Chapter 5

The Story of MIMIC

Roger Mark

Take Home Messages

• MIMIC is a Medical Information Mart for Intensive Care and consists of several
comprehensive data streams in the intensive care environment, in high levels of
richness and detail, supporting complex signal processing and clinical querying
that could permit early detection of complex problems, provide useful guidance
on therapeutic interventions, and ultimately lead to improved patient outcomes.
• This complicated effort required a committed and coordinated collaboration
across academic, industry, and clinical institutions to provide a radically open
access data platform accessible by researchers around the world.

5.1 The Vision

Patients in hospital intensive care units (ICUs) are physiologically fragile and
unstable, generally have life-threatening conditions, and require close monitoring
and rapid therapeutic interventions. They are connected to an array of equipment
and monitors, and are carefully attended by the clinical staff. Staggering amounts of
data are collected daily on each patient in an ICU: multi-channel waveform data
sampled hundreds of times each second, vital sign time series updated each second
or minute, alarms and alerts, lab results, imaging results, records of medication and
fluid administration, staff notes and more. In early 2000, our group at the
Laboratory of Computational Physiology at MIT recognized that the richness and
detail of the collected data opened the feasibility of creating a new generation of
monitoring systems to track the physiologic state of the patient, employing the
power of modern signal processing, pattern recognition, computational modeling,
and knowledge-based clinical reasoning. In the long term, we hoped to design

© The Author(s) 2016 43

MIT Critical Data, Secondary Analysis of Electronic Health Records,
DOI 10.1007/978-3-319-43742-2_5
44 5 The Story of MIMIC

monitoring systems that not only synthesized and reported all relevant measure-
ments to clinicians, but also formed pathophysiologic hypotheses that best
explained the observed data. Such systems would permit early detection of complex
problems, provide useful guidance on therapeutic interventions, and ultimately lead
to improved patient outcomes.
It was also clear that although petabytes of data are captured daily during care
delivery in the country’s ICUs, most of these data were not being used to generate
evidence or to discover new knowledge. The challenge, therefore, was to employ
existing technology to collect, archive and organize finely detailed ICU data,
resulting in a research resource of enormous potential to create new clinical
knowledge, new decision support tools, and new ICU technology. We proposed to
develop and make public a “substantial and representative” database gathered from
complex medical and surgical ICU patients.

5.2 Data Acquisition

In 2003, with colleagues from academia (Massachusetts Institute of Technology),

industry (Philips Medical Systems), and clinical medicine (Beth Israel Deaconess
Medical Center, BIDMC) we received NIH (National Institutes of Health) funding
to launch the project “Integrating Signals, Models and Reasoning in Critical Care”,
a major goal of which was to build a massive critical care research database. The
study was approved by the Institutional Review Boards of BIDMC (Boston, MA)
and MIT (Cambridge, MA). The requirement for individual patient consent was
waived because the study would not impact clinical care and all protected health
information was to be de-identified.
We set out to collect comprehensive clinical and physiologic data from all ICU
patients admitted to the multiple adult medical and surgical ICUs of our hospital
(BIDMC). Each patient record began at ICU admission and ended at final discharge
from the hospital. The data acquisition process was continuous and invisible to staff.
It did not impact the care of patients or methods of monitoring. Three categories of
data were collected: clinical data, which were aggregated from ICU information
systems and hospital archives; high-resolution physiological data (waveforms and
time series of vital signs and alarms obtained from bedside monitors); and death data
from Social Security Administration Death Master Files (See Fig. 5.1).

5.2.1 Clinical Data

Bedside clinical data were downloaded from archived data files of the CareVue
Clinical Information System (Philips Healthcare, Andover, MA) used in the ICUs.
Additional clinical data were obtained from the hospital’s extensive digital archives.
The data classes included:
5.2 Data Acquisition 45

Fig. 5.1 MIMIC II data sources

• Patient demographics
• Hospital administrative data: admission/discharge/death dates, room tracking,
billing codes, etc.
• Physiologic: hourly vital signs, clinical severity scores, ventilator settings, etc.
• Medications: IV medications, physician orders
• Lab tests: chemistry, hematology, ABGs, microbiology, etc.
• Fluid balance data
• Notes and reports: Discharge summaries; progress notes; ECG, imaging and
echo reports.

5.2.2 Physiological Data

Physiological data were obtained with the technical assistance of the monitoring
system vendor. Patient monitors were located at every ICU patient bed. Each
monitor acquired and digitized multi-parameter physiological waveform data,
processed the signals to derive time series (trends) of clinical measures such as heart
rate, blood pressures, and oxygen saturation, etc., and also produced bedside
monitor alarms. The waveforms (such as electrocardiogram, blood pressures, pulse
plethysmograms, respirations) were sampled at 125 Hz, and trend data were
updated each minute. The data were subsequently stored temporarily in a central
database server that typically supported several ICUs. A customized archiving
agent created and stored permanent copies of the physiological data. The data were
physically transported from the hospital to the laboratory every 2–4 weeks where
they were de-identified, converted to an open source data format, and incorporated
into the MIMIC II waveform database. Unfortunately, limited capacity and
46 5 The Story of MIMIC

intermittent failures of the archiving agents limited waveform collection to a

fraction of the monitored ICU beds.

5.2.3 Death Data

The Social Security Death Master files were used to document subsequent dates of
death for patients who were discharged alive from the hospital. Such data are
important for 28-day and 1-year mortality studies.

5.3 Data Merger and Organization

A major effort was required in order to organize the diverse collected data into a
well-documented relational database containing integrated medical records for each
patient. Across the hospital’s clinical databases, patients are identified by their
unique Medical Record Numbers and their Fiscal Numbers (the latter uniquely
identifies a particular hospitalization for patients who might have been admitted
multiple times), which allowed us to merge information from many different hos-
pital sources. The data were finally organized into a comprehensive relational
database. More information on database merger, in particular, how database
integrity was ensured, is available at the MIMIC-II web site [1]. The database user
guide is also online [2].
An additional task was to convert the patient waveform data from Philips’
proprietary format into an open-source format. With assistance from the medical
equipment vendor, the waveforms, trends, and alarms were translated into WFDB,
an open data format that is used for publicly available databases on the National
Institutes of Health-sponsored PhysioNet web site [3].
All data that were integrated into the MIMIC-II database were de-identified in
compliance with Health Insurance Portability and Accountability Act standards to
facilitate public access to MIMIC-II. Deletion of protected health information from
structured data sources was straightforward (e.g., database fields that provide the
patient name, date of birth, etc.). We also removed protected health information
from the discharge summaries, diagnostic reports, and the approximately 700,000
free-text nursing and respiratory notes in MIMIC-II using an automated algorithm
that has been shown to have superior performance in comparison to clinicians in
detecting protected health information [4]. This algorithm accommodates the broad
spectrum of writing styles in our data set, including personal variations in syntax,
abbreviations, and spelling. We have posted the algorithm in open-source form as a
general tool to be used by others for de-identification of free-text notes [5].
5.4 Data Sharing 47

5.4 Data Sharing

MIMIC-II is an unprecedented and innovative open research resource that grants

researchers from around the world free access to highly granular ICU data and in
the process substantially accelerates knowledge creation in the field of critical care
medicine. The MIMIC Waveform Database is freely available to all via the
PhysioNet website, and no registration is required. The MIMIC Clinical Database is
also available without cost. To restrict users to legitimate medical researchers,
access to the clinical database requires completion of a simple data use agreement
(DUA) and proof that the researcher has completed human subjects training [6].
The MIMIC-II clinical database is available in two forms. In the first form, inter-
ested researchers can obtain a flat-file text version of the clinical database and the
associated database schema that enables them to reconstruct the database using a
database management system of their choice. In the second form, interested
researchers can gain limited access to the database through QueryBuilder, a
password-protected web service. Database searches using QueryBuilder allow users
to familiarize themselves with the database tables and to program database queries
using the Structured Query Language. Query output, however, is limited to 1000 rows
because of our laboratory’s limited computational resources. Accessing and pro-
cessing data from MIMIC-II is complex. It is recommended that studies based on the
MIMIC-II clinical database be conducted as collaborative efforts that include clinical,
statistical, and relational database expertise. Detailed documentation and procedures
for obtaining access to MIMIC-II are available at the MIMIC-II web site [1]. The
current release of MIMIC-II is version 2.6, containing approximately 36,000 patients,
including approximately 7000 neonates, and covering the period 2001–2008. At the
present time approximately 1700 individuals worldwide in academia, industry, and
medicine have been credentialed to access MIMIC-II and are producing research
results in physiologic signal processing, clinical decision support, predictive algo-
rithms in critical care, pharmacovigilance, natural language processing, and more.

5.5 Updating

In 2008 the hospital made a major change in the ICU information system technology
and in ICU documentation procedures. The Philips CareVue system was replaced
with iMDsoft’s MetaVision technology. In 2013 we began a major update to MIMIC
to incorporate adult ICU data for the period 2008–2012. The effort required learning
the entirely new data schema of MetaVision, and merging the new data format with
the existing MIMIC design. The new MetaVision data included new data elements
such as physician progress notes, oral and bolus medication administration records,
etc. Updated data were extracted from hospital archives and from the SSA death files
for the newly added patients. Almost two years of effort was invested to acquire,
organize, debug, normalize and document the new database before releasing it.
48 5 The Story of MIMIC

MIMIC-III includes 20,000 new adult ICU admissions, bringing the total to
approximately 60,000. The new database is known as MIMIC-III, and the acronym
has been recast as “Medical Information Mart for Intensive Care” [7].

5.6 Support

Support of the MIMIC databases includes: credentialing new users, administration of

the authorized user list (i.e. users who have signed the DUA and have been granted
permission to access MIMIC-II), user account creation, password resets and
granting/revoking permissions. The servers providing MIMIC-II include authenti-
cation, application, database and web servers. All systems must be monitored,
maintained, upgraded and backed up; the maintenance burden continues to increase as
the number of database users grows. The engineering staff at LCP attempt to answer
user queries as needed. Common questions are added to list of frequently asked
questions on the MIMIC website and we regularly update our online documentation.

5.7 Lessons Learned

Building and distributing MIMIC-like databases is challenging, complex, and

requires the cooperation and support of a number of individuals and institutions.
A list of some of the more important requirements follows (Table 5.1).

Table 5.1 Health data 1. The availability of digitized ICU and hospital data including
requirements structured and unstructured clinical data and high resolution
waveform and vital sign data
2. A cooperative and supportive hospital IT department to assist
in data extraction
3. A supportive IRB and hospital administration to assure both
protection of patient privacy and release of de-identified data
to the research community
4. Adequate engineering and data science capability to design
and implement the database schema and to de-identify the
data (including the unstructured textual data)
5. Sophisticated signal processing expertise to reformat and
manage proprietary waveform data streams
6. Cooperation and technical support of equipment vendors
7. Adequate computational facilities for data archiving and
distribution
8. Adequate technical and administrative personnel to provide
user support and credentialing of users
9. Adequate financial support
5.8 Future Directions 49

5.8 Future Directions

The MIMIC-III database is a powerful and flexible research resource, but the
generalizability of MIMIC-based studies is somewhat limited by the fact that the
data are collected from a single institution. Multi-center data would have the
advantages of including wider practice variability, and of course a larger number of
cases. Data from international institutions would add still greater strength to the
database owing to the even larger variations in practice and patient populations.
Our long-term goal is to create a public, multi-center, international data archive
for critical care research. We envisage a massive, detailed, high-resolution ICU data
archive containing complete medical records from patients around the world. The
difficulty of such a project cannot be understated; nevertheless we propose to lay the
foundation for such a system by developing a scalable framework that can readily
incorporate data from multiple institutions, capable of supporting research on
cohorts of critically ill patients from around the world.

Acknowledgments The development and maintenance of the MIMIC and PhysioNet resources
have been funded by the National Institute of Biomedical Imaging and Bioengineering (NIBIB)
and the National Institute of General Medical Sciences (NIGMS) over the period 2003 to present.
Grants R01EB1659, R01EB017205, R01GM104987, and U01EB008577.

Open Access This chapter is distributed under the terms of the Creative Commons
Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/
4.0/), which permits any noncommercial use, duplication, adaptation, distribution and reproduction
in any medium or format, as long as you give appropriate credit to the original author(s) and the
source, a link is provided to the Creative Commons license and any changes made are indicated.
The images or other third party material in this chapter are included in the work’s Creative
Commons license, unless indicated otherwise in the credit line; if such material is not included in
the work’s Creative Commons license and the respective action is not permitted by statutory
regulation, users will need to obtain permission from the license holder to duplicate, adapt or
reproduce the material.

References

1. MIMIC-II Web Site. http://physionet.org/mimic2

2. MIMIC User Guide. http://physionet.org/mimic2/UserGuide/
3. WaveForm DataBase Data Format. http://www.physionet.org/physiotools/wfdb.shtml
4. Neamatullah I, Douglass M, Lehman LH, Reisner A, Villarroel M, Long WJ, Szolovits P,
Moody GB, Mark RG, Clifford GD (2008) Automated de-identification of free-text medical
records. BMC Med Inform Decis Mak 8:32. doi:10.1186/1472-6947-8-327
5. Deidentification Software. http://www.physionet.org/physiotools/deid/
6. Accessing MIMIC. http://www.physionet.org/mimic2/mimic2_access.shtml
7. MIMIC-III Website. http://mimic.physionet.org/