set to 24 hours by default.
Computing Eligibility and Generating a Popup. When LEPIS acquires
patient information from the EHR, it must compare it against the eligibility
criteria of any known trials. This is a simple process that currently involves
iteratively computing eligibility for each known trial and then selecting a random
one if multiple are found. A popup is then generated to notify the user. Fig. 2
shows a screenshot of the user interface.
144 G. Tyson et al.
Fig. 2. Screenshot of LEPIS Popup
3.4 Clinical Trial Management System (CTMS)
If a patient is interested in being recruited for a particular trial, it is then neces-

sary to actually perform the recruitment procedure. This is not handled by the
local agent; instead, an external website is used, called the Clinical Trial Man-
agement System (CTMS). The CTMS is securely accessed by the practitioner
and then used to register the patient’s interest in being recruited. Any necessary
steps can then be taken, e.g. contacting the patient, recording information etc.
4 Discussion, Future Work and Conclusion
The ePCRN-IDEA recruitment systems differs substantially from previous work

in that it is agent-based. Our early evaluation shows that the agent-based ap-
proach is promising and has several potential advantages over traditional client-
server approaches (e.g [2]). Through the use of agents, intelligence is decen-
tralised within the system so that both computation and decision making is
independently performed by each agent.
Primarily, we have used the approach to enable superior scalability. This is
critical, especially for patient recruitment in primary care. Even for common dis-
eases, eligible patients are thinly spread across many practices, which potentially
number several thousand. In the UK, alone, there are more than 10,000 practices.
A client-server approach could be realised in two ways: either all patient infor-
mation could be transmitted to a server, where it centrally computes eligibility;
or all trial information could be transmitted to all clinics for local computation.
The former is non-scalable as well as dangerous in terms of privacy and secu-
rity. The latter, however, is also highly non-scalable as the number of trials (e.g.
greater than 100,000) alongside the size of each trial description (e.g. 0.5 MB)
makes it impossible for all clinics to know of all trials. Consequently, to address
this, we embed intelligence within the agents to learn how to best select trials
for their host clinic, exploiting the local knowledge (and computational abilities)
of each agent, rather than burdening a central point.
From our initial phase-1 prototype we have identified a number of future lines
of work. First, we aim to complete a full system deployment within the UK
primary healthcare system, thereby enabling a detailed quantitative evaluation.
Beyond this, we also intend to extend the agent capabilities. Key research lines
include, (i) inter-agent collaboration: allowing agents to build societies to better
enable information and resource sharing (e.g. based on disease areas, localities);
(ii) interface adaptation: allowing agents to learn (and share) the behaviour of
users to adapt interaction; and (iii) trial negotiation: allowing agents to negotiate
with each other to best distribute trials based on runtime conditions.
References
1. McDonald, A., Knight, R., Campbell, M., Entwistle, V., Grant, A., Cook, J., El-
bourne, D., Francis, D., Garcia, J., Roberts, I., Snowdon, C.: What influences
recruitment to randomised controlled trials? a review of trials funded by two uk
funding agencies. Trials 7(1), 9 (2006)
2. Embi, P.J., Jain, A., Clark, J., Bizjack, S., Hornung, R., Harris, C.M.: Effect of
a clinical trial alert system on physician participation in trial recruitment. Arch.
Intern. Med. 195(19) (2005)
3. Rollman, B.L., Fischer, G.S., Zhu, F., Belnap, B.H.: Comparison of electronic
physician prompts versus waitroom case-finding on clinical trial enrollment. Jour-
nal of General Internal Medicine 23(4) (2008)
4. Luck, M., McBurney, P., Preist, C.: Agent Technology: Enabling Next Generation
Computing (A Roadmap for Agent Based Computing). AgentLink (2003)
5. Wooldridge, M.: Introduction to Multiagent Systems. John Wiley & Sons, Inc.
(2009)
6. Cruz-Correia, R., Vieira-Marques, P., Costa, P., Ferreira, A., Oliveira-Palhares, E.,
Araújo, F., Costa-Pereira, A.: Integration of hospital data using agent technologies
- a case study. AI Commun. 18 (August 2005)
7. González-Vélez, H., Mier, M., Julià-Sapé, M., Arvanitis, T., Garcı́a-Gómez, J.,
Robles, M., Lewis, P., Dasmahapatra, S., Dupplaw, D., Peet, A., Arús, C., Celda,
B., Van Huffel, S., Lluch-Ariet, M.: Healthagents: distributed multi-agent brain
tumor diagnosis and prognosis. Applied Intelligence 30 (2009)
8. Rodrı́guez, M.D., Favela, J., Preciado, A., Vizcaı́no, A.: Agent-based ambient in-
telligence for healthcare. AI Commun. 18, 201–216 (2005)
9. Koutkias, V.G., Chouvarda, I., Maglaveras, N.: A multiagent system enhancing
home-care health services for chronic disease management. IEEE Transactions on
Information Technology in Biomedicine 9, 528–537 (2005)
10. Rammal, A., Trouilhet, S., Singer, N., Pécatte, J.-M.: An adaptive system for
home monitoring using a multiagent classification of patterns. Int. J. Telemedicine
Appl. 2008, 3:1–3:8 (2008)
11. Speedie, S.M., Taweel, A., Sim, I., Arvanitis, T.N., Delaney, B., Peterson, K.A.: The
primary care research object model (pcrom): A computable information model for
practice-based primary care research. Journal of the American Medical Informatics
Association 15(5) (2008)
12. Vision, http://www.inps4.co.uk/vision/
The Use of Social Bookmarking by Health Care Students
to Create Communities of Practice
Ed de Quincey, Avril Hocking, Josephine O’Gorman, Simon Walker, and Liz Bacon
e-Centre, University of Greenwich

Old Royal Naval College, Park Row London, UK
e.de.quincey@gre.ac.uk
Abstract. Teaching and learning health and social care in a digital age produces
many challenges for students and their teachers. A common hurdle for
healthcare students and practitioners is the sheer amount of information that
they have to make sense of. Another challenge is where this information is
captured and stored, with people utilising personal, as well as institutionally
owned devices. A potential solution to these problems is the use of social
bookmarking applications such as “delicious”, where users can create a
centralised repository of online resources, share them with other users, and view
what others are bookmarking. This paper describes research conducted at the
University of Greenwich involving 160 participants across three Schools and 5
modules, including Health and Social Care who were encouraged to integrate
social bookmarking into their learning and teaching. Participants were
instructed to tag their resources with an appropriate module code tag e.g.
NURS1297 so that a repository of module specific bookmarks was created.
Over a 4 month period, 160 users created 1430 bookmarks with 5032 tags.
Further analysis of the bookmarking behaviour is discussed along with
reflections on the suitability of social bookmarking to create digitally literate
health care communities of practice.
Keywords: Social bookmarking, tagging, eLearning, Web 2.0, communities of

practice, connectivism, heutagogy, digital literacy.
1 Introduction
A common problem for students studying health related subjects in Higher Education
(HE) is that they can access the web on any number of devices which do not
necessarily have automatic syncing of bookmarked web pages enabled. There is also a
related issue of sharing, as well as discovering relevant online resources [1]. A
potential solution to these problems is a social bookmarking application, where users
can create a centralised repository of bookmarked resources, share them with other
users and view what others are bookmarking. The popularity of these systems has
become of increased interest to information architects and has prompted a number of
studies into the use social bookmarking and the related field of collaborative tagging
[2],[3],[4]. Results from these studies suggest that tagging and bookmarking share
Social Bookmarking by Health Care Students to Create Communities of Practice 147
similar features to more traditional indexing systems [2] but also contain extra
dimensions such as tags related to time e.g. “toread” and task or users’ emotional
responses to a document e.g. “cool”, which conventional indexing systems do not
support [2].
One of the most popular online resources that supports social bookmarking is
“delicious”, a social bookmarking website which “allows users to tag, save, manage
and share web pages from a centralized source” [5]. The main advantage of these
services over traditional bookmarking systems such as those included in the majority
of web browsers e.g. “favorites”, is that the need for synchronizing bookmarks
between multiple computers and browsers is no longer required. This is a common
problem for students in HE, in particular those from health and social care courses
who are on placements and who access the web on any number of machines and
devices, in any number of locations. Although a number of browsers have bookmark
synching capabilities (either in-built or via plugins), this is often not available on HE
or NHS maintained machines due to security or privacy concerns. A related benefit is
the ability to categorise resources in multiple categories, by the use of tags, which is
only partially supported by web browsers in the form of “bookmark folders”. A
consequence of the use of tagging that is potentially useful for students, lecturers and
practitioners is the ability to discover and share resources via these tags. For example
tagging resources using the course codes of a degree program gives students a central
access point (e.g. http://delicious.com/tag/NURS1297) for all relevant online
information for that course. If all students then use the appropriate course code tag
when bookmarking during their own personal research, both students and lecturers
will collaboratively produce a list of online resources for that course.
This paper describes the results of a project that aimed to utilise the delicious
website to investigate the use of social bookmarking and tagging in an educational
setting by students and lecturers.
2 Research Questions
The main objectives for this project were to investigate the best methods for
integrating social bookmarking into everyday practice for both students and teachers,
and to then determine behavioral usage and motivations. The following were the
proposed research questions for the project:
1. What patterns of user tagging activity emerge through analyses of tagging
frequency and co-word analysis? (based on [2])
2. What patterns of user bookmarking activity emerge through analyses of the
resources bookmarked and the tags used to bookmark them?
3. What is the temporal distribution of bookmarking during an academic
semester?
4. What types of tags are being used i.e. do students/lecturers utilise task and
time related tags?
148 E. de Quincey et al.
5. What are the levels of use of social bookmarking in relation to resource

discovery i.e. do students browse/follow fellow students bookmarks and tags
to discover resources?
6. What are students and lecturers perceptions of the advantages and
disadvantages of social bookmarking and tagging?
7. What are the motivations for using social bookmarking services?
8. What features are currently missing from social bookmarking websites?
The following section describes the two methods that were used to answer the
research questions described above.
3 Methodology
3.1 Participants
In total 160 people, comprising 5 lecturers and 155 students on 5 modules across the
Schools of Computing and Mathematical Sciences, Engineering and Health and
Social Care, participated in this study. Courses ranged from Masters (Level 7) to first
year undergraduate (Level 4) and comprised of around 10 students to over 100.
Students ranged from novice computer users to relative experts i.e. students who
already had an undergraduate computing related degree. Courses ran from various
points in September to the end of the semester in December 2010.
3.2 Materials
All students were provided with introductory materials, which were developed with a
pilot group, and delivered through lectures, tutorials and YouTube videos. These
resources introduced learners to the social bookmarking project and to the delicious
social bookmarking tool.
An initial interesting finding was the use of the term “bookmarking”. During the
first introductory talk to a group from the School of Health and Social Care, one of
the researchers compared “social bookmarking” to “browser based bookmarking”,
outlining the advantages and disadvantages of both approaches. It became apparent
however that students did not understand what “browser based bookmarking” was,
primarily due to the fact that the majority of students used Internet Explorer as their
default web browser, which utilises the term “favorites” instead of “bookmarks”1.
Following this, the term favorites was used in the introductory material, alongside
bookmarking, to avoid potential misunderstanding.
A further finding at this stage was the need for two different sets of instructions. A
step by step guide was produced with detailed instructions and screenshots of each
stage of the sign up, adding bookmarks and tagging resources process (totalling
around 14 pages). Although a number of students appreciated these materials, others
1
The term “bookmark” is used in the majority of other web browsers such as Firefox and
Chrome.
suggested that they would prefer a one-page set of instructions, which highlighted the
key stages, which was then produced.
3.3 Procedure
In September 2010, all lecturers introduced social bookmarking to their students using
the materials described above. The only prescribed usage was that students and
lecturers were instructed to tag any resources that were related to a particular module
with the appropriate module code e.g. NURS1297. This then enabled a module
specific set of resources to be created and made available at a single URL on the
delicious website e.g. http://delicious.com/tag/NURS1297. Students were encouraged
to use delicious to store and find useful resources for each module, with lecturers
employing a number of strategies to motivate continued use. These included the
production of a number of different visualisations via the tool described in section
3.3.1, as shown in figure 1 below which were shown to the students during lectures
and tutorials to demonstrate the course’s current bookmarking activity.
Fig. 1. Visualisation of bookmarking activity in the form of a tag cloud from NURS1297, a
course entitled “Principles of Learning Disability Nursing across the lifespan”
3.4 Automated Collection and Analysis Tool
One particular advantage for researchers in this field is that in addition to a user
facing service, delicious also enables programmatic access to the information stored
on the site. This is enabled by an Application Programming Interface (API), which
supplies a number of XML/JSON based web services. This means that the collection
and analysis of users’ bookmarking and tagging behaviour can then be automated.
As described by Kipp and Campbell [2], the basic component of delicious is the
bookmark entry made by each user upon encountering a website of interest. In
addition to the URL of the website, the user can enter a title, some notes and a number
of tags. All of these details along with the username and the date the bookmark was
added can be accessed via the delicious API. An analysis package was therefore
created with PHP and MySQL that stored any bookmarks that were tagged with the
relevant module codes, along with the username, the additional tags, any notes and
the timestamp.
This analysis package produces descriptive and visualised statistics, co-word

matrices and frequency data for all bookmarks created as part of this study, as well as
for individual courses.
Descriptive statistics included: the total number of bookmarks created by users; the
most popular bookmarked resources and websites; the total number of tags and the
total number of unique tags.
Visualisations included: annotated time line charts showing the temporal
distribution of bookmarks; tag clouds highlighting the most frequently used tags;
geographical location of bookmarked resources (estimated by the reported location of
the server of the site that had been bookmarked).
4 Results
4.1 Descriptive Statistics
160 users across the 3 Schools created 1,430 bookmarks with 5,032 tags from August
2010 to the end of January 2011. A certain amount of agreement between respondents
was demonstrated with only 1,069 unique tags being used (21%) and the 1,430
bookmarks being comprised of 882 distinct url’s. 58% of bookmarks (829) contained
notes about the resource that had been added by the user. The figure below shows the
normalised temporal distribution of bookmarking activity for the duration of the
project. There were initial peaks of activity in August, with the pilot group and then in
September when social bookmarking was introduced to students in all of the modules.
In the first full week of the semester (w/c 27th September) 468 bookmarks were
created.
Fig. 2. Normalised temporal distribution of bookmarking activity
Activity gradually declined from the initial launch in September with 488
bookmarks being created in October, 180 in November and 71 in December (although
this includes the Christmas holiday). Although all of the modules stopped in
December, the delicious website was still being used by students with 42 bookmarks
being created in January 2011.
The types of tags used were dependent on the domain of the module but generally
give a good descriptive overview of the subject area of that discipline. For example,
when considering the co-word matrices for each tag, the tag “NURS1297”, relating to
the module “Principles of Learning Disability Nursing across the lifespan”, co-occurred
most frequently with the tag “learningdisabilities” (40 times). Within that module, other
frequently co-occurring tags were “communicationdifficulties”, “intellectualdisability”
and “learningdisabilities”.
The number of tags used per bookmark ranged from 1 to 17, the distribution being
shown in Figure 3 below. The majority of bookmarks (266) were tagged with a single
tag, with the average number of tags per bookmark being 3.5.
Fig. 3. Total number of Tags per Bookmark
The following table shows numbers of users with how many bookmarks in total
they created e.g. 2 users created between 50 and 59 bookmarks.
Table 1. Number of bookmarks created by users

Number of Bookmarks Number of Users
140-149 1
50-59 2
30-39 1
20-29 12
10-19 36
1-9 108
Although the majority of users created between 1 and 9 bookmarks, over 59% of
users (94) created 5 or more tagged resources during the duration of the project.
Analysing the health related courses separately revealed similar activity levels and
behaviours to the results as a whole, across the 3 Schools.
4.2 Survey Results

At the end of each module in December 2010, students were asked to complete an
online survey to measure the impact on learning and teaching. 81% of respondents
said that they used delicious to bookmark web pages as good resources with 46%
saying they used delicious to share resources. Around 30% indicated that they used
delicious to find relevant resources. Interestingly only 49% found resources via the
module code tag with 77% finding resources via tags related to the module i.e. subject
related tags. Around 70% of students viewed other students’ bookmarked resources.
84% of respondents stated that they would use the delicious website again, with
89% of those students saying they would use it for University related activities.
Pleasingly, 70% of respondents said that they would recommend delicious to a friend
and 68% said they’d recommend it to other learners. 52% would recommend it to
professional colleagues. Although not a main objective of this project, 49% of
students felt that using delicious had improved their ICT skills.
The qualitative feedback generated from the questionnaire was generally positive,
relating to both storage and discovery of information.
5 Discussion and Conclusion
The results from this study indicate that social bookmarking has a number of positive
outcomes with regards to teaching and learning, across a number of disciplines related
to eHealth. Students and lecturers found the tool to be useful for storing, sharing and
discovering resources. Through the process of using specific course code tags, both
students and lecturers have created their own learning communities or communities of
practice. The formation of these learning communities enables them to share
information relating to their own specific course subjects, with their peers. The year
on year building of a repository of information in effect provides students with an
online searchable database for its members to access [6]. It also provides students the
opportunity to discover and share the views and perspectives of their fellow members.
In this study, students were not limited to accessing their own learning communities;
there were opportunities for students to create and join other learning communities. This
was achieved by users utilising their own specific descriptive words when tagging (as
“module codes” were chosen in the project to establish learning communities), or by
joining existing ones. The latter can be achieved by the students finding others who use
identical tags as themselves, relating to the students’ own interest. This means
individuals can share resources with other ‘like minded’ people who have similar, or the
same interests [7]. Through the tagging process students can discover additional
resources that they may not have necessarily found themselves (77% of students
reported that they found resources in this way), thus leading to a group of people
forming their own learning community [6].
The advantages of learning communities include individual users being able to
access these groups at a time and place which is convenient to them, on a 24/7 basis.
In this study, students were not restricted as to when and where they could access and
use these learning communities, which became increasingly important when students
were on placements. Additionally, through the process of using “notes”, students can
share views with one another about the various resources users have bookmarked,
helping students to develop their critical evaluation skills.
It is noted however, the level and extent of participation as a member within each of
these learning communities is dependent on the individual. Some students chose only to
participate a few times, whilst others used their communities more often. Users do not
receive any feedback about whether their tagged resources have been used by others [6],
perhaps explaining why there are some who chose only to consume the resources rather
than contribute to the community. Although bookmarking activity decreased during the
duration of the project, a key indicator of success is the building of the repository itself,
as opposed to the number of contributors. The work of Ortega et al. [8] suggests that in
collaborative resource creation applications such as Wikipedia, there is a great level of
inequality “with less than 10% of the total number of authors being responsible for more
than the 90% of the total number of contributions”. From this study, supposing that
creating over 10 tagged bookmarks is a reasonable level of contribution, then 33% of
users achieved this level. If we reduce that level to 5 tagged bookmarks then 59% of
users have made an active contribution.
The utilisation of social bookmarking is an example of how learning communities
can be created where its members can store and share a collective range of resources
for others to share. This could be seen as encouraging not only the development of
learner independence and autonomy but a range of related graduate attributes which
are valued by employers. Further to this, the emergence of these communities
demonstrates how social interaction within health care subjects in HE is progressing.
Future work will include identifying specific improvements that could be made to
the delicious website and functionality that is currently missing e.g. improved support
for critical evaluation of resources. The project will be continued in the following
academic year with other modules as well as an investigation into how social
bookmarking can be used within VLE’s such as Moodle.
References
1. Jenson, J.: It’s the information age, so where‘s the information? Why our students can’t
find it and what we can do to help. College Teaching 52(3), 107–112 (2004)
2. Kipp, M.E.I., Campbell, G.D.: Patterns and Inconsistencies in Collaborative Tagging
Systems: An Examination of Tagging Practices. In: Proceedings of the 2006 Annual
Meeting of the American Society for Information Science and Technology, Austin,
November 3-8 (2006)
3. Kipp, Margaret E.I.: @toread and Cool: Tagging for Time, Task and Emotion. In: Proc.
Information Architecture Summit 2007 (2007)
4. Golder, S., Huberman, B.A.: The structure of collaborative tagging systems. HP Labs
Technical Report (2006),
http://www.hpl.hp.com/research/idl/papers/tags/
5. delicious.com (2011), http://delicious.com/about
6. Benbunan-Fich, R., Koufaris, M.: An empirical examination of the sustainability of social
bookmarking website. Information Systems E-Business Management 8, 1310148 (2010)
7. Mason, R., Rennie, F.: E-Learning & Social Networking Handbook: The resources for
Higher Education. Taylor & Francis (2008)
8. Ortega, F., Gonzalez-Barahona, J.M., Robles, G.: On the Inequality of Contributions to
Wikipedia. In: Proceedings of the 41st Annual International Conference on Hawaii
International Conference on System Sciences, p. 304 (2008)
Engagement in Online Medical Communities of Practice
in Healthcare: Analysis of Messages and Social Networks
David Fowler1, Martin Szomszor1, Simon Hammond1,

John Lawrenson2, and Patty Kostkova1
1
City eHealth Research Centre, City University, London, UK
2
Optometry and Visual Science, City University, London, UK
David.Fowler.1@city.ac.uk
Abstract. We report on preliminary analysis on user engagement in two online

medical communities of practice. Despite the communities being independent
of each other, and dealing with different domains (field epidemiology and
therapeutic prescribing for optometrists), there are some clear similarities in the
networks of users, and in patterns of replies to user postings. We also draw
some initial conclusions to help maintain user engagement in these and similar
sites, and we suggest some future lines of research.
1 Introduction
Professional communities of practice (CoPs) have been the cornerstone for sharing
scientific knowledge and professional discourse. The internet has dramatically
changed the way communication and peer networking is managed: little overhead and
flat structures, easy online recording of scientific discussions, higher frequency of
postings, and virtually unlimited geographical coverage of the CoPs. However, online
communities may be vulnerable to stagnation and failure if the support tools are not
suitable, or if key members of the community are not able to take an active role.
1.1 Communities of Practice
The term “community of practice” has many definitions, although it originates in the
work of Lave and Wenger [1]. We will use the definition of CoPs as “groups of
people who share a concern, a set of problems, or a passion about a topic, and who
deepen their knowledge and expertise in this area by interacting on an ongoing basis.”
[2] (also cited in [3]). CoPs may be deliberately created or spontaneously emerge, and
be highly structured or informal [3].
1.2 The Two Communities: FEM Wiki and MSU

We examine two independent CoPs developed around medical scientific internet
portals: FEM Wiki (http://www.femwiki.com), dealing with field epidemiology, and
Medicines Support Unit for Optometrists (MSU, http://www.med-support.org.uk),
Engagement in Online Medical Communities of Practice in Healthcare 155
supporting therapeutic prescribing by optometrists. The user bases are geographically

dispersed (mainly throughout the UK for MSU and throughout Europe for FEM
Wiki). Both sites provide centrally authored information to specialists, and have
means for user discussion. Each was created to order, but FEM Wiki is more highly
structured than MSU. In FEM Wiki, users can directly edit the content, but to
guarantee quality, changes must be approved before the changes are made official. In
MSU, changes can be suggested informally via the forum.
2 Social Network Analysis

We collected the messages that were posted on the discussion forums of the
communities, and extracted networks of users. Each node corresponds to a user, with
arcs linking the nodes of users who were involved in the same discussion (Fig. 1).
Fig. 1. The user networks extracted from FEM Wiki (left), and MSU (right). The nodes with
most connections are highlighted.
Fig. 2 shows the number of connections for each node in the networks, and Table 1
summarises some key statistics. Each network has a number of users who are
involved in many discussions; these seem to be mainly senior project leaders or
administrators. There is an almost linear decline to users who were only involved in
one or two discussions (possibly they only had a specific question that was answered
to their satisfaction). Although the networks that are extracted are not a complete
picture of the knowledge sharing activities in the communities (for example, members
may share knowledge in person or via other media and the network does not measure
the quality of contributions), it may give a reasonable approximation. Users with
many connections are involved in many discussions, and therefore may have more
knowledge and experience to share.
156 D. Fowler et al.
Fig. 2. The numbers of neighbours for each node in Fig. 1
Table 1. Summary statistics for the two communities
FEMWiki MSU
Nodes 23 20
Edges 73 62
Average Degree 6.348 6.200
Diameter 3 4
Average Path Length 1.798 1.816
Graph Density 0.289 0.326
3 Message Analysis
In addition to examining the connections between users, we also looked at the
characteristics of the messages. There were striking similarities between FEM Wiki
and MSU in the distribution of replies to messages. Table 2 shows that the majority of
posts have a small number of replies (the median is 2 for both communities). This
seems to be typical behaviour for online forums, e.g. [4].
Table 2. Summary statistics for the numbers of replies to FEM Wiki and MSU posts
FEM Wiki (33 posts) MSU (45 posts)

Min 0 0
Q1 1 1
Median 2 2
Q3 3 3
Max 18 16
Engagement in Online Medical Communities of Practice in Healthcare 157
4 Discussion and Future Work
We have shown that there are underlying similarities in the user network structure and
distribution of numbers of replies to posts of two independent online CoPs. The two
sites also vary in their organisational structures and editing processes, so these results
might suggest some properties that are shared more widely between online CoPs. This
should provide some useful lines of enquiry, although it will require access to data
from a larger number of online CoPs. We will also need to investigate how the
properties of CoPs vary with size, as our examples were both in the small to medium
range.
The type of analysis in this paper may be helpful in identifying users whose
contributions are critical to keeping an online community active. If such users become
less active (for example, through pressures of other work), there is a risk that the
community will stagnate, and lose other users. There is some evidence that this has
happened recently with the MSU site (although with MSU there was another possible
cause for loss of activity: a spam attack on the discussion forum may have driven
away some users).
We are interested in tracking the activity of online CoPs over time to see how the
user networks vary, investigating what factors may affect the activity, and whether
there is an identifiable “critical point” at which community activity breaks down. We
are currently redesigning the MSU site, and plan to promote the site again to existing
and prospective users in order to increase activity.
Finally, we plan to investigate the factors that affect the user response to forum
messages. Section 3 showed high level similarities, and it will be interesting to see
which types of posts attract most discussion, and to draw comparisons between sites.
References
1. Lave, J., Wenger, E.: Situated Learning: Legitimate Peripheral Participation. Cambridge
University Press, Cambridge (1991)
2. Wenger, E., McDermott, R., Snyder, W.M.: Cultivating Communities of Practice: A Guide
to Managing Knowledge. Harvard Business School Press, Boston (2002)
3. Hara, N., Shachaf, P., Stoerger, S.: Online communities of practice typology revisited.
Journal of Information Science 35(6), 740–757 (2009)
4. Mishne, G., Glance, N.: Leave a reply: An analysis of weblog comments, In: Third Annual
Workshop on the Weblogging Ecosystem (part of WWW 2006) (2006)
Towards Delivering Disease Support Processes
for Patient Empowerment
Using Mobile Virtual Communities
Bert-Jan van Beijnum, Pravin Pawar, Lamia Elloumi, and Hermie Hermens
Faculty of Electrical Engineering, Mathematics and Computer Science

University of Twente, Enschede, The Netherlands
{b.j.f.vanbeijnum,p.pawar,l.elloumi,h.j.hermens}@utwente.nl
Abstract. In existing healthcare systems, the focus is on clinical processes to

assess the health condition of patients, making clinical decisions and applying
therapeutic procedures all under control of health professionals. However, patients
with chronic diseases usually face many disease related problems that are not
immediately under control or supervision of a health professional. Taking patients
with chronic cardiovascular diseases as an example, these patients are
recommended to make a number of lifestyle changes: increase physical activity,
change diet habits, quit smoking and adhere to a medication intake regime. In this
paper we propose the use of Mobile Virtual Communities to enhance the
empowerment of patients by providing the ICT mediated social support
functionalities that assist patients to realize the lifestyle changes needed.
Keywords: mobile virtual communities, telemedicine, cardiovascular diseases,

disease support processes, patient empowerment.
1 Introduction
In the past decade, the European Union has, and still is, investing a lot in changing the
way in which health services are delivered. Telemedicine is seen as a solution for the
problems in current and future healthcare delivery. Originally started with remotely
monitoring of the health condition of patients with chronic diseases, the emphasis
today is becoming more on the self-management of patients. A project that embraces
this vision is the BraveHealth project [1]. This project targets patients with chronic
cardiovascular diseases, these diseases have the highest contribution to the European
mortality rate (about 2 million per year), and account for about 192 billion Euro in
health expenditure and about 270 million lost working days.
This paper leverages on our previous work [2] in which the research challenges
and opportunities for the mobile virtual communities (MVC) in telemedicine are
described. In this paper we outline the vision for CVD patient empowerment using
MVCs. E-Support Groups or virtual communities for patients exists already for some
time. These virtual communities mostly focus on emotional support and informational
support. In this paper, we focus on the MVCs for CVD patient empowerment
including instrumental and feedback support as well.
Towards Delivering Disease Support Processes for Patient Empowerment Using MVC 159
This paper is organized as follows. Section 2 describes related work. Section 3

presents requirements for the MVC. Section 4 presents proposed concept of patient
empowerment using MVC. Section 5 elaborates MVC platform architecture in order
to fulfill targeted functionalities. Section 6 includes a brief discussion.
2 Related Work
Today, many dedicated virtual health communities exist, we have found over 40
different sites (English language only). Patients Like Me [3] is a social network for
patients, the focus of this network is on giving and receiving support and information
from the peers with similar health profiles. Face To Face Health [4] is a social
network to find and connect patients with similar health experiences on a one-to-one
basis. Main objective is to support storytelling and share experiences. IBM’s Patient
Empowerment System [5] is based on, and extends IBM’s long tradition in the
medical sector. The IBM vision is to merge the medical network with a social
network in order to empower patients. DailyStrength [6] is a social network website
where patients provide one another with emotional support by discussing their
struggles and successes with each other. The site contains multiple online
communities that deal with different medical conditions or life challenges. Medical
professionals are also available to contact and treatments for a variety of illnesses and
problems on some of the sites listed above. The virtual community sites we
investigated focus on providing emotional and informational (peer) support.
3 MVC Platform Requirements
A robust architecture and technical platform is necessary to support patient mobility

and empowerment needs of CVD patients. We have identified a collection of
functional and non-functional requirements of such a platform based on a preliminary
analysis of CVD patient needs, application scenario studies [7] and related work [8, 9,
10]. Here we only highlight some of the main functional requirements
• Community member roles: The platform must support user roles relevant for
providing support, e.g. patient, cardiologist, nurse, practitioner, relative and friend.
Also, creation of new roles and associated access rights must be supported by the
platform.
• Access: The MVC platform must be accessible using devices that have become
standard today, hence including smart phones and tablets. Human-machine
interfaces must adapt according to device capabilities.
• Language: the MVC platform must support multiple languages.
• Ease-of-use / usability: The user-interface must be intuitive and easy to use. User
role, skills, age, gender and any disability may affect the user-interface requirements.
• Information access control: Users must be able to audit who may access their
health related information and they must be able to exercise control over who is
allowed to see their health related information.
160 B.-J. van Beijnum et al.
4 Patient Empowerment Using the Mobile Virtual

Communities
The envisioned MVCs aim at accelerating disease support processes to empower

CVD patients in achieving their health goals, for instance to adhere to the
recommendations regarding physical activity, diet and medication [11]. Relevant
types of support are known from social theory [8,9] and include informational,
emotional, instrumental and appraisal / feedback support [8,9]. We see the added
value of our approach especially in the domain of instrumental and feedback support
from e.g. peers and aids in keeping patients motivated and ways to track their own
and group performance with respect to their health goals.
The expected patient empowerment outcomes by using the prospective MVC
functionalities are: being better informed, feeling more confident in the relationship with
their physician, improved acceptance of the disease, feeling more confident about the
treatment, enhanced self-esteem, enhanced social well-being and increased optimism
and control. These outcomes are adopted from the study of patient empowerment in the
online patient groups [9]. In the patient-centric view of MVCs, each community
specifically addresses a particular type of CVD lifestyle facet. In each community,
members take on a particular role and each role performs corresponding functions the
combined result of which is focused on patient empowerment.
5 MVC Platform Architecture and Modules
Regarding the logical architecture of the MVC, we adopt the well-known and proven
three-tier architectural pattern. To provide all the intended MVC disease support
processes, we identified the need for the following modules:
• MVC Platform Management: This module is responsible to provide functionalities
for managing the platform. The platform management tasks include deployment of
community services, platform performance monitoring and platform auditing
functionalities.
• MVC Template Management: The platform needs to be able to accommodate
multiple communities, based on disease facets and required support type. Through
this module new templates can be defined which are to be instantiated in order to
create a community. The community template specifies roles and services that are
to be part of communities and the rules that govern the operation and service use.
• MVC Generics: The MVC generics refers to the services that can be reused in
different contexts. E.g. chat capabilities, the management of publication (for the
purpose of information support or for educational purposes) are generic services
that may appear in various communities.
• Specific MVCs: Based on the above modules, dedicated communities are created
to address specific disease facets. These communities may be enhanced with new
functionalities specific for the disease facet addressed.
Towards Delivering Disease Support Processes for Patient Empowerment Using MVC 161
6 Discussion
In this paper we have described ongoing work on delivering patient empowerment

processes for CVD patients with the aim to maintain or improve patients’ health status
by focusing on CVD lifestyle disease facets. The disease facets include: diet, physical
exercise/activity, medication compliance, and smoking cessation. The concept of
patient empowerment, requirements, architecture and modules of proposed MVCs are
elaborated. Presently, we are working on the implementation of MVC modules and
performance metrics by which patients can objectively track their own performance
and that of peer patients in their community. In addition, metrics for group
performance are under development. In future phases of the project, the effectiveness
in terms of patient empowerment using instrumental and feedback support will be
evaluated in a clinical trial.
Acknowledgments. This study is supported by EUROPEAN UNION FP VII project

number 248694 (BRAVEHEALTH).
References
1. Bravehealth: Patient Centric Approach for an Integrated Adaptive, Context Aware Remote
Diagnosis and Management of Cardiovascular Diseases, http://bravehealth.eu
2. Beijnum, B.J.F., van Pawar, P., Dulawan, C., Hermens, H.: Mobile Virtual Communities
for Telemedicine: Research Challenges and Opportunities. International Journal of
Computer Science & Applications 6(2), 38–49 (2009)
3. Patientslikeme, http://www.patientslikeme.com/
4. FacetoFaceHealthCommunity, http://www.facetofacehealth.com/
5. Made in IBM Labs: IBM Reinvents the Patient Portal (March 2011),
http://www-03.ibm.com/press/us/en/pressrelease/33944.wss
6. DailyStrength, http://www.dailystrength.org/
7. Fedele, F., Sterfanis, P.D., Ribeiro, V.: Application Scenario Studies, BraveHealth project
Deliverable 1.1 (September 2010),
http://bravehealth.eu/pdf/Deliverable%201.1%20Application%20
Scenario%20Studies_FINAL.pdf
8. Tardy C.H.: Social Support Measurement. American Journal of Community Psychology
13(2) (1985)
9. van Uden-Kraan, C.F., et al.: Self-Reported Differences in Empowerment Between
Lurkers and Posters in Online Patient Support Groups. Journal of Medical Internet
Research 10(2) (2008)
10. Maloney-Krichmar, D., Preece, J.: A Multilevel Analysis of Sociability, Usability, and
Community Dynamics in an Online Health Community. ACM Transactions on Computer-
Human Interaction 12(2), 201–232 (2005)
11. Dickstein, K.: ESC Guidelines for the diagnosis and treatment of acute and chronic heart
failure 2008. European Heart Journal 29, 2388–2442 (2008)
E-Health Readiness Assessment for E-Health Framework
for Africa: A Case Study of Hospitals in South Africa
Alfred Coleman1, Marlien E. Herselman2, and Dalenca Potass3

1
Tshwane University of Technology, Soshanguve Campus, South Africa
2
Living Labs, Meraka Institute, CSIR, South Africa
3
School of Information and Communication Technology,
Nelson Mandela Metropolitan University, South Africa
colemana@tut.ac.za, mherselman@csir.co.za,
Dalenca.Pottas@nmmu.ac.za
Abstract. This study assessed e-healthcare readiness of rural and urban hospitals
in North West Province of South Africa. Outcome of assessment led to creation
of e-health architectural framework for e-health solutions. Assessment was
conducted in usage of ICT in patient healthcare record system, processes and
procedures in consultation among healthcare professionals, prescription of
medication, referral of patients and training of healthcare professionals in ICT
usage. The study was in two phases and six hospitals were selected. E-healthcare
readiness assessment focusing on need, technological, engagement and social
acceptance readiness were assessed. Data collected used group interviews and
qualitative questionnaires. Findings showed that computers were not used for
clinical duties and no e-health solutions were found. E-health Maturity Level
was at level zero. Recommendations and compilation of Provincial E-Health
Framework (PEHF) were made. The findings were unexpected and therefore, of
great benefit to healthcare institutions which intend to implement e-health
initiatives in hospitals.
Keywords: Technology readiness assessment, e-health, patient electronic

health records, electronic prescription.
1 Introduction
HIV pandemic in most African countries has not only caused considerable strain on
various national healthcare systems, but has increased the number of orphans, reduced
productive human capital and productivity, eroded knowledge and skills, put pressure
on national budgets, increased the poverty-stricken populace and reduced the quality
of life, health and wellness[1]. E-health as one of the supportive systems within the
healthcare system has great potential to address the challenges facing healthcare
systems in developing countries. The recognition of Information and Communication
Technology (ICT) in healthcare is not an end unto itself but a means to an end. The
successful introduction of ICT in healthcare requires the examination of complex
political, organisational and infrastructural factors, including a readiness factor [2].
E-Health Readiness Assessment for E-Health Framework for Africa 163
The White Paper on transformation of the public healthcare system in South Africa
(SA) reveals that majority of the population has inadequate access to basic healthcare
services and that the greater percentage of this population lives in rural communities
[3].The White Paper also indicates that majority of South Africans receive their
medical care from government-run clinics and hospitals [4]. The North West Province
Health Department (NWPHD) has invested substantial sums of money in the district
and provincial health services, health facilities management and healthcare support
services, including ICT, in an effort to improve the work processes of healthcare
professionals in order to promote efficient delivery of healthcare services. Despite the
high investment in the healthcare system, many healthcare professionals in NWP do
not receive the benefits for which ICT can provide to improve on their work
processes.
This article reports on how ICT can be used to promote efficiency in the work
processes within the hospitals to deliver quality healthcare services to the people in
North West Province (NWP). To operationalise this objective, an e-health readiness
assessment of rural and urban hospitals was conducted. Outcome of the assessment
led to compilation of e-health framework to improve healthcare professionals’ work
processes. The assessment investigated effective and efficient use of ICT in patient
record systems, processes in consultation, prescription, referrals and training.
This research has contributed to the body of knowledge in e-health in two ways.
Firstly, a set of recommendations has been drawn from the research data to help
bridge the gap between current levels of ICT usage in healthcare to attain a higher
level where there is integration of e-health adoption in the hospitals. Secondly, a
PEHF for the hospitals in NWP has been compiled, based on the data drawn from the
research.
1.1 E-Health and E-Healthcare Readiness Assessment Theories
The World Health Organization [5] defines e-health as ‘being the leveraging of ICT to
connect providers, patients and governments; to educate and inform healthcare
professionals, managers and consumers; to stimulate innovation in healthcare delivery
and health system management and to improve our healthcare system’. Eysenbach [6]
refers to e-health as ‘a concerted effort undertaken by leaders in healthcare and hi-
tech industries to fully harness the benefits available through convergence of the
Internet and healthcare’. The advent of e-health seems fitting to address both
opportunities and challenges in the healthcare sector but the question is, are hospitals
in NWP ready to embrace the full potential of e-health?
Currently there are six assessment models which are commonly used to assess e-
health readiness in the health environment. These models [3], [8], [7], provide
different dimensions which can be utilized in assessing e-healthcare readiness. These
are core readiness, engagement readiness, structural readiness and non-readiness.
Another component, Technology Acceptance Model [9], is included for use in
developing countries, especially in rural areas. The principal components adopted for
this e-health readiness assessment included technology acceptance construct which is
vital to the introduction of any technology in rural hospitals.
164 A. Coleman, M.E. Herselman, and D. Potass
2 Methods
Qualitative research design employing a multiple case study approach was used. Data
from selected urban and rural hospitals were collected, using group interview and
qualitative questionnaire. The questionnaire instrument was used in addition to the
group interview because it promoted reliability and validity of the data. The data were
analyzed by utilizing [10] case study analysis template and [11] guidelines for case
study analysis. The group interview and questionnaire items were formulated in
accordance with the categories of background/history of hospitals, hospital
infrastructure, ICT access level, including ICT availability, accessibility and usability;
and e-health solutions including availability, accessibility and usability.
There are five regions in the province and all the regions were taken into account
by selecting a hospital from each region. The selected hospitals were Rustenburg,
Taung, Ganyesa, Klerksdorp, Christian and Reivilo Hospital. Rustenburg and
Klerksdorp Hospitals are urban hospitals while Taung, Christiana and Reivilo
Hospitals are rural hospitals.
2.1 Participants
A purposive sampling was applied to select respondents from the hospitals in order to
achieve the goals of the study. A total of 48 respondents were selected. The selected
respondents from each hospital had the following categories: 2 administrators, 2
general doctors, 2 professional nurses and 2 assistant nurses. The respondents were
informed prior to the group interviews and the completion of the questionnaires that
participation in the research was voluntary and that any information provided would
be treated as confidential.
2.2 Research Instruments
Group Interview
A group interview was conducted in each hospital. The interviewees were of diverse
age, ethnicity, gender and educational levels. The purpose of the interview was to
determine how the healthcare professionals perceive the usefulness and potential
benefits of e-patient record, e-prescription, e-consultation, e-referrals and e-training
systems. The interview questions are attached to the full research report.
Questionnaire
The questionnaire instrument consisted of two sets of questions. The first set of
questions was administered to hospital administrators whilst the second was for
general doctors, nurses and assistant nurses. The purpose of the first set of questions
was to establish the background history, the settings of the hospitals and the existing
ICT infrastructure in these hospitals. The second set of questions was to establish
baseline data for processes and procedures in keeping patient health records,
consultation among healthcare professionals, prescription and referral processes. The
questions were drawn from e-health readiness assessment framework [9].
3 Results
This section presents the findings and proposes some recommendations which served
as guiding principles for the development of e-health solution.
3.1 Questionnaire Results
Availability of Healthcare Professionals in Rural and Urban Hospitals

The findings indicate that there is a widespread shortage of doctors in rural hospitals
in NWP. An average ratio of 1doctor to 18000 patients for rural hospitals as compared
to1 doctor to 9000 patients in urban hospitals was found. These hospitals are all state-
owned healthcare institutions.
ICT Availability and Accessibility in Rural and Urban Hospitals

The findings reveal that urban hospitals have more ICT equipment than rural
hospitals. Internet connection is more reliable in terms of connectivity and speed in
urban hospitals. In rural hospitals, connectivity and speed of Internet services are
often affected by poor telephone lines and interruption of electricity power supply.
The average ratio of number of computers to doctors in rural hospitals is 1:3 while it
is 1:2 in urban hospitals. However, data obtained from the e-health readiness
assessment revealed that there were no computers in the doctors’ consultation rooms.
It was ascertained that doctors do not use computers for their clinical duties. Again
nurses and assistant nurses do not use computers for their clinical work. These
computers are used by hospital administrative staff for capturing patients’
demographic information and revenue collection. This means that even though there
are computers in the hospitals, they are used for activities which are not directly
related to clinical work like e-patient health record keeping.
Healthcare professionals’ use of computer was also assessed. Healthcare
professionals in urban hospitals use computers/Internet daily for searching for
information but not for carrying out their clinical duties. Healthcare professionals in
rural hospitals use computers weekly or monthly.
3.2 Group Interview Results
E-health Solution, Availability and Accessibility in Rural and Urban Hospitals,

and Perceived Usefulness in Rural and Urban Hospitals
In comparing urban and rural hospitals in terms of e-health solution availability and
accessibility, the following findings emerged: There is no e-patient health record
system, e-consultation, e-prescription, e-referral and e-training systems in both urban
and rural hospitals in NWP. Participants indicated that installation of e-patient health
record system will save time when recording patient information, diagnosing and in the
treatment of patients. One participant stated “Patients who do not get their records after
a long time, the e-patient record system will help to safeguard this”. It became evident
that e-patient record system will improve communication and eliminate patients who
collect medication from multiple hospitals with duplicate paper prescription. One
respondent stated, “It will eliminate ghost patients and before a patient arrives at the
pharmacy, the pharmacist will have the information about the patient”.
It emerged that referrals of patients from a lower level hospital to a higher level
hospital were done through the use of paper notes. Patients are given referral letters
(except in emergency cases) which they take to the referred hospital. Patients often
misplace these letters and end up not going to the referred hospital. One participant
indicated, “E -referral will help because you refer patients but on the way they get lost
or don’t go. If it is done electronically, the referred hospital will know that such a
time we expect this patient”.
There is Internet facility in both urban and rural hospitals but the Internet is limited
to searching information and sending e-mails. Both urban and rural hospitals have
PAAB system which is used to collect and send patient demographic information to
NWPHD’s head office monthly. Urban hospitals have tele-radiography facility which
is used for sending x-ray images between Klerksdorp and Rustenburg Hospitals.
4 Discussion
4.1 Availability of Healthcare Professionals in Rural and Urban Hospitals

An average ratio of 1doctor to 18000 patients for rural hospitals as compared to1
doctor to 9000 patients in urban hospitals was noted. Since these are state-owned
healthcare institutions, the ratio of doctors to number of patients, both in rural and
urban hospitals, is higher than that quoted by the [12] report. The report [12] focused
on both private and state-owned hospitals in SA, while this research investigated only
state-owned hospitals in NWP in SA. The shortage of doctors in rural hospitals has to
drive investment into e-health technologies like e-consultation, to enable non-doctors
healthcare professionals such as nurses and midwifes to perform more advanced
functions through synchronous consultation with professional doctors.
4.2 ICT Availability and Accessibility in Rural and Urban Hospitals

It was found that more ICT equipment is needed in both rural and urban hospitals if e-
health solutions are to be implemented. Internet connection also needs upgrading.
ICT infrastructure in both urban and rural hospitals is not integrated to be able to
work together within and across urban and rural hospitals. This confirms research
conducted in the Eastern Cape Province of SA [13].
4.3 E-Health Solution, Availability and Accessibility in Rural and Urban

Hospitals, and Perceived Usefulness in Rural and Urban Hospitals
The findings showed that the Internet facility in both rural and urban hospitals is
limited to searching information and sending e-mails. Urban hospitals have tele-
radiography facility which enables them to send x-ray images. These findings place
both the urban and rural hospitals at Level 0 (The Baseline) on the E-health Maturity
Curve as illustrated in Figure 1 below.
Source: [14]
Fig. 1. E-health Maturity Curve
Figure 1 illustrates that the level of e-health application in rural hospitals is at the
Presence stage which is classified as Level 0 (The baseline). At this stage there is a
non-interactive website where the main intent is to disseminate information. Thus,
rural hospitals are able to receive and send information from and to NWPHD through
e-mails and the PAAB system.
The level of e-health application in urban hospitals is at the Interaction stage which
is also at Level 0 (The baseline). However, the Interaction stage offers services that
are more advanced than the Presence stage. Thus, urban hospitals are able to send and
receive x-ray images through the tele-radiography facility. Despite the availability of
these ICT infrastructure facilities, the ICT systems are not integrated to be able to
work together across departmental and organizational boundaries to use information
that has been provided by the patient.
5 The Need for Provincial E-Health Framework (PEHF)
A major setback which became evident is that despite the availability of the ICT
facilities and other ICT infrastructure in both urban and rural hospitals, the ICT
systems are not integrated to work together within and across the hospitals to allow
healthcare professionals to gain the benefits of e-health solutions and applications.
These findings place both rural and urban hospitals at Level 0 on the E-Health
Maturity Curve (cf Fig1). Therefore, it is imperative that a special e-health framework
be compiled based on these findings to move the e-health application usage in these
hospitals from Level 0 to Level 2 (Healthcare 2.0) on the E-Health Maturity Curve by
ensuring that Level 1(Integration) is effectively and efficiently achieved. Therefore,
the PEHF was compiled (cf Fig 2).
Fig. 2. PEHF Architecture
The PEHF architecture emphasizes the integration of services within local

hospitals and avails these services to the web through a provincial e-health services
hub. This component incorporates the business services of the hospitals and
transforms them from physical and paper phase to the electronic phase through the
service hub. This represents the migration from the physical normal business and
paper services of the hospitals to web-based services.
6 Conclusion
Having reviewed the literature on e-health and e-healthcare assessment, conducted the
e-healthcare assessment in the selected hospitals, and considering the findings derived
from the hospitals’ assessment, this article provides the following conclusion:
• ICT infrastructure in both urban and rural hospitals is not integrated to work
together within and across hospitals to allow healthcare professionals to gain
the benefits of e-health solutions and applications
• E-health applications in the work processes in the hospitals are at the
Presence and Interaction stages on E-Health Maturity Curve which is Level 0
(The Baseline).
Based on the above conclusion, the following recommendations are made:
• ICT systems within each hospital and across hospitals in NWP need to be
integrated in order to facilitate e-consultation by using an integrated network
which will assist healthcare professionals to consult with peers and
specialists for professional advice and information; and
• Inter-operability of systems to bring together the diverse systems and data

sources into a coherent, controlled environment, create a network of
healthcare delivery and incorporate existing e-health applications like PAAB
and tele-radiography into one e-health framework.
In conclusion, this article has provided the PEHF which will move the work processes
of healthcare professionals from Level 0 to Level 2 (Healthcare 2.0) on the E-health
Maturity Curve, leading to efficient and effective delivery of healthcare services to
the people in NWP.
References
1. Botswana Government: Status of the National Response to the UNGASS Declaration of
Commitment on HIV/AIDS. National Aids Coordinating Agency, Gaborone (2003)
2. Jennett, P., Jackson, A., Healy, T., et al.: A Study of a Rural Community’s Readiness for
Telehealth. J. Telecare 9, 259–263 (2003)
3. Department of Health (DoH). Documents and Reports.: North West Department of Health,
http://www.nwhealth.gov.za/nwhealth_web/doc_reports.php
4. Rasmeni, M.: Budget Speech for the Department of Health for the Financial Year
2007/2008. Presented by the MEC for Health, at the North West Provincial Legislature,
http://www.nwhealth.gov.za/nwhealth_web/dir_prog.php
5. World Health Organization (WHO).: Developing Health Management Information
Systems: A Practical Guide for Developing Countries, Geneva (2004)
6. Eysenbach, G.: What is e-Health? J. Med. Internet Research 3(2), 20 (2001)
7. Campbell, J.D., Harris, K.D., Hodge, R.: Introducing Telemedicine Technology to Rural
Physicians and Settings. J. Fam. Pract. 50, 419–424 (2001)
8. Demiris, G., Oliver, D.R.P., Porock, D., Courtney, K.: Home Telehealth: The Missouri
Telehospice Project: Background and Next Steps. Home Health Care Technology
Report 1(49), 55–57 (2004)
9. Ojo, S.O., Olugbara, O.O., Emuoyibofarhe, O.J.: Formal Model for e-Healthcare
Readiness Assessment in Developing Country Context. In: 4th International Conference on
Innovations in Information Technology, IIT 2007 (2007)
10. Creswell, J.W.: Qualitative Inquiry and Research Design, 2nd edn. Choosing among Five
Approaches. Sage Publications, London (2007)
11. Klein, H.K., Myers, M.D.: A Set of Principles for Conducting and Evaluating Interpretive
Field Studies in Information Systems. Management Information Systems Quarterly 23(1),
67–94 (1999)
12. United Nations Framework Convention on Climate Change (UNFCCC).: Development
and Transfer of Technologies. In: Workshop on Enabling Environment for Technology
Transfer, http://www.unfccc.int/resource/docs
13. Nkqubela, L.R., Herselman, M.E.: ICT Application as E-health Solutions in rural
Healthcare in the Eastern Cape of South Africa. Health Information Management J. 39(1)
(2010)
14. Jarvis, B., Fortunov, I., Bachour, T.: Connect Health Framework Architecture and Design
Blue Print, http://www.microsoft.com/.../articles
The Physical Activity Loyalty Card Scheme:
Development and Application of a Novel System
for Incentivizing Behaviour Change
Ruth F. Hunter1 , Michael Davis2 , Mark A. Tully1,3 , and Frank Kee1,3

1
Centre for Public Health, Queen’s University Belfast, Northern Ireland
2
Centre for Secure Information Technologies (CSIT), Queen’s University Belfast
3
UKCRC Centre of Excellence for Public Health, Queen’s University Belfast
ruth.hunter@qub.ac.uk
Abstract. In the Public Health White Paper “Healthy Lives, Healthy

People” (2010), the UK Government emphasised using incentives and
“nudging” to encourage positive, healthy behaviour changes. However,
there is little evidence that nudging is effective, in particular for increas-
ing physical activity. We have created a platform to research the effec-
tiveness of health-related behaviour change interventions and incentive
schemes. The system consists of an outward-facing website, incorporat-
ing tools for incentivizing behaviour change, and a novel physical activ-
ity monitoring system. The monitoring system consists of the “Physical
Activity Loyalty Card”, which contains a passive RFID tag, and a con-
tactless sensor network to detect the cards. This paper describes the
application of this novel web-based system to investigate the effective-
ness of non-cash incentives to “nudge” adults to undertake more physical
activity.
Keywords: incentives, nudging, physical activity, behaviour change,

Web 2.0, RFID tags, sensor technologies
1 Introduction
The prevalence of physical inactivity, rising obesity levels and associated health
conditions in children and adults is rising [1,2,6]. A continuing trend of a more
sedentary lifestyle has led to stark projections of future prevalence of obesity,
morbidity and mortality [1,2,12]. Recent European figures show that only 31%
of adults currently meet the physical activity recommendations [16]. This level
of inactivity has directly contributed to the rising rate of obesity worldwide. In
the UK, most adults are already overweight and by 2050, 60% of men and 50%
of women could be clinically obese, costing an extra £45.5 billion per year in
treating obesity-related disease [5].
More innovative interventions are required to halt the global increase in phys-
ical inactivity and obesity by sustaining healthy lifestyle behaviours for all
ages. “Nudging” and using incentives to promote positive, long-term healthy

The Physical Activity Loyalty Card Scheme 171
behaviour changes is now the UK government’s preferred strategy for promoting

public health [15]. Nudging is defined as:
. . . any aspect of the choice architecture that alters people’s behaviour in a

predictable way without forbidding any options or significantly changing
their economic incentives. To count as a mere nudge, the intervention
must be easy and cheap to avoid. Nudges are not mandates. Putting the
fruit at eye level counts as a nudge. Banning junk food does not. [19]
Nudging is not new and draws on behavioural economics and social psychol-
ogy to explain why people may behave in ways that deviate from rationality,
invoking various mechanisms of behaviour change including the changing of de-
fault options, message framing and the provision of social norm feedback. Given
the vicious cycles that confound our attempts to change our behaviour [18], it
is not surprising that a recent framework has underlined the value of combin-
ing approaches that can simultaneously affect our capability, opportunity and
motivation to change behaviour [14].
1.1 Previous Research
While there is growing evidence of effectiveness of these approaches for some

behaviours, the evidence base for nudging to promote physical activity is sparse.
Certainly, more behavioural economists are beginning to research the role of
incentives for physical activity, and both financial [3] and non-financial [9] in-
centives have been shown to increase levels of physical activity, at least in the
short term.
The Public Health White Paper [15] highlighted a number of nudge-type ini-
tiatives, such as the Step2Get scheme, developed by Intelligent Health Ltd.1
Based on individual incentives and innovative technology to monitor physical
activity levels, students were provided with a swipe card which they touched
on receivers placed on lampposts along a safe walking route to school. Each
completed walk to school was converted to points and these were redeemed as
rewards (e.g., cinema vouchers) at the end of the week. Post-intervention results
showed an 18% modal shift towards walking to school. Unfortunately, there are
no follow-up results available to determine if the large extrinsic incentive (cinema
vouchers) had led to a long-term intrinsic behaviour change (increased physical
activity).
This paper describes the development and application of a novel web-based
system for incentivizing behaviour change using innovative technology. The pa-
per will also outline the study design and preliminary findings of a randomized
controlled trial investigating the effectiveness of incentives (redeemable vouchers
for retail outlets) as a nudge to encourage adults to be more physically active.
1
http://www.intelligenthealth.co.uk/
172 R.F. Hunter et al.
(a) Swiping the PAL Card at a (b) The PAL Scheme website
sensor in the park
Fig. 1. Elements of the PAL Scheme
2 The Web-Based System

We have developed a web-based system in conjunction with The Physical Activity
Loyalty Card Scheme, based on guidance from previous research [7,8,11,17,20].
The primary purpose of this system is to increase physical activity and promote
a healthy lifestyle. The system incorporates a number of behaviour change tools
and nudge elements, which are described in more detail in this section.
2.1 Incentives and Nudge Elements

The PAL Scheme system integrates an innovative physical activity tracking sys-
tem (Intelligent Health Ltd.) with web-based monitoring and nudge elements
(Fig. 1). The tracking system uses Near-Field Communication Technology (NFC)
and a “Physical Activity Loyalty Card” (PAL Card) which contains a passive
Radio Frequency Identification (RFID) tag. This is a contactless proximity tech-
nology: when users present their card within 10 cm of the NFC sensor, a trans-
action is logged. As the sensors are placed in the outdoor environment, they are
housed within a wooden post. The post also contains a mobile phone, which
sends the transaction (Card ID, Sensor ID, and Timestamp) via SMS to a data
centre. In the back-end system, the transactions are assembled into paths (see
Fig. 3a for a visualisation of the network paths). As we know the elapsed time
between sensor swipes and the distance between sensors, we can calculate dis-
tance and average walking speed for each session. We also know the height and
weight of each user: these can be used to estimate the number of calories burned.
The system has been set up to give users real-time feedback on various aspects
of their recent bout of physical activity, including minutes completed, distance
covered and calories expended (Fig. 1b). Each user has a personal account where
they can view their own physical activity data. Feedback is presented graphically
to show users’ daily and aggregated physical activity parameters. Personal goals
can be set using this feature and shown on the graphical display. This system can
be used to incorporate some nudge elements—for example, minutes of activity
are converted to points and redeemed for various rewards.
We incorporated a feature that enabled automated messages to be tailored
to the individual users’ weekly physical activity level. For example, if a user
had fallen below the recommended weekly physical activity level, a prompt was
sent to encourage them to undertake more activity with links to some of the
behaviour change tools featured on the website to help them.
2.2 Behaviour Change Tools
The website contains a number of interactive tools offering personal advice,

support and aids for behaviour change. Two tools have been adapted to offer
individually tailored advice to the participant in order to increase their physical
activity levels. These include the Stages of Change Questionnaire [13] and an
adapted questionnaire for identifying and overcoming barriers to becoming more
physically active. In addition, the website contains tools for planning physical
activity opportunities as part of their working week.
2.3 Other Features
The website contains health promotion material for leading a healthy lifestyle,
including advice on physical activity, diet and smoking. We used Facebook and
Twitter to disseminate health promotion messages and regular encouragements
to keep participants engaged with the scheme. Additionally, the website incor-
porated a number of Web 2.0 features (e.g. forums, user comments) to receive
feedback on the system. Participants were also able to use the social network-
ing features to provide social support—for example, planning walks with other
users.
Finally, the web-based system also acts as a comprehensive research tool,
incorporating features to support each stage from recruitment to data analysis.
This includes electronic data collection, processing and aggregation.
3 Application of the System
3.1 Trial Design
We have built upon the non-randomized scheme highlighted in Sect. 1.1 and
designed an intervention with several nudge components (including modest in-
centives for physical activity participation). We also draw upon evidence-based
approaches from the behavioural science literature, including self-monitoring,
the provision of personal feedback and goal-setting resources. The aim of the in-
tervention is to provide an extrinsic incentive to nudge individuals to develop a
long-term intrinsic behaviour change. The system has been piloted in a 12-week
intervention with civil servants who work at Stormont Estate, Belfast, Northern
Ireland. Based on a sample size calculation, we recruited 406 employees aged
18–65 years old, who work on the Estate at least four days a week (minimum
of 6 hours per day) to participate in a randomized controlled trial investigat-
ing the effectiveness of incentives for encouraging physical activity in adults.
Participants were recruited via email, posters and flyers distributed around the
workplace. Participants were randomly allocated to one of two groups:
1. Participants received incentives for being physically active. Participants col-

lected points for each minute of physical activity that they completed by
swiping their “loyalty card” across sensors placed around the Estate. Points
could be earned by walking/running around the Estate, and by attending the
gym/exercise classes. Points were reimbursed for rewards at the end of week
6 and week 12. Participants could earn a maximum of 30 points per day, five
days per week (equivalent to the physical activity recommendations). The
rewards were various redeemable vouchers from a number of local businesses.
Participants received feedback regarding minutes of activity, points earned,
distance covered and calories burned.
2. Participants did not receive incentives for being physically active. Partici-
pants used their “loyalty card” to monitor their physical activity levels and
received feedback regarding time, distance, and calories. They were not able
to collect points or earn rewards.
3.2 Preliminary Results

There were 406 participants in the trial: 67% were female and the mean age
was 43.2 years. Fig. 2a shows the number of unique PAL Card users during
the 12-week intervention. There were 250 users in the first week. This gradually
declined from week 4, with approximately 100 users during the final week.
Fig. 2b shows the number of times each participant used their PAL Card
per week. Most users undertook activity (used their PAL Card) between 1 and 5
times per week. This data will be further analysed to investigate the effectiveness
of incentives for increasing physical activity and the predictors of card usage.
Fig. 3 represents the PAL Card transactions as network graphs. In Fig. 3a,
the vertices represent RFID sensors and edges represent patterns of mobility.
Sensors R1–R5 were placed along paths in the outdoor environment; there were
two additional sensors on the door to the Gym and the Fitness Studio. Edge
weights are proportional to the frequency that each path segment is traversed.
This allows us to analyse the mobility patterns of the participants in the study.
Fig. 3b shows the social connections between participants in the study. We
used a method similar to [10] to infer social connections based on users who
presented their cards at the same sensor within a few seconds of each other. If
two users present their cards in the same place at the same time, this suggests
a possible social connection. If this happens frequently for a given pair of users,
a connection is almost certainly present. The graph shows that certain users
(a) Unique active PAL Card (b) Number of PAL Card activity sessions per par-
users per week ticipant per week
Fig. 2. PAL Card activity during the 12-week intervention
(a) Sensor network (b) Social network
Fig. 3. PAL Card transactions represented as network graphs
formed clear clusters, whereas others preferred to exercise on their own. The
size of each node indicates the degree of connectedness; the shade represents the
level of physical activity (darker shades indicate less activity and lighter shades
indicate more activity). We propose to use this kind of graph to investigate
whether physical activity behaviour changes percolate through social networks.
Beyond visual analytics, we intend to investigate the formal properties of the
graph representations using graph-mining techniques such as [4].
4 Future Applications
There is scope for this web-based system to be used to investigate the effective-
ness of nudge interventions including the influence of social norms, competition,
and other lessons learnt from the behavioural economics literature. This innova-
tive technology can be applied to future schemes and research trials investigating
the use of incentives and nudges to encourage positive lifestyle behaviour changes
in terms of, for example, smoking and diet, and in various settings—for example,
school and workplace.
5 Conclusions
In this paper, we have described a sophisticated system which has been used
to investigate the effectiveness of incentives for encouraging adults to be more
physically active. Preliminary results indicate that this is a useful tool for this
purpose. The data will be further analysed to investigate the effectiveness of
incentives, the predictors of users of the system and the influence of peer re-
lationships in increasing physical activity in adults. We plan to develop a set
of software tools for the development and deployment of future nudge and be-
haviour change interventions that can be used by other researchers and public
health practitioners.
Acknowledgements. We would like to acknowledge funding from the National

Prevention Research Initiative (NPRI) and their funding partners (Alzheimer’s
Research Trust; Alzheimer’s Society; Biotechnology and Biological Sciences Re-
search Council; British Heart Foundation; Cancer Research UK; Chief Scientist
Office, Scottish Government Health Directorate; Department of Health; Diabetes
UK; Economic and Social Research Council; Engineering and Physical Sciences
Research Council; Health and Social Care Research and Development Division
of the Public Health Agency (HSC R&D Division); Medical Research Council;
The Stroke Association; Welsh Assembly Government; and World Cancer Re-
search Fund (http://www.npri.org.uk). We also wish to acknowledge funding
from the Department for Employment and Learning, Northern Ireland and the
support of Intelligent Health Ltd., London.
References
1. White house task force report on childhood obesity: Report to the president. solving
the problem of childhood obesity within a generation (May 2010),
http://www.letsmove.gov/
white-house-task-force-childhood-obesity-report-president
2. Branca, F., Nikogosian, H., Lobstein, T.: The challenge of obesity in the WHO
European region and the strategies for response. Technical report, WHO Regional
Office for Europe (2007)
3. Charness, G., Gneezy, U.: Incentives to exercise. Econometrica 77(3), 909–931
(2009)
4. Davis, M., Liu, W., Miller, P., Redpath, G.: Detecting anomalies in graphs with
numeric labels. In: 20th ACM Conference on Information and Knowledge Manage-
ment (CIKM 2011), Glasgow (October 2011)
5. DHSSPS. A fitter future for all: An obesity prevention framework for Northern
Ireland 2011–2021 (draft),
http://www.dhsspsni.gov.uk/obesity-prevention-framework-2011-2021.pdf
(accessed September 13, 2011)
6. Farhat, T., Iannotti, R.J., Simons-Morton, B.G.: Overweight, obesity, youth, and
health-risk behaviors. American Journal of Preventive Medicine 38(3), 258–267
(2010)
7. Ferney, S.L., Marshall, A.L.: Website physical activity interventions: preferences of
potential users. Health Education Research 21(4), 560–566 (2006)
8. Ferney, S.L., Marshall, A.L., Eakin, E.G., Owen, N.: Randomized trial of a neigh-
borhood environment-focused physical activity website intervention. Preventive
Medicine 48(2), 144–150 (2009)
9. Johannesson, M., Östling, R., Ranehill, E.: The effect of competition on physical ac-
tivity: A randomized trial. The B.E. Journal of Economic Analysis & Policy 10(1),
91 (2010)
10. Lauw, H.W., Lim, E.-P., Pang, H., Tan, T.-T.: Social network discovery by mining
spatio-temporal events. Comput. Math. Organ. Theory 11, 97–118 (2005)
11. Leslie, E., Marshall, A.L., Owen, N., Bauman, A.: Engagement and retention of
participants in a physical activity website. Preventive Medicine 40(1), 54–59 (2005)
12. Lieberman, D.A., Chamberlin, B., Medina, E., Franklin, B.A., Sanner, B.M., Vafi-
adis, D.K.: The power of play. In: Innovations in Getting Active Summit 2011,
vol. 123, pp. 2507–2516 (2011)
13. Marcus, B.H., Selby, V.C., Niaura, R.S., Rossi, J.S.: Self-efficacy and the stages of
exercise behavior change. Research Quarterly for Exercise and Sport 63(1), 60–66
(1992)
14. Michie, S., van Stralen, M., West, R.: The behaviour change wheel: a new method
for characterising and designing behaviour change interventions. Implementation
Science 6(42) (April 2011)
15. Department of Health. Healthy lives, healthy people: our strategy for public health
in England (2010), http://dera.ioe.ac.uk/1974/1/dh_122347.pdf
16. World Health Organization. 10 key facts on physical activity in the WHO European
region, http://tinyurl.com/10factsPA-EU (accessed September 13, 2011)
17. Pressler, A., Knebel, U., Esch, S., Kölbl, D., Esefeld, K., Scherr, J., Haller, B.,
Schmidt-Trucksäss, A., Krcmar, H., Halle, M., Leimeister, J.M.M.: An internet-
delivered exercise intervention for workplace health promotion in overweight seden-
tary employees: a randomized trial. Preventive Medicine 51(3-4), 234–239 (2010)
18. Swinburn, B., Egger, G.: The runaway weight gain train: too many accelerators,
not enough brakes. BMJ 329(7468), 736–739 (2004)
19. Thaler, R.H., Sunstein, C.R.: Nudge: Improving decisions about health, wealth and
happiness. Penguin, London (2008)
20. Vandelanotte, C., Spathonis, K.M., Eakin, E.G., Owen, N.: Website-delivered phys-
ical activity interventions. American Journal of Preventive Medicine 33(1), 54–64
(2007)
Tracking Media Reports on the Shiga Toxin-Producing
Escherichia coli O104: H4 Outbreak in Germany
Jens P. Linge1, Jas Mantero2, Flavio Fuart1, Jenya Belyaeva1,

Martin Atkinson1, and Erik van der Goot1
1
Joint Research Centre of the European Commission,
Institute for the Protection and Security of the Citizen, Ispra, Italy
2
European Centre for Disease Control and Prevention (ECDC), Stockholm, Sweden
{jens.linge,flavio.fuart,martin.atkinson,
erik.van-der-goot}@jrc.ec.europa.eu
jas.mantero@ecdc.europa.eu
jenya.belyaeva@ext.jrc.ec.europa.eu
Abstract. In May 2011, an outbreak of enterohemorrhagic Escherichia coli

(EHEC) occurred in northern Germany. The Shiga toxin-producing strain
O104:H4 infected several thousand people, frequently leading to haemolytic
uremic syndrome (HUS) and gastroenteritis (GI). First reports about the
outbreak appeared in the German media on Saturday 21st of May 2011; the
media attention rose to high levels in the following two weeks, with up to 2000
articles categorized per day by the automatic threat detection system MedISys
(Medical Information System). In this article, we illustrate how MedISys
detected the sudden increase in reporting on E. coli on 21st of May and how
automatic analysis of the reporting provided epidemic intelligence information
to follow the event. Categorization, filtering and clustering allowed identifying
different aspects within the unfolding news event, analyzing general media and
official sites in parallel.
Keywords: epidemic intelligence; event-based surveillance; E. coli; EHEC;

MedISys.
1 Introduction
Internet surveillance systems have increasingly been used for early event detection and
alerting of emerging public health threats [1]. The Medical Information System
(MedISys, http://medisys.newsbrief.eu) is a fully automatic public health surveillance
system to monitor reporting on emerging public health threats such as human and
animal infectious diseases, chemical, biological, radiological and nuclear (CBRN)
threats, food & feed contaminations and plant diseases [2]. The system retrieves news
articles from the internet, categorizes all incoming articles according to pre-defined
multilingual categories, identifies entities such as organizations, persons and locations,
clusters articles and calculates statistics to detect emerging threats. Users can screen the
categorized articles and display world maps highlighting event locations together with
Tracking Media Reports on the Shiga Toxin-Producing Escherichia coli O104 179
statistics on the reporting of health threats, countries and combinations thereof. Articles
can be further filtered by language, news source, and country.
Articles are classified in a category, if they satisfy the category definition which
may comprise Boolean operators, proximity operators and wild cards. Cumulative
positive or negative weights can be used with an adjustable threshold. For E. coli, a
category was introduced in MedISys in 2008 with expert input from DG SANCO and
ECDC [3]. It consists of a set of patterns covering several languages. Table 1 shows
the definition (as of 27 June 2011), with the weights for each pattern indicated in the
second column.
MedISys monitors the volume of news per category and per country to determine
sudden changes from the 14-day average number of articles in any given country-
category combination (e.g. Germany - E.coli). If the number of articles in the last 24
hours for a country-category combination (normalized by weekday fluctuations) is
significantly higher than the 14-day average, users are notified using ranking graphs
on the web site or email notifications. The statistics are language-independent because
of the multi-lingual patterns in the category definitions. This allows users to detect
any change in a category even before the event is reported in their own language.
MedISys also clusters all news items within a time window of four hours (or more,
depending on the number of recent articles) and presents the largest clusters as Top
Stories.
2 Media Reports on the Outbreak
On the 19th of May, the Robert-Koch-Institute was informed about a cluster of three
cases of HUS in Hamburg [4]. MedISys detected the first media report in the German
newspaper Die Welt on Saturday 21st of May at 12:14 CEST (Central European
Summer Time). Several other articles followed the developing story in the afternoon
(see Table 2), making reference to various press releases by public health authorities
in Germany, e.g. in Hamburg and Lower Saxony (in MedISys, users can distinguish
between general news media and official sources). Altogether, 23 German news items
triggered the MedISys E. coli category, of which 22 were about the outbreak (one
irrelevant article was about the water quality in German lakes in the region
Oberbergischer Kreis). Furthermore, there were two news reports in Farsi and one
report in Mandarin; these reports were from the science pages of Iranian and Chinese
newspapers and were not related to the outbreak. Due to the sudden increase in media
reports on E.coli on Saturday (in comparison to the average value of the last 14 days),
MedISys issued an automatic alert and also highlighted E.coli in combination with
Germany to the users (in the section called ”Most active topics”).
The Early Warning and Response System (EWRS) of the European Union [5]
received a first communication by the German authorities on Sunday 22 May at
11:40. ProMED-mail covered the event in a report on Monday 23 May [6]. All other
major early alerting systems (ARGUS, Biocaster, GPHIN, HealthMap, PULS)
reported the event as well.
180 J.P. Linge et al.
Table 1. Category definition (as of 27 June 2011) showing the patterns with their
corresponding weights (using wild cards such as % for zero, one or more characters, + for
whitespace/linebreak, and _ for one character). An incoming news article is selected, if the sum
of the weights of all triggered patterns exceeds the threshold (which was set to 40 for this
category); negative weights are used to exclude irrelevant news items.
Pattern Weight
escherichia+coli+enfek% 20
koli+basili% 20
pałeczka+okrężnicy 20
koli+basili 20
eşerişiya+koli 20
бактерията+Ешерихия+коли 20
кишечн%+палочк%+EHEC 20
палочк%+EHEC 20
EHEC 20
enterohämorragisches+Eschericha+Coli 20
Ehec-Infekt% 20
Ehec-Keim 20
Coliba -99
Colidiu -99
Ešerihija+koli 20
escherichia+coli 20
ешерихи%+кол% 20
Ešerihioze 20
大腸桿菌 20
大腸埃希氏桿菌 20
大肠杆菌 20
大肠埃希氏杆菌 20
‫اﻟﻘﻮﻟﻮﻧﻴ ﺔ‬+‫اﻹﺷ ﺮﻳﻜﻴﺔ‬ 20
‫ﻴﺎﮐﻠﻲ‬ ‫اﺷﺮﻳﺸ‬ 20
‫اﮐﻮﻟ ﻰ‬ 20
‫آﻮﻟ ﻲ‬+‫اﺳ ﭽﺮﻳﭽﻴﺎ‬ 20
‫ﯼﮐ ﻮﻻﯼا‬ 20
enterokrwotoczn% 20
escherichia+coli+O157_H7 20
escherichia+coli+O104_H4 20
escheric%+coli% 20
e.coli+bacteri% 20
Ешерихия+коли 20
кишечн%+палочк% 20
e-coli% 20
Sukelta 20
Enterohemoraginės 20
Lazdelės 20
Žarninės 20
escherich% 20
Nakkus 20
enteroh% 20
Colibacille 20
Κολοβακτηρίδιο 20
e+coli% 20
Ešerihioze 20
Kolibakteeri 20
Fig. 1. Number of articles in the E. coli category per day (data from 1 May to 4 July 2011; all
languages)
While the outbreak in northern Germany reached its peak between 21 and 23 May
[4], the media reporting showed a different behaviour. As illustrated in Fig. 1, the
highest number of articles per day was reached on 2 June 2011 (all languages, all
countries). The data can be further filtered by language, country of publication,
country mentioned in the text (using a multi-lingual category; see Fig. 2). The filter
functionality clearly shows how the media attention changed geographical focus over
time, following the developing situation.
In Fig. 2, we can clearly identify key aspects:
− the sudden rise on articles mentioning Germany (21-25 May), when the first
cases became public,
− a peak with articles on a Swedish tourist group who got infected in Lower
Saxony (26 May).
− the reporting on alleged E. coli contaminations in Spanish cucumber, tomatoes
and salad (peak on 27 May with 107 articles),
− the reporting on the financial impact on Spanish farmers and the announcement
that Spanish cucumbers had tested negative for E. coli (peak on 31 May with 300
articles),
− the discussion on trade restrictions in Russia for EU vegetable products (peak on
2 June),
− the announcements by German authorities that bean sprouts were the source of
infection (rise in volume on 5-10 June),
− the E.coli cluster of cases in Bordeaux, France (peak on 16 June with 203
articles), and
− the reporting on fenugreek seeds imported from Egypt in 2009 and 2010 (from
end of June on).
Fig. 2. Number of articles in the E. coli category from one of the countries Luxembourg (LU),
Sweden (SE), France (FR), Russia (RU), Spain (ES) and Germany (DE)
Fig. 3 illustrates how MedISys presented statistics on the outbreak on its web site
(screenshots taken on 26 May). It demonstrates how the deviation from the 14-day
average alerts the users to the E.coli-Germany combination. All data used for the
figures stem from MedISys.
In addition to the E. coli category, the outbreak was also visible in filters set up by
EFSA [7], e.g. EFSABacteria (which contains E. coli as potential pathogen),
EFSAEconomics and EFSACommerce in relation to Spain (impact on Spanish
farmers) and Russia (trade restrictions) and EFSAFoodFeedSafety. This demonstrates
that broader filters targeted at economics and commerce are able to detect changes in
media reporting.
Using entity extraction, the main organizations and people mentioned in the
articles can be identified by the system. As an example, we extracted a subset of
articles on E. coli that also mentioned the Robert-Koch-Institute, Commissioner John
Dalli, European Commission, ECDC, and EFSA. These entities were selected from an
automatically generated list of top entities (according to number of citations). Fig. 4
summarizes the data, highlighting the following aspects:
− Most articles from the early phase mention the Robert-Koch-Institute,

− The European Commission is mentioned during the discussions on the
alleged contaminations of Spanish cucumbers, tomatoes and salad;
Commissioner John Dalli is cited in the peaks on 2 June and 7 June (in
parallel to the European Commission),
− ECDC and EFSA are mentioned in many articles regarding the risk
assessment in terms of public health at EU level; a strong peak in reporting
on 30 June is due to press releases about scientific reports and risk
assessments, e.g. the scientific report [8] lead to more media attention (375
articles mentioning EFSA on 5 July).
Fig. 3. Screenshot of MedISys on 26th of May (above: country distribution; centre: statistics on
E.coli-country combinations in comparison to 14-day average values; below: daily number of
articles for the E. coli category)
Fig. 4. Number of articles in the E.coli category that also mention one of the entities Robert
Koch Institute, Commissioner John Dalli, European Commission, ECDC and EFSA
3 Conclusions
Media monitoring is a well-established technique for Epidemic Intelligence [9]. In the

case of the E.coli outbreak, the early media reports were identified quickly and
accurately; an alert was published on Saturday 21st of May via MedISys. Obviously,
the authorities in Germany were informed earlier via the established indicator-based
systems in Germany, and the EWRS message published on Sunday 22 notified all EU
member states about the outbreak. Thus, media monitoring was less used for early
alerting, but rather to identify key aspects of the developing story (which were further
disseminated, e.g. in reports for the European Commission). Since MedISys covers
both general media and official sites, users were able to analyze their reports in
parallel.
Acknowledgments. We thank Prof Dr Nikolaos Stilianakis and Dr Agnès Rortais for

fruitful discussions and comments on the manuscript.
References
1. Linge, J.P., Steinberger, R., Weber, T.P., Yangarber, R., van der Goot, E., Al-Khudhairy,
D., Stilianakis, N.I.: Internet surveillance systems for early alerting and health threats.
Euro. Surveill. 14(13), 19162 (2009)
2. Linge, J.P., Belyaeva, J., Steinberger, R., Gemo, M., Fuart, F., Al-Khudhairy, D., Bucci,
S., Yangarber, R., van der Goot, E.: MedISys: Medical Information System. In:
Asimakopoulou, E., Bessis, N. (eds.) Advanced ICTs for Disaster Management and Threat
Detection: Collaborative and Distributed Frameworks, pp. 131–142. IGI Global (2010)
3. Mantero, J., Belyaeva, J., Linge, J.P.: How to maximise event-based surveillance web-
systems: the example of ECDC/JRC collaboration to improve the performance of
MedISys. JRC Scientific and Technical Reports (EUR collection). Publications Office of
the European Union (2011) ISBN 978-92-79-19755-0. ISSN 1018-5593, doi:
10.2788/69804
4. Frank, C., Werber, D., Cramer, J.P., Askar, M., Faber, M., Heiden, M.A., Bernard, H.,
Fruth, A., Prager, R., Spode, A., Wadl, M., Zoufaly, A., Jordan, S., Stark, K., Krause, G.:
The HUS Investigation Team: Epidemic Profile of Shiga-Toxin-Producing Escherichia
coli O104:H4 Outbreak in Germany - Preliminary Report. N Engl. J. Med., June 22 (2011)
5. Guglielmetti, P., Coulombier, D., Thinus, G., Van Loock, F., Schreck, S.: The Early
Warning and Response System for communicable diseases in the EU: an overview from
1999 to 2005. Euro. Surveill. 11(12), 666 (2006),
http://www.eurosurveillance.org/
ViewArticle.aspx?ArticleId=666
6. ProMED-mail: E. coli VTEC - Germany: RFI. ProMED-mail 2011, May 23:
20110523.1566, http://www.promedmail.org (accessed July 3, 2011)
7. Linge, J.P., Belyaeva, J.: Customization of MedISys for the monitoring of food and feed
hazards (2011),
http://www.efsa.europa.eu/en/supporting/pub/140e.htm
8. EFSA: Tracing seeds, in particular fenugreek (Trigonella foenum-graecum) seeds, in
relation to the Shiga toxin-producing E. coli (STEC) O104:H4 2011 Outbreaks in
Germany and France. EFSA Report. Question No EFSA-Q-2011-00817 (2011),
http://www.efsa.europa.eu/en/supporting/pub/176e.htm (issued on
July 05, 2011)
9. Paquet, C., Coulombier, D., Kaiser, R., Ciotti, M.: Epidemic intelligence: a new
framework for strengthening disease surveillance in Europe. Euro. Surveill. 11, 665
(2006), http://www.eurosurveillance.org/
ViewArticle.aspx?ArticleId=665
Syndromic Classification of Twitter Messages
Nigel Collier and Son Doan
National Institute of Informatics, Tokyo, Japan

collier@nii.ac.jp
http://born.nii.ac.jp/dizie
Abstract. Recent studies have shown strong correlation between so-

cial networking data and national influenza rates. We expanded upon
this success to develop an automated text mining system that classifies
Twitter messages in real time into six syndromic categories based on key
terms from a public health ontology. 10-fold cross validation tests were
used to compare Naive Bayes (NB) and Support Vector Machine (SVM)
models on a corpus of 7431 Twitter messages. SVM performed better
than NB on 4 out of 6 syndromes. The best performing classifiers showed
moderately strong F1 scores: respiratory = 86.2 (NB); gastrointestinal =
85.4 (SVM polynomial kernel degree 2); neurological = 88.6 (SVM poly-
nomial kernel degree 1); rash = 86.0 (SVM polynomial kernel degree 1);
constitutional = 89.3 (SVM polynomial kernel degree 1); hemorrhagic =
89.9 (NB). The resulting classifiers were deployed together with an EARS
C2 aberration detection algorithm in an experimental online system.
Keywords: epidemic intelligence, social networking, machine learning,

natural language processing.
1 Introduction
Twitter is a social networking service that allows users throughout the world to
communicate their personal experiences, opinions and questions to each other
using micro messages (‘tweets’). The short message style reduces thought in-
vestment [1] and encourages a rapid ‘on the go’ style of messaging from mobile
devices. Statistics show that Twitter had over 200 million users1 in March 2011,
representing a small but significant fraction of the international population across
both age and gender2 with a bias towards the urban population in their 20s and
30s. Our recent studies into novel health applications [2] have shown progress
in identifying free-text signals from tweets that allow influenza-like illness (ILI)
to be tracked in real time. Similar studies have shown strong correlation with
national weekly influenza data from the Centers for Disease Control and Pre-
vention and the United Kingdom’s Health Protection Agency. Approaches like
these hold out the hope that low cost sensor networks could be deployed as early
1
http://www.bbc.co.uk/news/business-12889048
2
http://sustainablecitiescollective.com/urbantickurbantick/20462/twitter-usage-
view-america

DIZIE 187
warning systems to supplement more expensive traditional approaches. Web-

based sensor networks might prove to be particularly effective for diseases that
have a narrow window for effective intervention such as pandemic influenza.
Despite such progress, studies into deriving linguistic signals that correspond
to other major syndromes have been lacking. Unlike ILI, publicly available gold
standard data for other classes of conditions such as gastrointestinal or neuro-
logical illnesses are not so readily available. Nevertheless, the previous studies
suggest that a more comprehensive early warning system based on the same prin-
ciples and approaches should prove effective. Within the context of the DIZIE
project, the contribution of this paper is (a) to present our data classification and
collection approaches for building syndromic classifiers; (b) to evaluate machine
learning approaches for predicting the classes of unseen Twitter messages; and
(c) to show how we deployed the classifiers for detecting disease activity. A fur-
ther goal of our work is to test the effectiveness of outbreak detection through
geo-temporal aberration detection on aggregations of the classified messages.
This work is now ongoing and will be reported elsewhere in a separate study.
1.1 Automated Web-Sensing

In this section we make a brief survey of recent health surveillance systems
that use the Web as a sensor source to detect infectious disease outbreaks. Web
reports from news media, blogs, microblogs, discussion forums, digital radio, user
search queries etc. are considered useful because of their wide availability, low
cost and real time nature. Although we will focus on infectious disease detection
it is worth noting that similar approaches can be applied to other public health
hazards such as earthquakes and typhoons [3,4].
Current systems fall into two distinct categories: (a) event-based systems that
look for direct reports of interest in the news media (see [5] for a review), and
(b) systems that exploit the human sensor network in sites like Twitter, Jaiku
and Prownce by sampling reports of symptoms/GP visits/drug usage etc. from
the population at risk [6,7,8]. Early alerts from such systems are typically used
by public health analysts to initiate a risk analysis process involving many other
sources such as human networks of expertise.
Work on the analysis of tweets, whilst still a relatively novel information
source, is related to a tradition of syndromic surveillance based on analysis of
triage chief complaint (TCC) reports, i.e. the initial triage report outlining the
reasons for the patient visit to a hospital emergency room. Like tweets they
report the patient’s symptoms, are usually very brief, often just a few keywords
and can be heavily abbreviated. Major technical challenges though do exist:
unlike TCC reports tweets contain a very high degree of noise (e.g. spam, opinion,
re-tweeting etc.) as well as slang (e.g. itcy for itchy) and emoticons which makes
them particularly challenging. Social media is inherently an informal medium
of communication and lacks a standard vocabulary although Twitter users do
make use of an evolving semantic tag set. Both TCC and tweets often consist of
short telegraphic statements or ungrammatical sentences which are difficult for
uncustomised syntactic parsers to handle.
188 N. Collier and S. Doan
In the area of TCC reports we note work done by the RODS project [9] that
developed automatic techniques for classifying reports into a list of syndromic
categories based on natural language features. The chief complaint categories
used in RODS were respiratory, gastrointestinal, botulinic, constitutional, neu-
rologic, rash, hemorrhagic and none. Further processes took aggregated data
and issued alerts using time series aberration detection algorithms. The DIZIE
project which we report here takes a broadly similar approach but applies it to
user generated content in the form of Twitter messages.
2 Method
DIZIE currently consists of the following components: (1) a list of latitudes

and longitudes for target world cities based on Twitter usage; (2) a lexicon of
syndromic keywords used as an initial filter, (3) a supervised machine learning
model that converts tweets to a word vector representation and then classifies
them according to six syndromes, (4) a post-processing list of stop words and
phrases that blocks undesired contexts, (5) a MySQL database holding historic
counts of positive messages by time and city location, used to calculate alerting
baselines, (6) an aberation detection algorithm, and (7) a graphical user interface
for displaying alerts and supporting evidence.
After an initial survey of high frequency Twitter sources by city location
we selected 40 world cities as candidates for our surveillance system. Sampling
in the runtime system is done using the Twitter API by searching for tweets
originating within a 30km radius of a city’s latitude and longitude, i.e. a typical
commuting/shopping distance from the city centre. The sampling rate is once
every hour although this can be shortened when the system is in full operation.
In this initial study we focussed only on English language tweets and how to
classify them into 6 syndromic categories which we describe below.
Key assumptions in our approach are that: (a) each user is considered to be a
sensor in the environment and as such no sensor should have the capacity to over
report. We controlled over reporting by simply restricting the maximum number
of messages per day to be 5 per user; (b) each user reports on personal obser-
vations about themselves or those directly known to them. To control (a) and
(b) and prevent over-reporting we had to build in filtering controls to mitigate
the effects of information diffusion through re-reporting, particularly for pub-
lic personalities and mass events. Re-tweets, i.e. repeated messages, and tweets
involving external links were automatically removed.
2.1 Schema Development
A syndrome is a collection of symptoms (both specific and non-specific) agreed

by the medical community that are indicative of a class of diseases. We chose
six syndrome classes as the targets of our classifier: constitutional, respiratory,
gastrointestinal, hemorrhagic, rash (i.e. dermatological) and neurological. These
were based on an openly available public health ontology developed as part
DIZIE 189
of the BioCaster project [10] by a team of experts in computational linguists,

public health, anthropology and genetics. Syndromes within the ontology were
based on RODS syndrome definitions and are linked to symptom terms - both
technical and laymen’s terms - through typed relations. We use these symptoms
(syndromic keywords) as the basis for searching Twitter and expanded them
using held out Twitter data.
2.2 Twitter Data

After defining our syndromes we examined a sample of tweets and wrote guide-
lines outlining positive and negative case definitions. These guidelines were then
used by three student annotators to classify a sample of 2000 tweets per syn-
drome into positive or negative for each of the syndrome classes. Data for training
was collected by automatically searching Twitter using the syndromic keywords
over the period 9th to 24th July 2010. No city filtering was applied when we
collected the training data. Typical positive example messages are: “Woke up
with a stomach ache!”, “Every bone in my body hurts”, and “Fever, back pain,
headache... ugh!”. Examples of negative messages are: “I’m exercising till I feel
dizzy”, “Cabin fever is severe right now”, “Utterly exhausted after days of house-
work”. Such negative examples include a variety of polysemous symptom words
such as fever in its senses of raised temperature and excitement and headache
in its senses of a pain in the head or an inconvenience. The negative examples
also include cases where the context indicates that the cause of the syptom is
unlikely to be an infection, e.g. headache caused by working or exercising. The
training corpus is characterised using the top 7 terms calculated by mutual asso-
ciation score in Table 1. This includes several spurious associations such as ‘rt’
standing for ‘repeat tweet’, ‘botox’ which is discussed extensively as a treatment
for several symptoms and ‘charice’ who is a new pop idol.
The final corpus was constructed from messages where there was total agree-
ment between all three annotators. This data set was used to develop and eval-
uate supervised learning classifiers in cross-fold validation experiments. A sum-
mary of the data set is shown in Table 2. Inter-annotator agreement scores
between the three annotators are given as Kappa showing agreement between
the two highest agreeing annotators. Kappa indicates strong agreement on most
syndromic classes with the noteable exception of gastrointestina and neurologi-
cal.
2.3 Classifier Models

DIZIE employs a two stage filtering process. Since Twitter many topics unre-
lated to disease outbreaks, DIZIE firstly requests Twitter to send it messages
that correspond to a set of core syndromic keywords, i.e. the same sampling
strategy used to collect training/testing data. These keywords are defined in the
BioCaster public health ontology [10]. In the second stage messages which are
putatively on topic are filtered more rigorously using a machine learning ap-
proach. This stage of filtering aims to identify messages containing ambiguous
Table 1. Top 7 terms by syndrome calculated by mutual information score. * indicates

a spurious association.
Resp Gastro Const Hemor Rash Neuro

throat stomach botox∗ pain road headache
sore ache body hemorrhage heat coma
cough gib charice∗ muscle arm worst
flu feel jaw tired tired gave
nose rt∗ hurts pray rash giving
rt∗ bad stomach brain itcy vertigo
cold worst sweating guiliechelon∗ face pulpo∗
Table 2. Structure of the annotated syndrome corpus of Twitter messages
Syndrome Positives (P) Negatives (N) P/N Kappa

Respiratory 627 738 0.85 0.67%
Gastrointestinal 489 676 0.72 0.49%
Neurological 549 434 1.26 0.42%
Rash 914 592 1.54 0.86%
Hemorrhagic 320 711 0.45 0.92%
Constitutional 1043 338 3.09 0.78%
words whose senses are not relevant to infectious diseases and messages where
the cause of the symptoms are not likely to be infectious diseases. About 70%
of messages are removed at this second stage.
To aid in model selection our experiments used two widely known machine
learning models to classify Twitter messages into a fixed set of syndromic classes:
Naive Bayes (NB) and support vector machines (SVM) [11] using a variety of
kernel functions. Both models were trained with binary feature vectors repre-
senting a dictionary index of words in the training corpus. i.e. a feature for the
test message was marked 1 if a word was present in the test message which
had been seen previously in the training corpus otherwise 0. No normalisation
of the surface words was done, e.g. using stemming, because of the high out of
vocabulary rate with tools trained on general language texts.
Despite the implausibility of its strong statistical independence assumption
between words, NB tends to perform strongly. The choice to explore keywords
as features rather than more sophisticated parsing and conceptual analysis such
as MPLUS [12] was taken from a desire to evaluate less expensive approaches
before resorting to time consuming knowledge engineering.
The NB classifier exploits an estimation of the Bayes Rule:

P (ck ) × m
i=1 P (fi |ck )
fi (d)
P (ck |d) = (1)
P (d)
where the objective is to assign a given feature vector for a document d consist-
ing of m features to the highest probability class ck . fi (d) denotes the frequency
count of feature i in document d. Typically the denominator P (d) is not com-
puted explicitly as it remains constant for all ck . In order to compute the highest
DIZIE 191
value numerator NB makes an assumption that features are conditionally inde-

pendent given the set of classes. Right hand side values of the equation are
estimates based on counts observed in the training corpus of classified Twit-
ter messages. We used the freely available Rainbow toolkit3 from CMU as the
software package.
SVMs have been widely used in text classification achieving state of the art
predictive accuracy. The major distinction between the two approaches are that
whereas NB is a generative classifier which forms a statistical model of each
class, SVM is a large-margin binary classifier. SVM operates as a two stage
process. Firstly the feature vectors are projected into a high dimensional space
using a kernel function. The second stage finds a maximum margin hyperplane
within this space that separates the positive from the negative instances of the
syndromic class. In practice it is not necessary to perfectly classify all instances
with the level of tolerance for misclassification being controlled by the C param-
eter in the model. A series of binary classifiers were constructed (one for each
syndrome) using the SVMLight software package 4 . We explored polynomial de-
gree 1, 2, 3 and radial basis function kernels.
2.4 Temporal Model
In order to detect unexpected rises in the stream of messages for each syndrome
we implemented a widely used change point detection algorithm called the Early
Aberration and Reporting System (EARS) C2 [13]. C2 reports an alert when its
test value St exceeds a number k of standard deviations above a historic mean:
St = max(0, (Ct − (μt + kσt ))/σt ) (2)
where Ct is the count of classified tweets for the day, μt and σt are the mean and
standard deviation of the counts during the history period, set as the previous
two weeks. k controls the number of standard deviations above the mean where
an alert is triggered, set to 1 in our system. The output of C2 is a numeric score
indicating the degree of abnormality but this by itself is not so meaningful to
ordinary users. We constructed 5 banding groups for the score and showed this
in the graphical user interface.
3 Results
3.1 Classifying Twitter Messages
Results for 10-fold cross validation experiments on the classification models are
shown in Table 3. Overall the SVM with polynomial degree 1 kernel outper-
formed all other kernels with other kernels generally offering better precision at
a higher cost to recall. Precision (Positive predictive) values ranged from 82.0 to
3
http://www.cs.cmu.edu/ mccallum/bow/rainbow/
4
http://svmlight.joachims.org/
93.8 for SVM (polynomial degree 1) and from 83.3 to 99.0 for NB. Recall (sensi-
tivity) values ranged from 58.3 to 96.2 for SVM (polynomial degree 1) and from
74.7 to 90.3 for NB. SVM tended to offer a reduced level of precision but better
recall. In the case of one syndrome (Hemorrhagic) we noticed an unusually low
level of recall for SVM but not for NB.
SVM’s performance seemed moderately correlated to the positive/negative
ratio in the training corpus and also showed weakness for the two classes (Hem-
orrhagic and Gastrointestinal) with the smallest positive counts. Naive Bayes
performed robustly across classes with no obvious correlation either to posi-
tive/negative ratio or the volume of training data. Low performance was seen
in both models for the gastrointestinal syndrome. This was probably due to the
low number of training examples resulting from the low inter-annotator agree-
ment on this class and the requirement for complete agreement between all three
annotators.
Table 3. Evaluation of automated syndrome classification using naive Bayes and Sup-
port Vector Machine models on 10-fold cross validation. P - Precision, R - Recall, F1
- F1 score. 1 SVM using a linear kernel, 2 SVM using a polynomial kernal degree 2, 3
SVM using a polynomial kernal degree 3, R SVM using a radial basis function kernel.
Naive Bayes SVM1 SVM2 SVM3 SVMR

Synd. P R F1 P R F1 P R F1 P R F1 P R F1
Resp. 90.3 82.4 86.2 85.4 82.5 83.8 83.0 71.0 76.5 86.4 61.3 71.7 66.7 3.2 6.2
Gast. 83.3 75.5 79.2 85.9 78.4 81.8 92.7 79.2 85.4 91.4 66.7 77.1 73.1 39.6 51.3
Neur. 98.2 74.7 84.8 83.2 95.0 88.6 77.9 98.2 86.9 62.4 98.2 76.3 90.0 63.0 74.1
Rash 94.5 76.1 84.3 82.0 90.6 86.0 76.9 91.2 83.4 67.7 94.5 78.9 60.7 100.0 75.5
Hem. 89.4 90.3 89.9 93.8 58.3 71.7 100.0 50.0 66.7 100.0 50 66.7 87.5 43.8 58.3
Con. 99.0 79.8 88.4 83.6 96.2 89.3 83.6 93.3 88.2 78.6 99.0 87.7 76.5 100 86.7
3.2 Technology Dissemination

An experimental service for syndromic surveillance called DIZIE has been im-
plemented based on the best of our classifier models and we are now observ-
ing its performance. The service is freely available from an online portal at
http://born.nii.ac.jp/dizie. As shown in Figure 3.2 the graphical user interface
(GUI) for DIZIE shows a series of radial charts for each major world city with
each band of the chart indicating the current level of alert for one of the six
syndromes. Alerting level scores are calculated using the Temporal Model pre-
sented above. Each band is colour coded for easy recognition. Alerting levels
are calculated on the classified twitter messages using the EARS C2 algorithm
described above. Data selection is by city and time with drill down to a selection
of user messages that contributed to the current level. Trend bars show the level
of alert and whether the trend is upwards, downwards or sideways. Charting is
also provided over an hourly, daily, weekly and monthly period. The number of
positively classified messages by city is indicated in Figure 3.2 for a selection of
cities.
DIZIE 193
Fig. 1. Radial graphs showing syndromic alert levels for major world cities. Colour
coding on the radial segments indicates the alerting degree automatically assigned to
a syndrome in a city based on the previous hour’s Twitter counts and the previous 2
weeks as a baseline. The page is updated every hour. Clicking on the graph for a city
displays the frequency graph and also the matching tweets for the current hour.
Fig. 2. Number of Tweets by a sample of major world cities classified by DIZIE during
the period 2nd March 2011 to 31st August 2011
Navigation links are provided to and from BioCaster, a news event alerting
system, and we expect in the future to integrate the two systems more closely to
promote greater situation awareness across media sources. Access to the GUI is
via regular Web browser or mobile device with the page adjusting automatically
to fit smaller screens.
4 Conclusion
Twitter offers unique challenges and opportunities for syndromic surveillance.
Approaches based on machine learning need to be able (a) to handle biased
data, and (b) to adjust to the rapidly changing vocabulary to prevent a flood
of false positives when new topics trend. Future work will compare keyword
classifiers against more conceptual approaches such as [12] and also compare the
performance characteristics of change point detection algorithms.
Based on the experiments reported here we have built an experimental appli-
cation called DIZIE that samples Twitter messages originating in major world
cities and automatically classifies them according to syndromes. Access to the
system is openly available. Based on the outcome of our follow up study we
intend to integrate DIZIE’s output with our event-based surveillance system
BioCaster which is currently used by the international public health community.
Acknowledgements. This work was in part supported by grant in aid support

from the National Institute of Informatics’ Grand Challenge Project (PI:NC).
We are grateful to Reiko Matsuda Goodwin for commenting on the user interface
in the early stages of this study and helping in data collection for the final system.
References
1. Java, A., Song, X., Finin, T., Tseng, B.: Why we twitter: Understanding microblog-
ging usage and communities. In: Proc. 9th WebKDD and 1st SNA-KDD Workshop
on Web Mining and Social Network Analysis, August 12. ACM (2007)
2. Collier, N., Nguyen, S.T., Nguyen, M.T.N.: OMG U got flu? analysis of shared
health messages for bio-surveillance. Biomedical Semantics 2(suppl. 5), S10 (2011)
3. Earle, P.: Earthquake twitter. Nature Geoscience 3(4), 221–222 (2010),
doi:10.1038/ngeo832
4. Sakaki, T., Okazaki, M., Matsuo, Y.: Earthquake shakes twitter users: real-time
event detection by social sensors. In: Proc. of the 19th International World Wide
Web Conference, Raleigh, NC, USA, pp. 851–860 (2010)
5. Hartley, D., Nelson, N., Walters, R., Arthur, R., Yangarber, R., Madoff, L., Linge,
J., Mawudeku, A., Collier, N., Brownstein, J., Thinus, G., Lightfoot, N.: The land-
scape of international biosurveillance. Emerging Health Threats J. 3(e3) (January
2010), doi:10.1093/bioinformatics/btn534
6. Szomszor, M., Kostkova, P., De Quincey, E.: swineflu : Twitter predicts swine flu
outbreak in 2009 (December 2009)
7. Lampos, V., De Bie, T., Cristianini, N.: Flu Detector - Tracking Epidemics on
Twitter. In: Balcázar, J.L., Bonchi, F., Gionis, A., Sebag, M. (eds.) ECML PKDD
2010. LNCS, vol. 6323, pp. 599–602. Springer, Heidelberg (2010)
8. Signorini, A., Segre, A.M., Polgreen, P.M.: The use of twitter to track levels of dis-
ease activity and public concern in the U.S. during the influenza a h1n1 pandemic.
PLoS One 6(5), e19467 (2011)
9. Wagner, M.M., Espino, J., Tsui, F.C., Gesteland, P., Chapman, W., Ivanov, W.,
Moore, A., Wong, W., Dowling, J., Hutman, J.: Syndrome and outbreak detec-
tion using chief-complaint data - experience of the real-time outbreak and disease
surveillance project. Morbidity and Mortality Weekly Report (MMWR) 53(suppl.),
28–31 (2004)
DIZIE 195
10. Collier, N., Doan, S., Kawazoe, A., Matsuda Goodwin, R., Conway, M., Tateno, Y.,
Ngo, Q., Dien, D., Kawtrakul, A., Takeuchi, K., Shigematsu, M., Taniguchi, K.:
BioCaster:detecting public health rumors with a web-based text mining system.
Bioinformatics 24(24), 2940–2941 (2008), doi:10.1093/bioinformatics/btn534
11. Joachims, T.: Text categorization with support vector machines: Learning with
many relevant features. In: Nédellec, C., Rouveirol, C. (eds.) ECML 1998. LNCS,
vol. 1398, pp. 137–142. Springer, Heidelberg (1998)
12. Christensen, L.M., Haug, P.J., Fiszmann, M.: Mplus: A probabilistic medical lan-
guage understanding model. In: Proceedings of the Workshop on Natural Language
Processing in the Biomedical Domain, Philadelphia, USA (July 2002)
13. Hutwagner, L., Thompson, W., Seeman, M.G., Treadwell, T.: The bioterrorism
preparedness and response early aberration reporting system (EARS). J. Urban
Health 80(2), i89–i96 (2003)
Towards Spatial Description-Based Integration
of Biomedical Atlases
Nurzi Juana Mohd Zaizi1 and Albert Burger1,2

1
Department of Computer Science, Heriot-Watt University, Edinburgh, UK
2
MRC Human Genetics Unit, Edinburgh, UK
Abstract. Biomedical imaging has become ubiquitous in both, basic re-

search and the clinical sciences. Technology advances, and the resulting
multitude of imaging modalities, have led to a sharp rise in the quantity
and quality of such images. Whether for epidemiological studies, educa-
tional uses, monitoring the clinical progress of a patient or translational
science purposes, being able to integrate and compare such image-based
data has developed into an increasingly critical component in the Life
Sciences and eHealth domain. Image processing-based solutions have dif-
ficulties when the underlying morphologies are too different. Ontology-
based solutions often lack spatial precision. In this paper, we describe a
compromise solution which captures location in biomedical images via
spatial descriptions using so-called fiducial points. The work is discussed
in the context of biomedical atlases and includes, in addition to the in-
troduction of the basic method, some experimental performance results.
Keywords: spatial description, integration of biomedical images/atlases.
1 Introduction
Patients are now routinely undergoing a variety of medical imaging investiga-
tions, such as Magnetic Resonance Imaging (MRI) and Computed Tomography
(CT) scanning, and the images resulting from these investigations become part
of the patients’ medical records. The same as well as other imaging techniques,
e.g. optical imaging, are also used in preclinical studies and the Life Sciences.
The work presented in this paper is rooted in the latter and uses examples
from biomedical atlases, where we consider an atlas to consist of the image data
components, a set of labels describing structures in the images, and mappings
between them. There are questions of data integration within the domain of
clinical images, within the Life Sciences image datasets, as well as between hu-
man and model organism data. The latter being particularly of interest in the
translational sciences.
Biologists have access to a variety of biomedical atlases. Many of these atlases
are data sources for the same experimental field, for example, mouse gene expres-
sion data. Though storing the same type of data, different experimental designs,
varying analysis of results and different update routines have caused the data in
these atlases to be different. The consequence is that these atlases may provide

Biomedical Atlases 197
different results even for the same query. It is vital that multiple resources in
the same field are used so that full and complete results can be generated for
the query. The comparative clinical issue is the integration of different medical
images for a single patient, or the comparison of images of multiple patients with
the same disease.
A biomedical atlas consists of a graphical model, the ontology associated with
the graphical model and a mapping between these two. The ontology contains
a collection of anatomical domains and relations between these domains. The
graphical model is the image for a mammalian with those anatomical domains.
This paper proposes the integration of these data sources by mapping the im-
ages of biomedical atlases using spatial descriptions. Given two images I1 and I2,
mapping one image onto another means that, for each anatomical space in image
I1, we try to find a corresponding space, which has the same intended mean-
ing, in image I2. For this study, we circumvent the extra complexity of image
segmentation by considering anatomical domains that can be easily segmented.
More specifically, we explore 2D image space of mouse embryo domains.
Mappings anatomical spaces concern a number of issues. Different images may
have a different number of segmented regions causing one structure to correspond
to parts of several structures, and vice versa. Furthermore, even if these images
may have the same anatomical structures, the morphology may vary with scale,
orientation and the position of the structure. In addition, different biomedical
atlases may have the same segmented images but may use different anatomical
names causing interoperability issue of finding correspondences anatomical re-
gions between these images. An efficient representation structure is necessary to
conceptualize anatomical space of an image to guide the mapping process. This
paper explores spatial description-based approach for the linking of images for
the integration of biomedical atlases.
Section 2 presents an overview of image mapping approaches. The proposed
integration approach is described in Section 3. Section 4 provides experimental
results of the proposed approach. In Section 5, a discussion is presented. Finally,
a conclusion in Section 6.
2 An Overview of Image Mapping Approaches
This section discusses two approaches for mapping. In particular, its focus is
on the following approaches: (1) ontology based mapping (2) image processing
based mapping. Ontology based mapping depends on spatial relations between
anatomical regions, whereas, mapping using image processing depends on fiducial
points.
2.1 Spatial Relations: Ontology Based
This section describes the mapping between anatomical spaces across images us-
ing ontologies. Mappings based on an ontology start by segmenting the image of
a biomedical atlas according to its anatomical regions. Subsequently, the regions
198 N.J.M. Zaizi and A. Burger
can be linked to the appropriate concepts in the atlas’ anatomy ontology. Two
regions are then mapped according to the similarity of their spatial relationships.
Given two atlas anatomy ontologies O1 and O2 , if anatomical structure A1 in
ontology O1 has the relationships A1 is adjacent to B1 , and A1 is adjacent to C1
then its equivalent anatomical structure, A2 , in ontology O2 , must be adjacent
to B2 and C2 , where the latter two correspond to B1 and C1 , respectively. The
integration of biomedical atlases can then be achieved by linking between their
respective anatomy ontologies.
The concepts of spatial relations have been well employed in ontologies by
both FMA [1] and Bittner et al. 2008 [2] to describe anatomical space in the
biomedical domain. In general, spatial relations between anatomical entities are
described using relationships from the following categories:
Mereological relations describe the concept of parthood between the whole
and its parts, e.g., finger is part of hand, hand is part of the arm etcetera.
Topological relations describe the concept of adjacency, discreteness, and
connectedness among entities. Two entities are described as being adjacent
when they are close to each other, but not connected. Discrete entities are
not connected. If two entities have a common anatomical space, such that
they partially coincide or are externally attached with one another, they
are said to be connected., e.g., two entities are externally connected if the
distance between them is zero and do not overlap, for example, in human
major parts of the joint, the synovial cavity is externally connected to the
synovial membrane [2].
Location relations describe the concept of relative location between entities
that may coincide wholly or partially without being part of one another, for
example, the brain is located in (but not part of ) cranial cavity.
Based on spatial relations, for example, anatomical region x in Figure 1(a) is
mapped to the result region y in Figure 1(b) if x is described as:
’adjacent(x, midgut), adjacent(x, liver)’
Fig. 1. Based on spatial adjacency between (a) anatomical region x with other anatom-
ical regions will map x to (b) the result region y
2.2 Fiducial Points: Image Processing Based
This section discusses the mapping between biomedical atlases based on image
processing techniques. These methods start by examining the pixels in an image
to classify them into regions, e.g. [3,4]. Classification is by the pixel’s intensity
level. Subsequently, a registration algorithm is required to identify equivalent
regions, across images, based on pixels. In addition, based on the pixel classifica-
tion, fiducial points can be located. A fiducial point is a point in space in either
2D or 3D, typically an anatomical landmark which is easily recognizable in an
image, usually identified by human experts and possibly assisted by auto/semi-
automated image processing algorithms. These fiducial points are typically used
for registration experimentation image of canonical atlas. Izard and Jedynak [5]
describes a registration approach which employs a Bayesian model to detect these
points in order to map between regions across images. Registration technique as
proposed by Khaissidi et al. 2009 [6] uses the Hough Transform algorithm to
align medical images, based on fiducial points extracted from the two compared
images. However, the drawback of image processing based mapping in general
is that it has the possibility to fail if there is a large variation in pixel/voxel
intensity [7].
3 Concept of Spatial Description
3.1 Spatial Description Based on Fiducial Points and a Set of

Spatial Relations
The proposed approach of mapping involved the concepts of a query region,

fiducial points and fiducial lines. A fiducial point is a point in space. A fiducial
line is made up by creating a straight line through a pair of fiducial points. A
query region is made up of connected multiple single-elements within a closed
boundary. Given two images I1 and I2, mapping one image onto another starts
by selecting the same fiducial points in both images. We then describe a query
region using spatial relations between the query region with respect to the fidu-
cial points. Two regions from different images are then mapped according to the
similarity of their spatial description. For example, if query region X in image I1
is described as X is north of fiducial point P1 and X is west of fiducial point P2,
then its equivalent region in image I2 must be a region that is located north of
fiducial point P1 and west of fiducial point P2. Figure 2 depicts the framework
of the approach. The image processing based mapping inspires the idea of using
fiducial points. Because a fiducial point can become a point of reference for an
anatomical location, this paper intends to describe a query region based on these
points using spatial relations. The idea of using spatial relations, on the other
hand, is inspired by ontology based mapping. Because a spatial relation can de-
scribe the location of a region in space, this paper proposes to describe a query
region using fiducial points and a set of spatial relations. By the use of spatial
relations, this approach works independently of image pixel/voxel intensities.
Fig. 2. Ontology-based mappings align images by identifying correspondences among

elements of two ontologies based on spatial relations between anatomical structures.
The image processing based mappings align images based on equivalent pixel/voxel
intensities corresponding the fiducial points
On the other hand, the use of fiducial points allows this approach to work in-
dependently of spatial relations between segmented regions. The approach does
not intend to include a large number of concepts in spatial relations as that
replicates the ontology mapping approach. The entire spatial area of an image
should be conceptualized with a small number of fiducial points such that the
attempt is not a replicate to the image processing mapping approach. We now
summarize the formalism of the approach. We define directional relations as
D = {northOf, eastOf, southOf, westOf } (1)
We describe a query region x in an image as
SQ = {r(x, fi ) | r ∈ D and fi ∈ (Fpoint ∪ Fline )} (2)
where SQ is the spatial description for query region x with respect to a fiducial
point Fpoint = {p1 , p2 ,..., pn } or a fiducial line Fline = {l1 , l2 ,..., ln }
Figure 3 depicts two images of mouse embryo with 6 fiducial points and 15
fiducial lines. The simplified description for query region X is described as:
’southOf(X, P6P2), eastOf(X, P1P4), northOf(X,P2P5), westOf(X, P1P3)’
Note that, in the description, we label a fiducial line according to its pair of
fiducial points. The location highlighted in Figure 3(b) denotes the matched
location corresponding to the description.
4 Experimental Results
A series of experiments were conducted to demonstrate how fiducial points and
a set of spatial relations can be used to describe locations. For the experiments,
an image representing the mouse embryo was used and 102 spatial regions were
annotated in the image. The image generated 97104 query regions each of size
Fig. 3. Spatial description based on fiducial points and a set of spatial relations maps
(a) query region X to (b) result region Y
50x50 squared pixels, 68154 query regions each of size 100x100 squared pixels,
44204 query regions each of size 150x150 squared pixels, and 25254 query regions
each of size 200x200 squared pixels. For all query regions of size 50x50 squared
pixels, the first query region starts at the top-left corner of the image and is
increased every time by one pixel in order to generate the following query region
and so on. Query regions of other sizes are also generated by following this step
of one pixel. The idea of using query regions is to test the mappings of pixels in a
query region of one image to pixels in a region of another image based on fiducial
points. The percentage of accuracy is calculated by dividing the total number
of pixels in X by the total number of pixels in result region Y , and multiply
by 100 (see Figure 3). Figure 4(A) depicts the average percentage of accuracy
served by number of fiducial points. Results show that the more fiducial points
100 100
Set A
Average % of Accuracy
Average % of Accuracy

Set B
80 80
Set C

60
60
Query Region of Size 200x200 Pixels 40
40 Query Region of Size 150x150 Pixels
Query Region of Size 100x100 Pixels 20
Query Region of Size 50x50 Pixels
20
0 80 240 400 560 720 880 50x50 100x100 150x150 200x200
A Number of Fiducial Points B Area Size of the Query Region
Fig. 4. (A) Average percentage of accuracy served by number of fiducial points. The
more fiducial points are included the higher the average percentage of accuracy gets.
Moreover, the average percentage of accuracy substantially increases as the query region
area size gets larger. (B) Average percentage of accuracy in three different positioning
sets of 8 fiducial points served by query region area size. The same number of fiducial
point place at different positions produce different average percentage of accuracy.
were included, the more accurate the mapping was. In addition, the mapping ac-
curacy substantially increases as the query region area size gets larger. Fiducial
points provide qualitative spatial relations to describe locations. Therefore, the
more fiducial points are used the more spatial relations are available to describe
locations, which increases the average percentage of accuracy. In general, spatial
descriptions will return a location which is either larger or equal to the actual
area location. Thus, for cases where spatial descriptions for the corresponding
query regions do not return locations that are exactly equal to the actual loca-
tion, the larger the size of the query region, the more accurate it is to the actual
area, by which contribute to much higher accuracy value compared to the smaller
one. Figure 4(B) depicts the average percentage of accuracy in three different
positioning sets of 8 fiducial points served by query region size. Results show
that the same number of fiducial points placed at different positions produce
different accuracy. The positions of fiducial points determine spatial relations
made available to describe locations. Because the location is determined by spa-
tial descriptions, different positioning set for the same number of fiducial points
certainly contributes to different spatial descriptions to describe locations, which
produce different average percentages of accuracy. Overall, with the appropriate
number of fiducial points used and better selection of fiducial point location,
mappings can be improve in terms of accuracy.
5 Discussion
The definition for best match criteria is important in any mapping algorithm.
Because anatomical structures exist at different range of scale, arrangement and
the position, there is a possibility for an exact copy of location corresponding
the query region in one image to be unavailable in another image. The proposed
spatial description approach at the current state perform mappings by returning
a location that satisfies all spatial relation constraints corresponding to a query
region. However, this may not be necessary. Therefore, the google-style matches
can be considered. This can be done by specifying a range, for example, allowing
for a distance limit from a fiducial line, which will return a location given by the
range.
A preliminary experiment has also been conducted to compare the perfor-
mance of spatial descriptions based on fiducial points and a set of spatial re-
lations with the following approaches: (1) spatial description based on spatial
relationships between segmented regions (2) spatial description based on fiducial
points and a set of spatial relations, integrated with spatial relations between
segmented regions. Experimental results verified that mapping using spatial de-
scription based on spatial relationships between segmented regions managed to
produce better accuracy compared to spatial description based on fiducial points
and a set of spatial relations. However, this result cannot be used to benchmark
the overall mapping performance produced by spatial description based on fidu-
cial points. Depending on better selection of fiducial point locations or by in-
creasing the number of fiducial points used, the mapping accuracy can be further
increased. Furthermore, experimental results have verified that the approach of

mapping using spatial description based on fiducial points and a set of spatial
relations, integrated with spatial relations between segmented regions can yield
significantly higher mapping accuracy compared to using either approach alone.
6 Conclusion and Future Work

This paper explores spatial description based approach to facilitate data inte-
gration across biomedical atlases. The most important feature of our approach
is that the spatial description, which is rule-based, can provide the means to fa-
cilitate the mapping between images of biomedical atlases. Future work includes
research on selection of fiducial points where the combination can give perfor-
mance, as effective as both ontology-based and image processing algorithm; and
analyse the capability of spatial description to facilitate data integration be-
tween (1) images (i.e. from biomedical atlases), (2) natural-language description
of space (i.e. free text from biomedical literature) (3) database warehouses (i.e.
structured database of biomedical facts).
Acknowledgments. This work is partially funded by EU projects CUBIST

(FP7-ICT-2010-257403) and RICORDO (FP7-ICT-2009-248502).
References
1. Rosse, C., Mejino, J.L.V.: The Foundational Model of Anatomy Ontology. In:
Anatomy Ontologies for Bioinformatics: Principles and Practise. Springer, Heidel-
berg (2008)
2. Bittner, T., Donelly, M., Goldberg, L.J., Neuhaus, F.: Spatial Representation and
Reasoning. In: Anatomy Ontologies for Bioinformatics: Principles and Practise.
Modeling Principles and Methodologies. Springer, Heidelberg (2008)
3. Boccignone, G., Napoletano, P., Ferraro, M.: Embedding diffusion in variational
bayes: A technique for segmenting images. International Journal of Pattern Recog-
nition and Artificial Intelligence 22(1), 811–827 (2008)
4. Wyawahare, M.V., Patil, P.M., Abhyankar, H.K.: Image Registration Techniques:
An overview. International Journal of Signal Processing, Image Processing and Pat-
tern Recognition 2(3), 1–5 (2009)
5. Izard, C., Jedynak, B.: Bayesian Registration for Anatomical Landmark Detection.
In: IEEE International Symposium on Biomedical Imaging, Arlington, VA (2006)
6. Khaissidi, G., Tairi, H., Aarab, A.: A fast medical image registration using feature
points. ICGST-GVIP Journal 9(3), 19–24 (2009)
7. Walia, E., Suneja, A.: A Conceptual Study on Image Matching Techniques. Global
Journal of Computer Science and Technology 10(12), 83–88 (2010)
8. Human-fetal week 10 sagittal plane,
http://php.med.unsw.edu.au/embryology/
index.php?title=Embryonic Development
Author Index
Aidemark, Jan 85 Hunter, Ruth F. 170

Alves, Bruno 17 Huptych, Michal 26
Argyropaidas, Panagiotis 76 Huser, Martin 26
Arvanitis, Theodoros N. 105
Askenäs, Linda 85 Ioanitescu, Radu 109
Atkinson, Martin 178
Janku, Petr 26
Bacon, Liz 146
Bahati, Raphael 1 Katt, Basel 97
Belyaeva, Jenya 178 Kee, Frank 170
Breu, Ruth 97 Kontogiannis, Vasilis 76
Bromuri, Stefano 128 Kostkova, Patty 80, 154
Burger, Albert 196 Krampf, Johannes 128
Bursa, Miroslav 26
Lagani, Vincenzo 76
Chen, Liming 9 Langford, Gavin 105
Chronaki, Catherine 76 Lawrenson, John 154
Chudacek, Vaclav 26 Lhotska, Lenka 26
Coleman, Alfred 162 Lim Choi Keung, Sarah N. 105
Collier, Nigel 58, 186 Linge, Jens P. 178
Luck, Michael 138
Davis, Michael 170
Delaney, Brendan 138
Mair, Richard 97
de Quincey, Ed 146
Mantero, Jas 178
Detschew, Vesselin 34
Miles, Simon 138
Di Tullio, Eugenio 43
Mulvenna, Maurice 9
Doan, Son 58, 186
O’Gorman, Josephine 146
Eiring, Øystein 51
Ogunsina, Ire 105
Ekinci, Okan 34
Elloumi, Lamia 158
Papı́ková, Vendula 122
Fernando, Juanita 93 Pawar, Pravin 158
Fowler, David 154 Peters, Nicolette 93
Fuart, Flavio 178 Potass, Dalenca 162
Gattnar, Eva 34 Quinn, Darren 9

Grasso, Floriana 43
Guy, Stacey 1, 114 Ratzki-Leewing, Alexandria 1, 114
Gwadry-Sridhar, Femida 1, 114 Rentea, Victor 109
Ruiz, Juan 128
Hammond, Simon 154
Hermens, Hermie 158 Sandmark, Hélène 67
Herselman, Marlien E. 162 Schabetsberger, Thomas 97
Hocking, Avril 146 Schumacher, Michael 17, 128
206 Author Index
Serbanati, Luca Dan 109 van Beijnum, Bert-Jan 158

Slaughter, Laura 51 van der Goot, Erik 178
Smedberg, Åsa 67 Van Staa, Tjeerd 138
Spilka, Jiri 26 Vasilateanu, Andrei 109
Szomszor, Martin 154 Vo, Bao-Khanh Ho 58
Taweel, Adel 138 Walker, Simon 146

Trojer, Thomas 97
Tully, Mark A. 170 Zaizi, Nurzi Juana Mohd 196
Tyler, Edward 105 Zhao, Lei 105
Tyson, Gareth 138 Zvolský, Miroslav 122

Electronic Healthcare

Uploaded by

Copyright:

Available Formats

Electronic Healthcare

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Electronic Healthcare

Uploaded by

Copyright:

Available Formats

Lecture Notes of the Institute

for Computer Sciences, Social Informatics

ISSN 1867-8211 e-ISSN 1867-822X

Library of Congress Control Number: 2012934259

CR Subject Classification (1998): K.4, I.2, J.3-4, H.4, C.2, H.5

It is our great pleasure to introduce the special issue of LNICST comprising

Ed de Quincey and Gawesh Jawaheer chaired a joint posters and demos

Knowledge Transfer Chair

Scientific Media Chair

Social Media Chair

IEEE EMB Representative

Senior Program Committee

Technical Program Committee

David Isern Universitat Rovira i Virgili, Tarragona, Spain

Pekka Ruotsalainen National Institute of Health and Welfare,

Session 1: Social Media Analysis

An Examination of the Behaviour of Young and Older Users

Session 2: Knowledge Integration and EPR

Obstetric Medical Record Processing and Information Retrieval . . . . . . . . 26

Session 3: Personalisation and Patient Support

A Model for a Motivational System Grounded on Value Based Abstract

An Assessment of the Potential for Personalization in Patient Decision

Session 4: Early Warning Systems and Mobile

Web-Based Stress Management System Goes Mobile:

Use of SMS for Tsunami Early Warnings at a Table Top Exercise . . . . . . 76

Session 5: Games and Learning

Patient-Centered Care as a Learning Process . . . . . . . . . . . . . . . . . . . . . . . . 85

Wireless Computer Games and Applications in the Medical Education

Session 6: Security, Privacy and Prevention

Modelling a User Authorisation and Data Access Framework

Software Engineering-Inspired Approach to Prevention Healthcare . . . . . . 109

Session 7: Online Support for Professionals and

Cardio Online Reader/COR: A Web 2.0-Based Tool Aimed at Clinical

Session 8: Agents in eHealth

Session 9: Online Communities of Practice

Session 10: eHealth Solutions

Session 11: Social Media Surveillance

Syndromic Classiﬁcation of Twitter Messages . . . . . . . . . . . . . . . . . . . . . . . . 186

Towards Spatial Description-Based Integration of Biomedical Atlases . . . 196

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

Stacey Guy, Alexandria Ratzki-Leewing, Raphael Bahati,

Lawson Health Research Institute,

Abstract. Social media represents a new frontier in disease surveillance.

Keywords: social media, review, population surveillance, data mining.

This paper is organized as follows. The methodology used in this systematic

2.1 Data Sources

2.2 Data Extraction and Synthesis

references within the retrieved publications and Google searching) revealed 16

3.1 Study Characteristics

3.2 Results of Individual Studies

In realizing the potential of infodemiology in healthcare, it is important to consider

4.1 Advantages and Disadvantages of Mining Social Media

improve public access to health surveillance information. As data-mining sources

4.2 Study Limitations

4.3 Research Directions

Summary. Evolving the use of infodemiology in healthcare will involve the

Darren Quinn, Liming Chen, and Maurice Mulvenna

School of Computing and Mathematics, University of Ulster,

Abstract. Increased use in recent years of social networking applications such

Keywords: Interaction Analysis, Social Networking, Social Computing.