A TOOLKIT FOR NATIONAL

DENGUE BURDEN ESTIMATION

© World Health Organization 2018
Some rights reserved. This work is available under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 IGO licence
(CC BY-NC-SA 3.0 IGO; https://creativecommons.org/licenses/by-ncsa/3.0/igo).
Under the terms of this licence, you may copy, redistribute and adapt the work for non-commercial purposes, provided the work
is appropriately cited, as indicated below. In any use of this work, ther should be no suggestion that WHO endorses any specific
organization, products or services. The use of the WHO logo is not permitted. If you adapt the work, then you must license your
work under the same or equivalent Creative Commons licence. If you create a translation of this work, you should add the following
disclaimer along with the suggested citation: “This translation was not created by the World Health Organization (WHO). WHO
is not responsible for the content or accuracy of this translation. The original English edition shall be the binding and authentic
edition”.
Any mediation relating to disputes arising under the licence shall be conducted in accordance with the mediation rules of the World
Intellectual Property Organization.
Suggested citation. A Toolkit for national dengue burden estimation, Geneva: World Health Organization; 2018. Licence: CC
BY-NC-SA 3.0 IGO.
Cataloguing-in-Publication (CIP) data. CIP data are available at http://apps.who.int/iris. Sales, rights and licensing. To pur-
chase WHO publications, see http://apps.who.int/bookorders. To submit requests for commercial use and queries on rights and
licensing, see http://www.who.int/about/licensing.
Third-party materials. If you wish to reuse material from this work that is attributed to a third party, such as tables, figures or images,
it is your responsibility to determine whether permission is needed for that reuse and to obtain permission from the copyright holder.
The risk of claims resulting from infringement of any third-party-owned component in the work rests solely with the user.
General disclaimers. The designations employed and the presentation of the material in this publication do not imply the expres-
sion of any opinion whatsoever on the part of WHO concerning the legal status of any country, territory, city or area or of its
authorities, or concerning the delimitation of its frontiers or boundaries. Dotted and dashed lines on maps represent approximate
border lines for which there may not yet be full agreement.
The mention of specific companies or of certain manufacturers’ products does not imply that they are endorsed or recommended by
WHO in preference to others of a similar nature that are not mentioned. Errors and omissions excepted, the names of proprietary
products are distinguished by initial capital letters.
All reasonable precautions have been taken by WHO to verify the information contained in this publication. However, the pub-
lished material is being distributed without warranty of any kind, either expressed or implied. The responsibility for the interpretation
and use of the material lies with the reader. In no event shall WHO be liable for damages arising from its use.
Design and layout: Patrick Tissot WHO/NTD
WHO/CDS/NTD/VEM/2018.05
CONTENTS

Acknowledgements .............................................................................................................................v
Abbreviations ...................................................................................................................................vi
Introduction ...................................................................................................................................... 1
Aim ............................................................................................................................................... 1
Measuring burden ............................................................................................................................ 2
Levels of disease severity of dengue infections............................................................................... 2
Why current surveillance systems may misestimate the clinical burden of dengue ................................. 4
Toolkit overview ............................................................................................................................ 7
Suggested dengue surveillance standards for burden estimation ............................................................... 10
1.1. Collating and reporting dengue surveillance data ................................................................. 10
1.2. Completeness assessment of national dengue surveillance data............................................... 13
1.3. Over-diagnosis analysis ................................................................................................... 15
1.4. Under-diagnosis analysis ................................................................................................. 18
1.5. Febrile cohorts ............................................................................................................... 20
Febrile illness surveys ................................................................................................... 20
Measuring inapparent incidence within febrile illness cohorts ............................................... 22
1.6. Measuring total infections using cross-sectional seroprevalence surveys ..................................... 23
Burden calculation .......................................................................................................................... 24
2.1. Using the burden calculator .............................................................................................. 24
2.2. Comparison with global dengue burden modelling estimates .................................................. 29
Further extension studies .................................................................................................................... 30
3.1. Spatial variation in burden ............................................................................................... 30
3.2. Economic burden of dengue ............................................................................................. 31
3.3. Combined arboviral burden ............................................................................................. 32
3.4. Reducing uncertainty in burden estimates ............................................................................ 33
3.5. Evaluating change in burden over time ............................................................................... 34
References .................................................................................................................................... 36

iii
iv
ACKNOWLEDGEMENTS

This toolkit is the product of a series of meetings of the Hombach (WHO), Thomas Jaenisch (University of Hei-
World Health Organization (WHO) Advisory Group on delberg), Mohamed Jamsheed (WHO), Jacqueline Lim
dengue burden estimation. The document was developed (International Vaccine Institute), Robert Reiner Jr (Institute
on behalf of WHO by Oliver J. Brady (London School of for Health Metrics and Evaluation), Thomas Scott (Uni-
Hygiene & Tropical Medicine). versity of California, Davis), Donald Shepard (Brandeis
University), Jeff Stanaway (Institute for Health Metrics and
All members of the Advisory Group also contributed to
Evaluation) and Raman Velayudhan (WHO).
and approved the final version of the toolkit document
and are listed here (in alphabetical order): Mathieu Ban- Input was also provided by various individuals from the
gert (WHO), Gamaliel Gutierrez Castillo (Pan American health ministries and WHO offices in Brazil, Cambodia,
Health Organization), Ng Lee Ching (National Envi- Maldives, Mexico and Sri Lanka as well as by Maïna
ronment Agency’s Environmental Health Institute), Derek L’Azou (Sanofi Pasteur).
Cummings (University of Florida), Zoey Diaz (BMGF),
Funding for the production of this toolkit and its devel-
Neil Ferguson (Imperial College London), Simon Hay
opment was gratefully received from the Bill & Melinda
(Institute for Health Metrics and Evaluation), Joaquim
Gates Foundation (BMGF).

v
 ABBREVIATIONS

DENV dengue virus ELISA enzyme-linked immunosorbent assay

PPV positive predictive value HI haemagglutination inhibition
UFI unidentified febrile illness PRNT plaque reduction neutralization test
DHS Demographic and Health Surveys FOI force of infection
IgM/G immunoglobulin M or G WHO World Health Organization
PCR polymerase chain reaction

vi
INTRODUCTION

Dengue is a viral disease vectored by Aedes mosqui- Understanding how these levels of burden are distributed
toes that has spread throughout the tropical world since over time, age and space in a particular country is im-
the mid twentieth century. Existing reactive control efforts portant for:
have failed to stop the expansion of dengue virus trans- • determining how to allocate optimally the limited re-
mission and many areas now have endemic circulation sources available for dengue prevention and control;
of all four dengue virus (DENV) serotypes. New strategies
• calculating the economic burden of dengue to build
are needed to reverse this trend; and to be effective,
the case for investment in surveillance and control
they must be based on accurate quantitative information
measures;
about the burden of dengue. The WHO Global strate-
gy for dengue prevention and control, 2012–2020 (1) • in assisting policy-makers in allocating resources for
highlights the urgent need to generate new estimates of sustained support to intervention programmes, based
the burden of dengue that are acceptable to all Member on impact;
States. • identifying gaps in surveillance or best practice clini-
cal management;
While few dengue infections result in death, the high
number of non-fatal cases imposes a heavy burden of • evaluating the impact of control activities locally and
morbidity and makes dengue an increasing priority in the internationally;
more than 120 countries that it affects. Measuring the • improving understanding of the local epidemiology
full burden imposed by dengue requires estimating the and the potential for DENV spread;
incidence of all levels of disease severity from minor to • predicting the likely impact of new vaccination and
major. Severe cases can result in death or require lengthy vector control strategies, either alone or in combina-
hospitalization, with potential long-term side-effects; clin- tion;
ical dengue cases can overwhelm healthcare facilities; • increasing recognition and treatment of potentially un-
sub-clinical infections can limit attendance at work or diagnosed dengue cases in order to improve disease
school; and inapparent infections can act as a reservoir outcomes; and
of infection that undermines surveillance and control ef-
forts. • improving the targeting of vaccination programmes to
areas where they will be most effective.

1
 AIM

The aim of this toolkit is to guide countries on how to best ment from an official healthcare provider, i.e. a hospital
estimate their current burden of dengue by combining or a clinic, and instead is self-managed or managed by
existing data from dengue surveillance systems with on- friends or family possibly with non-prescription medica-
going research efforts to measure the community burden tion until recovery.
of dengue.
3. Non-severe dengue
MEASURING BURDEN The case classification of non-severe dengue is defined in
the 2009 WHO guidelines (2) as travel to or resident in
a dengue endemic area plus two of the following criteria:
LEVELS OF DISEASE SEVERITY OF DENGUE nausea or vomiting, rash, aches and pains, tourniquet test
INFECTIONS positive, leukopenia or any warning sign. This definition
includes the classifications of dengue with and without
warning signs and can include cases in whom dengue
Human infection with dengue viruses can result in a broad
is either diagnosed using only clinical criteria (probable
range of disease manifestations of varying levels of sever-
dengue in (2)) or those that are laboratory confirmed.
ity (Fig. 1). These levels of infection severity are largely
distinguished by how and where the patient is treated or
managed. They include1: 4. Severe dengue
The case classification of severe dengue is defined in the
1. Inapparent dengue 2009 WHO guidelines (2) as patient presentation with
severe plasma leakage, severe haemorrhage and severe
A dengue virus infection that does not result in any dis-
organ impairment.
ruption to the daily routine of the individual. This includes
infections that are entirely asymptomatic as well as those
that have very mild symptoms that do not have an impact 5. Fatal dengue
on the affected individual’s routine but that may be detect- A death in which acute dengue virus infection is the sole
able upon detailed examination or questioning. or leading cause.

2. Self-managed dengue
1
While further subcategories of infection severity are also often recog-
A dengue virus infection that results in disruption to the
nized (e.g. hospitalized, non-hospitalized), for the purpose of burden
daily routine of the individual (e.g. missing school or estimation, the above categories make best use of the most commonly
work) but that does not result in seeking diagnosis or treat- available data types.

2
 Death

ce
lan Severe dengue

Cl
eil

in
ica
v
sur

lb
ve

urd
ssi

en
Pa

Non-severe dengue

Co
e
nc

Self-managed dengue

mm
illa

un
ve

ity
sur

bu
e
tiv

rde
Ac

Inapparent dengue

n
Fig. 1 Disease outcomes of a dengue virus infection divided by the clinical burden (blue line) that is measured by routine
passive surveillance and the community burden (red line), which can only be measured by community-based sampling, i.e.
enhanced disease detection, longitudinal cohorts and cluster studies.

3
 More severe disease outcomes are more rare, but affected WHY CURRENT SURVEILLANCE SYSTEMS
individuals are also more reliable to diagnose and more MAY MISESTIMATE THE CLINICAL BURDEN
likely to seek treatment. Routine nationwide healthcare
facility-based surveillance (hereafter routine surveillance)
OF DENGUE
is the regular reporting of disease data by all institutions
that see patients and are part of a reporting network (3). Established national dengue surveillance systems provide
By design, these systems are only able to report cases the most comprehensive and frequently available source
that seek treatment (upper four levels in Fig. 1) and are of data for estimating the burden of dengue. However,
typically less sensitive and specific as the severity of the these passive surveillance systems will not capture all den-
disease becomes milder. To detect infections in a commu- gue infections, and the types of dengue infections they
nity that does not seek treatment (bottom two levels in Fig. do capture are usually unrepresentative in severity, age
1) requires community-based sampling, where individuals distribution, geography and time.
are visited at home, school or their place of work and First, there are epidemiological reasons why a dengue
questioned about recent fever events and serologically virus infection may not be detected. Because secondary
tested for dengue virus infection. Such community-based infection with a different dengue virus serotype is more
surveys are often highly logistically and financially costly likely to lead to a more severe dengue disease outcome,
and are technically unfeasible at a national scale; as a younger individuals experiencing their first dengue virus
result, they are only implemented at a few key sites. infection are more likely to have an inapparent disease
The vast majority of routine surveillance data for dengue outcome and may not seek care or treatment for their mild
comes from passive surveillance, with some countries illness. These age biases will vary in different areas that
additionally having a small number of sentinel sites that have a different history of dengue virus type invasion and
conduct community-based sampling to measure the full persistence as well as different population demographics.
spectrum of dengue disease severity in selected loca- Second, there are sociodemographic factors that will af-
tions. This toolkit assesses why existing surveillance sys- fect whether an individual with a symptomatic dengue
tems may not accurately represent the clinical burden of infection seeks treatment at an official healthcare facility
dengue, explains the methods for correcting inaccuracies (where they can be correctly diagnosed and treated) or
and then advises on community-based surveys to measure chooses to self-treat (referred to here as treatment-seeking
the community burden of dengue. behaviour). Treatment-seeking behaviour can vary consid-
erably depending on cost, accessibility and availability
of care and, as a result, will differ in different areas and
at different times (4).
Assuming that a symptomatic dengue infected individual
seeks treatment at an official healthcare facility and thus
has the opportunity to be recorded by routine passive sur-

4
veillance, there are further reasons why such individuals High Low
may be missed or incorrectly recorded. The rest of this representativeness representativeness
section explains these common gaps in dengue surveil-
lance and how they can be minimized so that surveil- A B
lance data can be used to give a reliable measure of the

High coverage
clinical burden of dengue.
There are three main reasons why existing surveillance
data may not give a good estimate of the true clinical
dengue burden (see also the examples given in Table 1):
+ +
1. Incomplete coverage of the surveillance system + +
2. Cases detected are not representative of the total
DENV infected population
C D
3. Inconsistency of diagnosis, surveillance and treat-
ment seeking in different places or over time

Low coverage
No surveillance system will have perfect coverage, rep-
resentativeness and consistency, but that does not mean + +
that its data cannot be useful for burden estimation. If
+ +
current surveillance data have low, but known, coverage
and good representativeness, reliable extrapolations can
be made to areas where coverage is lacking (Fig. 2C). Legend
If a surveillance system has good coverage but low rep- Dengue case + Dengue death
resentativeness, it can be used to provide reliable esti-
mates of subsections of dengue burden (e.g. cases but Fig. 2 The concept of surveillance system coverage and representa-
not deaths, Fig. 2B). Estimates of other subsections (e.g. tiveness; surveillance system coverage (orange dotted lines) of den-
gue cases (filled circles) and deaths (crosses) can vary with respect
deaths) can then be made using other data sources (e.g. to coverage and representativeness – changes in coverage and rep-
data from sentinel sites, or from the published literature). resentativeness over time or space lead to changes in consistency.
While most surveillance systems will be somewhere be-
tween these extreme examples, quantifying coverage,
representativeness and consistency is essential to reliably
estimate dengue burden and how it changes. The meth-
ods explained in the first section of this toolkit explain
how these characteristics of the surveillance system can
be practically measured.

5
 Table 1. Understanding reasons why data from current dengue surveillance may under- or mischaracterize dengue burden

1. Coverage
Reasons why a surveillance system might miss dengue cases
1.1. Geography • Isolated clinics are not part of the surveillance system
1.2. Policy • Not legally notifiable
• Private clinics are not part of the surveillance system
1.3 Capability and incentives • Some facilities may not have sufficient laboratory support to accurately diagnose dengue
• Some facilities may see reporting as a time burden that detracts from their time for patient
care, or consider that the cases reflect unfavourably on their institution
1.3. Reporting fidelity • Staff do not complete case notification
1.4. Misdiagnosis • Dengue cases diagnosed as other febrile illnesses
• Co-morbidities disguise dengue or lead to it not being reported as a contributing factor to
disease
2. Representativeness
Reasons why a surveillance system might be more likely to miss some types of dengue infections more than others
2.1. Severity • More severe cases are more likely to be correctly diagnosed
• Reporting of deaths is subject to a higher degree of priority and independent review and thus
deaths are correctly reported more frequently
2.2. Age • Clinical dengue signs are more recognizable in children than adults
• Sentinel surveillance may be confined to specific age groups, e.g. paediatric hospitals
3. Consistency
Reasons why surveillance systems may vary in coverage or representativeness
3.1. Time • Time constraints during outbreaks mean staff do not have time to notify cases
• Surveillance systems improve over longer time periods to increase the number of healthcare
centres that are part of the system and provide greater capacity for laboratory diagnosis of
dengue
• Dengue clinical case definitions may change
3.2. Standardization • Different physicians may or may not diagnose a dengue case or may classify it at a different
level of severity based on differing personal professional opinion or prior experience.

6
TOOLKIT OVERVIEW

The aim of this toolkit is to estimate the national annual and community) as it brings many advantages for current
burden of dengue when applied in a given country or and future control efforts (Table 2). This can be achieved
subnational area. This is achieved through a series of six by reviewing previously published research efforts to con-
sub-objectives that: assemble existing data (1.1), amend duct fever cohorts and seroprevalence surveys or by part-
gaps in surveillance completeness (1.2), and correct for nering with researchers to conduct new surveys that fulfil
both over (1.3) and under (1.4) diagnosis to estimate the requirements of burden estimation set out in this toolkit.
the true clinical burden of dengue. These clinical burden Key parameter estimates from each of these separate
estimates should then be combined with new or existing studies can then be entered into a simple spreadsheet
community-based surveys to estimate the symptomatic calculator to estimate the full burden of dengue at all
(1.5) and inapparent (1.6) community-based burden of levels of infection severity. Model-based estimates of key
dengue (Fig. 3). parameters will also be provided for comparison and for
The toolkit describes the use of simple data analysis tech- guidance when primary data are not yet available.
niques that can be implemented by programme man- Countries are encouraged to document the burden esti-
agers in the health ministry and be paired with existing mation process in a formal, publicly available report. This
datasets. Even countries without extensive and well es- report should include the sources for each piece of evi-
tablished surveillance systems are encouraged to at least dence used and key details about how it was collected
complete the clinical burden estimation components (Ta- and processed; each piece of evidence should conform
ble 2). However, most countries should be able to extend to the Guidelines for accurate and transparent health es-
these efforts to estimate the full burden of dengue (clinical timates reporting, or GATHER, criteria (5). Technical sup-
port for implementation can be provided if requested.

7
 Death

Dengue surveillance data

Under-diagnosis analysis
Over-diagnosis analysis
Completeness analysis
e
Severe dengue

nc

Cl
illa

ini
ve

c al
sur

bu
ve

rde
ssi

n
Pa

Non-severe dengue

Febrile cohort
Cm
e
nc

Self-managed dengue

m
illa

un
e

ity
ur v

bu
es

rd
tiv

en

survey
(Sero)
Ac

Sero
Inapparent dengue

Key

New clinical New community

Existing data Data analysis
study study

Fig. 3 Overview of data collection and analyses recommended in this toolkit and levels of dengue disease severity for which they
provide burden estimates

8
Table 2. Feasibility, requirements and rationale for completing clinical or full burden estimation using the toolkit; the clinical
burden of dengue can be estimated by completing activities (in orange), while calculating the full dengue burden requires
completion of activities (in both orange and blue)

Dengue burden Analysis component Data needed Rationale for burden estimation
component Clinical burden Full burden
Clinical burden Surveillance data Existing data • Gaps in surveillance • Full economic burden
Completeness analysis Existing data • Gaps in best treatment • Optimize targeting of
practice interventions
Under-diagnosis analysis Existing data • Direct medical costs • Evaluate effectiveness
Over-diagnosis analysis Minor new study of control
• Predict the effects of
Existing published data / new interventions (e.g.
Febrile cohorts collaboration with vaccines)
research partners
• Set achievable and
Community burden Existing published data measureable future
/ collaboration with goals
Seroprevalence surveys research partners / in
preparation for dengue
vaccination

9
 SUGGESTED DENGUE
SURVEILLANCE STANDARDS
FOR BURDEN ESTIMATION
The methods suggested for estimating dengue burden in Disease severity is disaggregated into “dengue”, “severe
this toolkit are best suited to surveillance systems with the dengue” and “fatal dengue” following the case classifi-
following criteria: cation guidelines set out in the 2009 WHO criteria for
• Countries with regular dengue transmission with a dengue diagnosis (2). The dengue category includes cas-
national passive surveillance system established in all es both with and without warning signs. Fatal dengue, for
areas at risk. the purpose of burden estimation, is defined as a death
in which acute dengue infection was the sole or one of
• Countries in which at least a subset of dengue cases
the primary causes. Cases should have mutually exclusive
are confirmed using dengue-specific rapid or labora-
final outcomes, i.e. one infection results in dengue or se-
tory-based diagnostic tests (Table 8 (2)).
vere dengue, not both.
• Countries in which the number of dengue cases can
It is recognized that there have been known issues
be disaggregated by both disease severity (at least
around the practicality and utility of this revised case defi-
fatal/non-fatal) and age (at least children/adults).
nition (6, 7), which have led many countries to adapt
The focus of this toolkit is to estimate the average burden their own dengue case definitions or continue to use the
of dengue for a given country and over a given year. 1997 WHO case definition (8). These inconsistencies
Dengue burden and the way in which it is measured complicate international comparison of dengue burden
will vary considerably both sub-nationally and at differ- and, ideally, case definitions would be globally stand-
ent times of the year. It is therefore recommended that ardized. However, given that such standardization or
subsequent burden studies further investigate spatial and reclassification of existing cases is outside the scope of
temporal variations in burden for surveillance so as to this burden analysis, countries are encouraged to report
target control efforts appropriately (see 4.1–4). However, cases as originally diagnosed and then to detail the case
for the purposes of this first assessment we assume that definitions used in the supplementary reporting question-
dengue burden is being estimated at the national level naire (Table 4).
for one given year.
Dengue cases at different levels of severity are to be dis-
aggregated by patient age, and confirmation method.
1.1 COLLATING AND REPORTING DENGUE Notified cases refer to patients who have been reported
SURVEILLANCE DATA as dengue through the routine notifiable diseases report-
ing system. Depending on how the surveillance system
is designed, these cases may include some or all of the
As a first step, existing data on dengue cases from the following: suspected dengue (where dengue is one of
past epidemiological year should be disaggregated by multiple suspected differential diagnoses), clinically con-
age, disease severity and method of confirmation, as firmed dengue (where the case is confirmed on the basis
shown in Table 3. of the patients symptoms) or test confirmed dengue (where

10
dengue has been virologically or serologically confirmed testing, including the criteria for who should be tested for
through testing of patient blood or serum). The criteria for dengue (e.g. everyone or just clinically complex cases),
national standard practice in notification should be in- what tests are used to confirm dengue and the threshold
cluded in the surveillance system questionnaire (Table 4). used to define dengue positive from dengue negative (or
In addition to notification, the disaggregation of the num- indeterminate test result) should also be included in the
bers of patient samples that were tested (either rapid test surveillance system questionnaire (Table 4).
or laboratory-based) for suspected dengue infection and If recorded age groups are more aggregated in nation-
the numbers of these tested samples that had a positive al dengue data than presented below (e.g. <15 years,
result should also be recorded (Table 3). To enable in- 15+ years), then cases should be divided equally among
ternational comparison of burden estimates, details of all sub age bands.

Table 3. Data template for reformatting of national dengue surveillance data for clinical burden analysis

Age (in Dengue Severe dengue Fatal dengue

years)
Notified Tested* Test* confirmed Notified Notified
(clinical and/or test (clinical and/or test (clinical and/or test
confirmed) confirmed) confirmed)
<1
1–4
5–9
10–14
15–19
20–29
30–39
40–49
50–59
60–69
70+

* Applicable tests include dengue-specific point of care (rapid) or laboratory-based tests that are appropriate for the day of illness on which samples
are taken (see Table 7).

11
 Table 4. Dengue surveillance system questionnaire to accompany national dengue surveillance data detailing each country’s
specific notification, clinical and laboratory case definitions

Notification criteria

At what point in a patient’s diagnostic history does it become com- Suspected dengue ☐
pulsory for them to be notified to the national surveillance system Clinically confirmed dengue ☐
(“notified dengue”)? (tick all that apply)
Test* confirmed dengue ☐
Other (please specify):________________________________

Clinical case definitions**

What clinical case definition is used for diagnosis of dengue?

What clinical case definition is used for diagnosis of severe

dengue?

What criteria are used to determine if a death was due to dengue?

Dengue testing

What types of tests or assays are most frequently used to confirm

dengue infection? And what percentage of samples are tested by
each method?
(e.g. NS1 rapid test, IgM/IgG ELISA, PCR)

What are the threshold criteria for positivity for each laborato-
ry-based assay? (e.g. copy number, PRNT50, titre)

What are the criteria for when testing should be done? All suspected dengue cases
(circle one) Only sentinel sites
Only clinically complex cases
Only vulnerable individuals (e.g. pregnant)
Other (please specify): ________________________________

*Applicable tests include dengue-specific point of care (rapid) or laboratory-based tests that are appropriate for the day of illness on which samples
are taken (see Table 7).
** Clinical diagnosis refers to dengue diagnosis based on patient symptoms and basic non dengue-specific laboratory procedures, e.g. full blood
count or haematocrit.

12
1.2 COMPLETENESS ASSESSMENT OF If private healthcare facilities are not required to notify
NATIONAL DENGUE SURVEILLANCE DATA dengue cases, the proportion of cases treated in the pri-
vate sector can be estimated by comparing the total num-
ber of febrile illness episodes treated by major private
The first step in burden estimation is to estimate the num- healthcare providers with those in the public sector in
ber of clinical dengue cases that have been diagnosed equivalent catchment areas, e.g. cities or counties. Such
by physicians but not notified to the national surveil- a sample should ensure it includes a variety of socioec-
lance system. This can occur due to a variety of factors. onomic levels (e.g. income levels) and service provision
Table 5 summarizes the most important reasons of rele- environments (e.g. rural vs urban). Alternatively, patients
vance to burden estimation. seeking treatment at public and private hospitals can be
Measures to reconcile each of these completeness gaps interviewed using a structured questionnaire that asks
can be obtained through an internal assessment of the about their treatment seeking pathway. Such question-
dengue surveillance system. Such assessments are a com- naires are important for identifying individuals who might
mon feature of many dengue surveillance systems to en- seek primary care and sometimes diagnosis in the public
sure compliance with national notification practices (3). sector before choosing treatment in the private sector, or
Identifying healthcare facilities that are not part of the vice versa.
national notification system can be done by comparing Notification fidelity can be estimated in a subsample of
the total number of healthcare centres that are registered healthcare centres through retrospective analysis to count
and considered to be in an area at risk of dengue trans- dengue cases in records of treatment (e.g. line lists of
mission with the number that have reported any dengue admitted patients or patient billing records) that can be
cases in the past five years. Outpatient only clinics should compared with the total number of notified cases record-
be excluded from this analysis if it is not routine practice ed in the dengue surveillance system over the same time
to report dengue outpatients or if all dengue patients they period.
do see are referred to centres that are able to report den-
gue. If outpatient clinics are included, it should be con-
sidered that they may have lower notification fidelity than
other healthcare providers; and it should be ensured that
calculations of notification fidelity include a representative
number of outpatient clinics.

13
 Table 5. Key reasons why dengue cases may be correctly diagnosed but not notified to the national surveillance system

Reason Definition Example

Gaps in surveillance Logistical barriers to the notification of • Resource constraints mean small clinics lack the personnel
system coverage correctly diagnosed dengue patients or equipment (e.g. computer) to notify case
• Inconsistent Internet access in remote areas prevent isolated
clinics using a web-based reporting system
• Only sentinel sites may record specific details about cases,
such as disease severity

Private sector treatment Dengue patients who are diagnosed in • Individuals with private healthcare insurance may choose to
the private healthcare sector, which is not be treated in a private healthcare facility
required to notify dengue cases • An individual may choose to pay for private treatment to
avoid long waiting times at public healthcare facilities
during a dengue outbreak

Notification fidelity Cases that have been correctly diagnosed • Insufficient training on how to notify leads to forms being
as dengue but who are not notified or are incorrectly and inconsistently completed
incorrectly notified in the national surveil- • Lack of resources during outbreaks mean staff prioritise
lance system patient treatment over completion of notification forms

Table 6. Dengue surveillance completeness questionnaire

Question Answer

What percentage of all primary, secondary and tertiary healthcare Dengue: __%
centres are able to notify the following types of cases to the national Severe dengue: __%
surveillance system?

Are private healthcare centres required to report all dengue cases to YES / NO
the national surveillance system?
If NO, what percentage of all febrile illness cases are treated only in __%
private healthcare facilities?

What is the notification fidelity* for the following types of cases? Dengue: __%
Severe dengue: __%

*Notification fidelity is defined as the percentage of diagnosed dengue cases in a healthcare facility that has access to the notification system that
are correctly notified in the national surveillance system.

14
1.3 OVER-DIAGNOSIS ANALYSIS
Box1
Other febrile illnesses that can be misdiagnosed as den-
As a result of imperfect diagnosis, other febrile illnesses gue
(Box 1) may be misdiagnosed and notified as dengue Depending on the context and phase of illness, dengue may
(over-diagnosis). This is especially common in the mid to commonly be misdiagnosed as:
late phase of dengue outbreaks where dengue diagnosis • other arboviral infections (e.g. Zika virus disease, yellow
is largely dependent on presumptive contextual evidence, fever, chikungunya)
especially when testing facilities are overwhelmed by ex- • measles
cess requests (9). Misdiagnosis is also an issue in the • rubella
context of outbreaks of other arboviral infections such as • adenovirus infections
Zika virus disease and chikungunya which often share the • influenza
same seasonal timing and non-specific clinical symptoms, • typhoid fever
while serological diagnostic tests for Zika virus disease • malaria
and dengue can often cross-react (10). • leptospirosis
The positive predictive value (PPV), or precision, of a den- • viral hepatitis
gue case definition is the proportion of true dengue cases • Rickettsial infections
among those that are notified as dengue (Box 2). A high • bacterial sepsis
PPV (close to 1) indicates low rates of over-diagnosis. Source: reference (11).
PPV can be calculated using existing dengue rapid and
laboratory-based testing data as long as they are repre-
Box2
sentative of all clinically suspected cases. Some samples
tested for dengue may be unrepresentative of all notified
cases because testing is reserved for clinically complex Calculation of sensitivity and positive predictive value of a
dengue case definition
cases, precautionary testing in vulnerable individuals
(e.g. pregnant or the elderly) or may be underutilized in Clinical diagnosis
the later stages of dengue outbreaks due to laboratory Dengue Not dengue
capacity constraints.
Diagnostic Dengue A B
test Not dengue C D

PPV = A/(A+C)

Sensitivity = A/(A+B)

15
 To reduce these biases when calculating PPV, the cases analysis should be selected based on maximizing the per-
included from the laboratory testing dataset should have formance of each test for the number of days between
the same age and weekly distribution as the notified cas- fever onset and sample collection (Table 7).
es dataset (i.e. all cases). This can be achieved by sam- “True” dengue cases for the over-diagnosis analysis
pling selected cases from the laboratory dataset with their should be reserved for cases that satisfy the “highly sug-
probability of being chosen dependent on the proportion gestive” or “confirmed” definition in the WHO 2009
of adult and children samples in each week in the notified guidelines (2).
cases dataset.
The PPV should be calculated separately for adults and
It must also be considered that dengue tests also have children to reflect known differences in the specificity of
variable degrees of sensitivity and specificity (see Labo- dengue diagnosis (12); however, if dengue is primarily a
ratory diagnosis and diagnostic tests (Chapter 4) of the paediatric disease in the area concerned only estimates
WHO 2009 guidelines (2) and samples chosen for the for children should be made to make this study more fea-
sible (Table 9).

16
Table 7. Time window for accurate diagnosis of dengue using different diagnostic tests

Method Optimal time window for detection (days after onset of symp-
toms)

Viral isolation 1–5 days

Nucleic acid detection (PCR) 1–5 days

Antigen detection (NS1) 1–6 days

IgM ELISA or IgM rapid test After 5 days

IgG (paired sera) by ELISA, HI or neutralization test Acute sera 1–5 days; convalescent after 15 days

ELISA, enzyme-linked immunosorbent assay; HI, haemagglutination inhibition; Ig, immunoglobulin; PCR, polymerase chain reaction
Source: reference (6).

Table 8. Criteria for highly suggestive and confirmed dengue infections; note that IgM and IgG serological methods may cross-re-
act with antibodies to other flaviviruses and therefore may be less accurate in regions where multiple flaviviruses co-circulate

Highly suggestive Confirmed

One of the following: One of the following:

1. IgM-positive in a single serum sample 1. PCR-positive
2. IgG-positive in a single serum sample with an HI titre of 1280 or 2. Virus culture-positive
greater 3. IgM seroconversion (four-fold rise in titre) in paired sera
4. IgG seroconversion in paired sera or fourfold IgG titre in-
crease in paired sera

HI, haemagglutination inhibition; Ig, immunoglobulin; PCR, polymerase chain reaction

Source: reference (2).

Table 9. Recording the results of the over-diagnosis analysis of notified dengue cases

Population (age in years) Notified cases that tested posi- Notified cases that tested nega- PPV of case definition (positive
tive dengue tive for dengue samples / total samples tested)

Children (< 19)

Adults

PPV, positive predictive value

17
 1.4 UNDER-DIAGNOSIS ANALYSIS 3. Exclusion criteria for enrolment should include nega-
tive for all other obvious causes of infection (e.g. exist-
ing chronic illnesses, other common febrile illnesses if
While PPV of a case definition quantifies over-diagnosis relevant- malaria).
and over notification of dengue, case definition sensitivity
also needs to be known to quantify under-diagnosis of 4. Samples (blood, serum or plasma depending on rou-
dengue. Dengue, particularly in its more milder disease tine practice and diagnostic method used) should be
forms, is frequently misdiagnosed as other febrile illness- obtained at the relevant disease time-point (see Table
es, especially in areas with high prevalence of other 7) from all enrolled patients and sent for diagnostic
acute febrile illnesses (Box 1). Furthermore, many peo- testing for dengue using the routine testing protocols.
ple with dengue infections do not develop highly specific 5. Clinical diagnosis of enrolled patients should be re-
clinical symptoms, and those that do seek care do not corded with no modification to standard clinical prac-
receive a final diagnosis – undifferentiated febrile illness, tices.
or UFI (13–15). 6. Diagnostic test-negative patients can be discarded
Despite their often mild disease outcomes, UFI dengue from the study.
infections are an important part of the clinical burden of 7. Sensitivity is calculated as the proportion of true den-
dengue because: gue infections that receive a clinical diagnosis of den-
• they are improperly treated, typically with antibiotics gue (Box 2).
or antimalarials resulting in a greater risk of anti-micro- 8. Under-diagnosis (as measured by case definition sen-
bial resistance emergence; and sitivity) should be disaggregated by age of patients
• they frequently overwhelm healthcare infrastructure because these are known confounders in clinical di-
during outbreaks which compromises care for other agnosis (16) (Table 10). However, if dengue is pri-
patients marily a paediatric disease in the country in which
To estimate under-diagnosis of notified clinical dengue burden estimation is being performed, case efforts
requires new data collection under a prospective clinical should focus on maximizing sample size in paediatric
study. This new study requires minimal new resources as populations.
samples can be collected and tested using the already Countries may have already undertaken research to es-
established routine surveillance infrastructure and testing timate the sensitivity of their clinical case definitions for
facilities. dengue. If this has already been performed, this data can
The protocol for such a study should be based on the alternatively be used to fill out the sections in Table 10,
following major steps: however it is important that such studies use the same de-
nominator as suggested above, i.e. all febrile illness pa-
1. Selected healthcare facilities around the country tients without obvious alternative cause not, for example,
should be chosen to take part at different times of the patients with dengue-like illness which may overestimate
year (see Box 3 for area sampling strategies). clinical case definition sensitivity.
2. A clinical cohort of febrile illness patients should be
enrolled at the point of first seeking care (e.g. emer-
gency room, outpatient clinic, triage, etc.).

18
Table 10. Recording the results of the under-diagnosis analysis

Population Test positive dengue cases in Test positive and clinically Sensitivity of dengue case defi-
(age in years) UFI sample diagnosed dengue cases in UFI nition (clinical and test positive
sample / test positive)

Children (< 19)

Adults

UFI, unidentified febrile illness

Box3
in the mid to late stages of an outbreak and to make studies
more practical, this might change to every 5th case one month
Sampling strategy for new clinical and community-based before the typical dengue season time, then transition to one
surveys every 20th case once the dengue outbreak* begins.
For the purpose of this burden estimation exercise these new Who should be sampled?
surveys are intended to generate estimates of under-diagnosis For clinical studies, samples should reflect who seeks care, so
(1.4) and incidence (2.1 and 2.2) that are nationally and there is no need to stratify by any particular criteria. Febrile
annually representative. Further studies can also be planned to illness and seroprevalence cohorts should ideally be representa-
estimate subnational or intra-annual variation in burden. tive of the resident population. However convenience sampling
Where should these studies take place? is often required to make such cohorts feasible and as a result
Ideally to be nationally representative, study sites (healthcare school children or employees of a particularly company are
centres and their catchment populations) should be sampled typically chosen for follow-up. Community-based studies should
with their probability of being chosen proportional to their aim for measuring the incidence of symptomatic and inapparent
catchment population (i.e. more sites in densely populated dengue infection in at least one adult and one child cohort.
areas). However, given the need for collaboration between Sample sizes?
established research projects and existing routine surveillance Required sample size for both clinical and community-based
activities, an element of convenience sampling is advised and studies will be highly dependent on sub-national variance in
integration of studies should take priority over strict geo- transmission intensity and diagnostic standardization making
graphically representative sampling. Countries should aim for exact numbers difficult to generalise to different settings. Past
representation from each major dengue-endemic administrative community-based febrile illness cohorts and serological cohort
subregion (e.g. North, South, East, West, or provinces/states studies have included around 1000 individuals followed for a
if feasible) and include at least some rural or peri-urban sites period of at least one year. For the under-diagnosis analysis,
proportional to the percentage of the national population that countries should aim for between 1000–2000 samples with
lives in each of these areas. further samples required if highly variable sensitivity results are
When should these studies take place? found.
To be annually representative, sampling should occur across Further details
a 12-month period with sample number proportional to when For more details on sampling strategy and sample size calcula-
most dengue infections occur (i.e. more samples in the dengue tion please consult WHO guidelines on dengue serosurveys for
season). This can be achieved in clinical studies by sampling vaccine targeting (17).
every 10th febrile illness patient (depending on sample num-
ber). However, because of under-diagnosis of dengue during * The definition of a dengue outbreak will vary from country-to-country but is
the early phase of an outbreak and over-diagnosis of cases typically when the weekly number of cases exceed two standard deviations
of previous five non-outbreak year’s equivalent weeks case numbers.

19
 1.5 FEBRILE COHORTS
Box4
Gold standard locations for burden estimation
A high proportion of symptomatic dengue infections (es-
Integration of studies
timated at around 70%; range 40–82%) (17) will not
Each of the studies and data analyses recommended in these
guidelines measures unique and non-overlapping segments of
seek formal healthcare but will still develop disease symp-
dengue’s burden. Therefore it is important that different types toms. Estimating this subclinical symptomatic burden is
of studies (1.1–4. and 2.1–2) occur in the same locations and important because affected individuals may still have
ideally at the same times. This will generate “Gold standard” chronic effects, e.g. prolonged joint pain, that may lead
sentinel sites for burden estimation where the burden of dengue to significant productivity losses in a very large number
at all levels of severity from inapparent infection to death is
of individuals. Many of these individuals may also seek
known.
informal healthcare and subsequently have worse or ex-
Maximising existing data
tended disease outcomes.
Community-based surveys are typically costly and usually only
conducted in research settings as opposed to routine public A self-managed dengue infection is defined as an infec-
health activities. However, many febrile illness cohorts and tion of sufficient severity to disrupt the daily routine of the
serological cohort studies have been previously carried out in individual and will likely result in not attending work or
a variety of settings. Countries are encouraged to integrate school. As such events occur outside formal healthcare
burden estimation activities with the past or on-going efforts settings, a community-based fever survey is needed to
to measure community burden. This could include research
projects, but also preparatory or control-arm studies for vaccine
measure their incidence.
or drug trials (18,19).
This will establish a network of gold standard burden estimation FEBRILE ILLNESS SURVEYS
sites in the same location as previous cohort studies. Integration
of clinical and community burden measurement activities has 1. Selected schools (20) and workplaces (if dengue in
the potential to reduce the cost and improve the accuracy of adults is common in the affected country) (21) should
burden estimation. be identified, and pupils and employees recruited into
a fever cohort where each individual’s age, gender
and address are recorded (Table 11).
2. Individual school or work absenteeism should trigger
a home visit to record the reason for absence.
3. If the absence is due to acute febrile illness, blood
samples should be obtained from the individual on the
day of the visit.
4. No attempts should be made to increase or decrease
existing barriers to treatment for the infected individ-
ual, and any treatment seeking should be initiated
by the individuals themselves and not directed by the
study organizers.

20
5. The infected individual should be revisited (at home or 6. All samples should be tested to confirm or reject den-
school/workplace) at least 2 weeks after the first visit. gue infection (see Table 8).
During this visit the subject (or subject’s parents in the 7. The fever cohort study should run for a minimum time
case of schoolchildren) should complete some brief period of one calendar year. Additional individuals
questions on: may need to be recruited into the cohort over time
o Did they seek treatment for this acute illness epi- (ideally, at annual intervals) due to dropout.
sode? Alternatively, fever cohorts can be household based with
o If so, where did they seek treatment (e.g. public or a similar design (20). In these household-based febrile
private hospital, emergency department or outpa- cohorts, households are enrolled at baseline and, as op-
tient clinic)? posed to absenteeism triggering follow-up visits, house-
o If so, when did they seek treatment (e.g. 2 days holds are visited at fixed time intervals (e.g. every 6
after onset of illness)? months) and each participant completes a questionnaire
to retrospectively collect data on fever and treatment seek-
o If so, what was your final diagnosis when you
ing. These types of experimental designs work best when
sought treatment (e.g. dengue or other non-den-
also taking periodic serological samples (see below) as
gue illness)?
this allows confirmation of reported dengue-like illness
If IgM/IgG ELISA methods are being used to con- through monitoring an individual’s seroconversion. These
firm infection, then convalescent samples should approaches may give more representative estimates of
also be obtained from the individual during this burden in adults as workplace-based cohorts may intro-
visit. Additionally, if the study is done in an area duce selection bias for higher socioeconomic status than
with known circulation of other flaviviruses, then se- the general population.
rological ELISA methods should be cross-validated
using more sensitive methods (see (22) for details).

Table 11. Output measures for reporting febrile illness cohort results

Measure (within cohort) Number (child cohort) Number (adult cohort)

Person-years of observation (number in cohort multiplied by time they

were observed for)

Confirmed apparent dengue infections

Confirmed apparent dengue infections that sought treatment*

Incidence of notified dengue cases in the area and during the time of
the cohort study per 100,000 residents

* Should only include treatment seeking to healthcare centres that are part of the national dengue surveillance system.

21
 MEASURING INAPPARENT INCIDENCE WITHIN Many dengue febrile illness cohorts have already been
FEBRILE ILLNESS COHORTS conducted in a variety of areas worldwide, and stake-
Inapparent incidence can also be measured within the holder countries are encouraged to make use of existing
same febrile illness cohorts with minimal additional sam- and ongoing research activities to obtain this information
pling and testing commitments. Measuring the incidence (Box 4). Finally, it should be noted that other methods are
of inapparent infection requires two additional steps to available for measuring apparent dengue infection inci-
the above fever cohort: dence that may give estimates of comparable accuracy.
In particular, index household studies (23) where patients
1. At enrolment, a blood sample is taken from each indi-
are recruited at the clinic level, then members of their
vidual and tested for prior dengue exposure with IgG
immediate family and local neighbours are tested for re-
ELISA.
cent dengue infection, may give comparable estimates of
2. Each subsequent year, every individual in the cohort is incidence, particularly in areas where a high proportion
sampled and tested for IgG and IgM ELISA. of individuals seek care. It should also be noted that in
Further details on fever study implementation can be the absence of dengue-specific data on treatment seek-
obtained from previous published febrile illness cohort ing, treatment seeking rates for fever in children can be
studies (20, 21), particularly with regards to laboratory obtained from Demographic Health Surveys (DHS, www.
test quality control, informed consent, or participants and dhsprogram.com) and national health surveys, which are
maximizing participation rate. frequently conducted in many countries and made pub-
licly available.

Table 12. Serological fever cohort study additional output measures

Measure (within cohort) Number (school-based Number (adult-based cohort)

cohort)

Total primary dengue infections in the cohort (IgG negative at enrol-

ment then IgG positive at annual follow up)

Total post-primary dengue infections in the cohort (IgG positive at en-

rolment then IgG and IgM positive at annual follow up or confirmed
dengue following house visit triggered by school or workplace absen-
teeism)

Total dengue infections

(primary dengue infections plus post-primary dengue infections)

22
1.6 MEASURING TOTAL INFECTIONS methods has its advantages and disadvantages (summa-
USING CROSS-SECTIONAL SEROPREVA- rized in Table 13).
LENCE SURVEYS While seroprevalence surveys have conventionally been
considered a research-based activity, the growing recog-
nition of the importance of the community burden of den-
Due to the high prevalence of inapparent dengue in- gue to wider dengue control means there is a growing
fections, the only reliable method for detecting the total need to incorporate seroprevalence surveys into routine
number of dengue infections is through community-based dengue surveillance.
serological surveys.
Comprehensive guidelines on how to conduct age-strati-
Measuring the total number of dengue infections (includ- fied cross-sectional seroprevalence surveys are contained
ing asymptomatic infections) is important for predicting in Informing dengue vaccination programs: best practices
the impact of interventions such as vaccines and vector for conducting a serosurvey (22), which includes advice
control that aim to prevent people from being infected. on site choice, sample size and standard operating pro-
Determining how many infections need to be prevented cedures. Guidance on serological cohort study design
to avert a clinical case or a case of severe dengue can can be found in previously published sources such as in
be used to set goals and to evaluate new interventions or (24) and (25).
control strategies.
To be useful for burden estimation across all age ranges,
The number of total dengue infections can be measured the seroprevalence survey results must be used to calcu-
through cohort-based longitudinal surveys where serocon- late force of infection (FOI). FOI measures the annual rate
version is directly observed in paired samples from the at which susceptible individuals acquire infection and
same individual (section 1.5), or cross-sectional age-strat- methods for its calculation can be found in section 4.1.4.
ified seroprevalence surveys where accumulation of den- in (22). The extracted FOI estimate can then be used to
gue antibody over time (age) is measured. Each of these calculate infection burden using the burden calculator.

Table 13. Advantages and disadvantages of serological cohort studies and age-stratified seroprevalence surveys

Serological cohort studies Seroprevalence surveys

Advantages Advantages
• Incidence directly observed • More suited for estimating long-term average incidence
• More precise estimate of how incidence varies within the year • Can measure (or predict) incidence across many age groups
• Can monitor serotype-specific incidence (if a subset are • Usually cheaper, quicker and more simple than cohort studies
cross-validated with PRNT [plaque reduction neutralization test])
• Can be an addition to planned or existing febrile illness cohorts Disadvantages
• Incidence indirectly observed (have to make assumptions of
Disadvantages stable incidence and age-independent susceptibility)
• Often limited to narrow age groups • Not serotype specific (unless done with PRNT, which limits sam-
• Expensive, time-consuming and may require specialist equip- ple size and feasibility)
ment if results need to be serotype specific

23
 BURDEN CALCULATION

2.1 USING THE BURDEN CALCULATOR* Raw data should be entered in the green coloured cells,
while orange coloured cells indicate automatically calcu-
lated intermediate or final outputs.
The burden calculator is a spreadsheet-based tool that
combines the information supplied in sections 1.1–6 to While ideally data from all of the studies suggested in
estimate national dengue burden. Each sheet in the bur- sections 1.1–6 should be entered, if some of this informa-
den calculator corresponds to a different type of data. tion is not available, approximations can be made from
within the range of data from previous studies shown in
red coloured cells. It is also important to represent uncer-
tainty in some of these measurements. This is achieved
*Please contact WHO for the excel based burden calculator. in this burden calculator through the use of additional
Email: dengue@who.int columns for multiple experiments or surveys.

Step 1: Enter age and severity stratified notified case data (see Table 3 and section 1.1) in Sheet 1 of the burden
calculator.

D
TA B C G

! 1
1.1. Routine surveillance data

4 Non-severe dengue Severe dengue Fatal dengue

Age Clinical and/or test Clinical and/or test Clinical and/or test
5 confirmed Samples tested Samples positive confirmed confirmed
6 0-1 223 8 2 2 0
7 1-4 3314 123 31 27 7
8 5-9 5976 221 55 48 12
9 10-14 5350 198 49 43 11

10 J15-19 4869 179 45 39 10

11 20-29 12039 446 111 97 24
12 30-39 7041 260 65 57 14
13 40-49 4431 164 41 36 9
14 50-59 2825 105 26 23 5
15 60-69 717 27 7 6 2
16 110+ 717 27 7 6 1
17 Total 47502 1758 439 384 95
18

A C 0 G H
24
1.2. Com�el<?n.e:ss metrics

Suggested Value Value Value Value V�h.1e

Question raOR,e of values (Experiment 1) (Experiment 2) (Experiment 3} {Experiment 41 (Experiment SJ
 12 30-39 7041 260 65 57 14
13 40-49 4431 164 41 36 9
D
TA
14 50-59
B
2825
C
105 26 23 5
G

!
15 60-69
1.1. Routine surveillance data
16 1110+
717
717
27
27
7
7
6
6
2
1
17 Total 47502 1758 439 384 95
4
Step 2: Enter completeness metrics Non-severe
(seedengue
Table 6 Severe
and section 1.2) indengue
Sheet 2 ofFatal
thedengue
burden calculator stratified
18
Age severity.
by disease Clinical and/or test Clinical and/or test Clinical and/or test
5 confirmed Samples tested Samples positive confirmed confirmed
6 0-1 A 223 8 C 2 0 2 0 G H

7 1-4 3314 123 31 27 7

1.2. Com�el<?n.e:ss metrics
8 5-9 5976 221 55 48 12
9 10-14 5350 198 49
Suggested Value 43Value Value 11 Value V�h.1e
raOR,e of values (Experiment 1) (Experiment 2) (Experiment 3} {Experiment 41 (Experiment SJ
10 J15-19 4869
I
179 45 39
Question
10
:.��;:,";���:�:!
11 20-29 :���=12039 ::::.":::n':::. 01 cases No
the national surveilance s.ys1em? 7041
12 1030-39
n-severe dengue S0-100111
446
Severe dengue
260 20-100
65
100 97
10.8 57
100
10.8
240
140
0

0
0
0
Are private healthcare oentres required to cepc>rt an der.gi.ie cases 10 the national
40-49 $ystem? U YES: leave blank,
13 wrvel11ance 4431If NO, what percentage of164
al febfi!e l!lne$$ 10-SO 41 36 9
50-59
14 ca�!;
60-69
15 What
{Ire treated only In pr 2825
ivate hQsoital!.?

IS U,e percenta,ge notification fidel� 717

I
105
Non-severe
27
for the follo'Mng types deno�
80-100
26
7 80
23
6
5
20 0

I
90 0
16 lof cases? 717 27 dengue 90-100 7 6 10
9 110+
Severe
lOC 1()( 0 0
Total
17 Total
10 completeness": Non-severe47502 dengue 1758 439 80 384 90. 95
18
11 Total comoleteness %: Severe den1ue 10.8 10.8
12
13
14
A C 0 G H
• notificaoon fidelity is defined a-s. the percen199e ol diagnosed
del)Que cases In a heal1hcare facility that has access to the
1.2. Com�el<?n.e:ss
notification systemmetrics
lhat are correctly notified in the national
lS surva,iUance syslem,
16 Suggested Value Value Value Value V�h.1e
raOR,e of values (Experiment 1) (Experiment 2) (Experiment 3} {Experiment 41 (Experiment SJ

I
Question

:.��;
If private :,";���:�:!
national surveilance
10 the sector
:���=
healthcareA
s.ys1em? facilities are required
n evere dengue
::::.":::n':::. 01 cases No -s
B dengueto noti-
Severe
S0-100
20-100 C If data10.8
100
D
are unavailable
10.8
100
E section,
for0 any 0
0
G
reasonable
0
0
val-
0

Are private healthcare oentres required to cepc>rt an der.gi.ie cases 10 the national
fy all dengue cases to the national surveillance system, ues can be chosen based on previous surveys. These
I
wrvel11ance $ystem? U YES: leave blank, If NO, what percentage of al febfi!e l!lne$$ 10-SO
leave2 row 1.3. 7Over-diagnosis
ca�!; {Ire treated only In pr
blank. ivate analysis
hQsoital!.?
Non-severe
range of values are shown in the red boxes. If estimates
3
80-100
What IS U,e percenta,ge notification fidel� for the follo'Mng types deno� from these 80 red boxes 0
are used, it is 0recommended that a
I
90 0
If morelof than
cases? one completeness assessment has been per-
9 Suggested range
Severe dengue 90-100
broad range
lOC of values
1()( be 0used as 0opposed0 to a simple
formed
104 Totalorcompleteness":
there areNon-severe signifidenguecant regional differences
of values
in
Diagnostic test positive Diagnostic
80 test
90.negative Total
11 Total comoleteness
completeness, it is%: advised
Severe den1ue to enter the separate results average 10.8to appropriately representPPV uncertainty in this esti-
125 Notified dengue (children)
10.8
0.5-0.95 1758 439 2197 0.8
I Notified den1we
from136separate surveys (adults)in different columns 0.4-0.9 (“Experiment
mate (e.g.
1758
enter 50, 60 and 70 in separate experiment
43g 2197 0.8
14
1”, “Experiment
7 • notificaoon fidelity2”, etc.). columns as opposed to just 60 in one column).
is defined a-s. the percen199e ol diagnosed
8 • see Tables 7 and 8 for suggestions on what appropriate tests to include.
del)Que cases In a heal1hcare facility that has access to the
notification system lhat are correctly notified in the national
lS surva,iUance syslem,
Step 3: Enter
16
A
dengue testing data (see Table 9D and section 1.3) stratifiedG by children
H
and adults.

A
1.4. Under-diacnosis analysis B C D E G
Exoeriment t Experiment 2 Exoeriment 3
2 1.3. Over-diagnosis analysis frue Oe�ue c.tses Tn,e Der«;ue cas.cs T,ue Oen1ue CilSCS
dene,ue c:oireclly diae:nosed deneue co,recttydiaenosed dene,ue c0trec1ly dl,1gnosed
3 Suggc-stc-d t;lf"CC' ol e::iS-C'S inUFI 1n dC'nguc- positivC' e.asc-sinUFI in dC'nguc- positi,'C' e:ISC'S in UFI in dcng!JC' f)OSltrvt
values for Sensitivity Suggested
sam le range UFlsamale Sensititi1itv samole UFI !>amole Sensitivitv samDle UFI samale SemitNitv
4 of values Diagnostic test positive Diagnostic test negative Total PPV
Notified dentue (children] 0.S-095 0.8 1000 700 0.7 1000
Notified dengue (children) 800
1000 900 0.9
5 INot1fted de:nlOC'! (aduftsl 0.◄-0.
9
0.5-0.951000 800
1758
0,8 1000 439 2197
700 0,7
0.8
1000 900 0.9
6 I Notified den1we (adults) 0.4-0.9 1758 43g 2197 0.8
7
8 • see Tables 7 and 8 for suggestions on what appropriate tests to include.
A 8 0

A D G H 25
2.1. Febti1e ill�s c.ohottt.
1.4. Under-diacnosis analysis

Exoeriment t Experiment 2 Exoeriment 3

Measure (within cohort) Number (child cohort) Number (adult cohort) Sueeested ranee of values
 del)Que cases In a heal1hcare facility that has access to the
notification
What system lhatnotification
IS U,e percenta,ge are correctly notified
fidel� forinthe
thefollo'Mng
I Non-severe
nationaltypes deno� 80-100
80 0 0

I
90 0
lS ofsurva,iUance syslem,
l cases? Severe dengue 90-100
916 lOC 1()( 0 0 0
10 Total completeness": Non-severe dengue 80 90.
11 Total comoleteness %: Severe den1ue 10.8 10.8
12
A B C D E G
13
14
If data disaggregated by children and adults are una-
2 del)Que
1.3. cases
Over-diagnosis analysis
• notificaoon fidelity is defined a-s. the percen199e ol diagnosed
In a heal1hcare facility that has access to the
3 notification system lhat are correctly notified in the national
vailable, enter all data in the “children” rows; the same
lS surva,iUance syslem,
values will be automatically assumed and calculated for Suggested range
of values Diagnostic test positive Diagnostic test negative Total
16
4 PPV
adults.
5 Notified dengue (children) 0.5-0.95 1758 439 2197 0.8
6 I Notified den1we (adults)
A 0.4-0.9B C 1758 D 43g E2197 0.8 G
7
Step
8 •4:
seeEnter
Tables 7over-diagnosis
and 8 for suggestions analysis data (see
on what appropriate tests toTable
include.10 and section 1.4) stratified by adults and children.
2 1.3. Over-diagnosis analysis
3
A
Suggested range D G H
4 of values Diagnostic test positive Diagnostic test negative Total PPV
Notified
5 1.4. dengue
Under-diacnosis (children)
analysis 0.5-0.95 1758 439 2197 0.8
6 I Notified den1we (adults) 0.4-0.9 Exoeriment t 1758 Experiment 2 43g 2197 Exoeriment 3
0.8
7 frue Oe�ue c.tses Tn,e Der«;ue cas.cs T,ue Oen1ue CilSCS
dene,ue c:oireclly diae:nosed deneue co,recttydiaenosed dene,ue c0trec1ly dl,1gnosed
8 • see Tables 7 and 8 for Suggc-stc-d
suggestions on what appropriate tests to include.
t;lf"CC' ol e::iS-C'S inUFI 1n dC'nguc- positivC' e.asc-sinUFI in dC'nguc- positi,'C' e:ISC'S in UFI in dcng!JC' f)OSltrvt
values for Sensitivity sam le UFlsamale Sensititi1itv samole UFI !>amole Sensitivitv samDle UFI samale SemitNitv

A (children]
Notified dentue 0.S-095 1000 D 800 0.8 1000 G 700 H 0.7 1000 900 0.9
INot1fted de:nlOC'! (aduftsl 0.◄-0.
9 1000 800 0,8 1000 700 0,7 1000 900 0.9
1.4. Under-diacnosis analysis

Exoeriment t Experiment 2 Exoeriment 3

Enter data for multiple
A
experiments frue in separateOe�ue c.tses
8
column Tn,e Der«;ue cas.cs T,ue Oen1ue CilSCS
0
dene,ue c:oireclly diae:nosed deneue co,recttydiaenosed dene,ue c0trec1ly dl,1gnosed
groups. Suggc-stc-d t;lf"CC' ol e::iS-C'S inUFI 1n dC'nguc- positivC' e.asc-sinUFI in dC'nguc- positi,'C' e:ISC'S in UFI in dcng!JC' f)OSltrvt
values for Sensitivity sam le UFlsamale Sensititi1itv samole UFI !>amole Sensitivitv samDle UFI samale SemitNitv
2.1. Febti1e ill�s c.ohottt.
Notified dentue (children] 0.S-095 1000 800 0.8 1000 700 0.7 1000 900 0.9
INot1fted de:nlOC'! (aduftsl 0.◄-0.9 1000 800 0,8 1000 700 0,7 1000 900 0.9
Step 5: Enter fever cohort data (see tables 11–12, section 1.5) stratified by children and adults.
6
Measure (within cohort) Number (child cohort) Number (adult cohort) Sueeested ranee of values

Person-years of observaUon (Nurri>er in cohort

800 0
• years eacri were oosmeo for)
8 Confirmed apparent denoue A infections 8 27 0 person years
0 2·30 per 1000 cohort
Con11-med apparent dengue WecUons that
9 sounht 1rea�men1 22 O 0·15 per 1000 cohort person years
Incidence
2.1. Febti1eof notlied
ill�s dengue cases in the area of
c.ohottt.
10 the cohort sltrlv Mr 100 000 ,esldents 230 0 10-1000 oer 100,000 residents
11 Apparent dengue caaea per notified case 14.67391304 14.67391304
Proportion of apparent
Measure (withineases that seek
cohort) Number (child cohort) Number (adult cohort) Sueeested ranee of values
12 0.814814815 0.814814815
6 t111atment
13 Person-years of observaUon (Nurri>er in cohort
14 •Serologlcal cohort study calculator 800 0
years eacri were oosmeo for)
Total dengue
8 Confirmed infections
apparent (pl'imaty
denoue and
infections 27 0 2·30 per 1000 cohort person years
'
secon$ry, asvmptomatic
lS Con11-med apparent dengue WecUons and symptomatic:}
that •d 0 10-300 per 1000 cohort per$on ','e'1rs
916 sounht
Oercue 1rea�men1
infections per notified case 22
21.73913043 O 0·15 per 1000 cohort person years
21.73913043
17 Incidence of notlied dengue cases in the area of
10 the cohort sltrlv Mr 100 000 ,esldents 230 0 10-1000 oer 100,000 residents
11 Apparent dengue caaea per notified case 14.67391304 14.67391304
Proportion of apparent eases that seek
12 t111atment 0.814814815 0.814814815
13
14 Serologlcal cohort study calculator
Total dengue infections (pl'imaty and
'
lS secon$ry, asvmptomatic and symptomatic:} •d 0 10-300 per 1000 cohort per$on ','e'1rs
16 Oercue infections per notified case 21.73913043 21.73913043
17

If serological measures are part of the febrile cohort, also

add seroconversion results.

26
 Step 6: Enter seroprevalence and corresponding notified case data (see section 1.6).

Also enter national population age distribution from cen-

sus data in the lower table.

27
 Step 7: Examine final burden estimates in the final sheet of the burden calculator (all automatically calculated).

Simple estimates of the upper and lower bounds of the If estimated values are used for any section, it is recom-
final estimated burden are provided based on the upper mended that users conduct a series of experiments with
and lower ranges of results of the different experiments different values for parameters for which they are uncer-
or surveys. tain and assess what affect these changes have on both
the median burden estimate but also the upper and lower
bounds of the prediction.

28
2.2 COMPARISON WITH GLOBAL DENGUE estimating different levels of severity of dengue burden.
BURDEN MODELLING ESTIMATES The burden estimates made using the burden calculator
can be compared to each of these modelled burden esti-
mates by selecting the relevant country in the brown “Se-
Model-based burden estimates use data from many differ-
lect country” box.
ent contexts in combination with epidemiological theory
and risk factors for the disease to produce estimates of dif- Some of these burden estimation methods can also be
ferent types of dengue burden (Table 14). Each of these used to provide improved parameter estimates for sec-
different modelling approaches has its own strengths and tions in the burden calculator for which there are no coun-
weaknesses, and each approach is more or less suited to try-specific data (26).

Table 14. Summary of the different independent model-based burden global dengue burden methods

Institutional home: Oxford Global Burden of Disease Brandeis

Main data sources Serological cohort studies Databases of all-cause Fever cohort studies and
and environmental risk maps mortality and dengue risk reported case numbers in
factors selected countries

Further details (reference): Bhatt et al., 2013 (17) www.healthdata.org Shepard et al., 2016 (28)
and Stanaway et al., 2016
(27)

Estimates

Year for which burden estimates 2012 2016 2013

were made

Deaths – Yes (37,737) Yes (13 586)

Apparent Clinical Yes (96 m) Yes (101 m) Yes (37.4 m)

infections
Non-clinical Yes (21 m)

Deaths Yes (13 586)

Inapparent infections Yes (294 m) – –

29
 FURTHER EXTENSION STUDIES

3.1 SPATIAL VARIATION IN BURDEN The burden estimation approach detailed in this toolkit
should then be repeated within each of these risk strata to
give separate estimates of dengue burden in high, medi-
Transmission of dengue is highly heterogeneous and na- um and low-risk states.
tional burden numbers are unlikely to reflect subnational
To obtain burden estimates at finer spatial scales (e.g.
variations. Generating more fine-scale predictions of den-
municipality or neighbourhood level), it is not practical
gue burden is important for developing targeted strate-
to conduct the full programme of recommended burden
gies that can improve the way limited surveillance and
estimation activities in every spatial unit. Instead, a re-
control efforts are deployed.
gression approach is more suitable where administrative
The required degree of spatial disaggregation in burden regions within the country are stratified into high, medi-
will depend on each country’s dengue epidemiology um, or low transmission intensity categories based on his-
and capacity to adopt different surveillance and control torical dengue incidence and environmental or socioeco-
strategies in different areas. These are most frequently nomic characteristics. The full range of burden estimation
constrained to different levels of administrative divisions activities is then carried out in at least one of the high,
(Admin1/state/province, Admin2/municipality/county, medium and low administrative regions, with the results
Admin 3/neighbourhood). extrapolated to other regions in the same category. This
Once the desired scale has been chosen, the first step method is explained in more detail in the WHO guid-
should be to stratify each unit into a suspected risk catego- ance on seroprevalence survey design (22). A number of
ry based on available epidemiological and entomolog- potentially explanatory covariates that should also be col-
ical data. For example, if estimating burden at the state lected from all units may explain variation in overall den-
level, states may be categorized into high, medium or gue incidence or in the way it is diagnosed and reported
low suspected risk based on total notified case incidence (Table 15). These covariates can typically be extracted
estimates e.g. > 1000, 100–1000 or < 100 notified from national census records or meteorological datasets.
cases per 100 000 residents per year. The exact thresh- Separate linear regression models should be created for
olds chosen should balance the need for an even number each of the measured burden parameters measured in
of units in each risk category with other epidemiological sections 1.2–4 and 2.1–2. Each of these models should
considerations, such as separating highly seasonal trans- be fitted with all combinations and permutations of ex-
mission units from those with year-round transmission.

30
planatory covariates to create a list of candidate models. For more detail on stratification, sample size estimation
The model with the lowest Akaike Information Criterion and the regression approach refer to the WHO guidance
should be selected as the final model. This final model on seroprevalence survey design (22). More fine-scale
can then be used to make predictions in all units where a mapping and burden estimation or assistance with this
burden assessment did not take place. regression approach may require the assistance of a ge-
ospatial modeller.

Table 15. Examples of potentially explanatory covariates for subnational dengue burden estimation

Covariates for dengue transmission Covariates for surveillance system quality

Number of notified cases % of notified cases that are test confirmed

Rainfall Measures of individual income

Temperature Healthcare budget

Urbanization Doctors per person

3.2 ECONOMIC BURDEN OF DENGUE The direct and indirect cost of each dengue illness episode
will vary depending on severity of illness. For the purpose
of economic burden analysis, it may be more beneficial
Estimation of dengue burden provides an opportunity to to re-categorize non-fatal dengue cases (non-severe den-
also consider the economic burden of the disease, which gue and severe dengue) by treatment setting (inpatient
can be used to make the case for new investments in its or outpatient) as it is a principle determinant of direct
surveillance and control. medical costs.
The types of costs most relevant to the impact of dengue Methods for gathering cost data can be found by referring
can be subdivided into direct and indirect costs. Direct to previous studies (29). Once the cost per illness episode
costs include costs directly related to the treatment of an is derived, the total economic burden can be obtained by
acute dengue illness episode and can include medical multiplying the cost per illness episode with the number of
costs (e.g. medical care, diagnostic tests and medicines illness episodes at each level of severity (see Table 16).
paid for by the public healthcare sector or the patient) Further disaggregation or detail of economic burden may
and non-medical costs (e.g. cost of travel to seek treat- require collaboration with a health economist.
ment). Indirect costs represent losses of productivity (typi-
cally work wages) due to illness or premature death.

31
 Table 16. Calculating the national economic burden of dengue

Type of case Cost per illness episode Number of illness episodes Total cost
(from burden estimation)

Hospital (direct costs)

Hospital (indirect costs)

Outpatient (direct costs)

Short-term costs
Outpatient (indirect cost)

Outside medical sector

(indirect)

Fatal child (indirect)

Long-term costs
Fatal adult (indirect)

Total

3.3 COMBINED ARBOVIRAL BURDEN Health Organization has published guidelines for differ-
entiating dengue, chikungunya and Zika using clinical di-
agnosis (section 5) and laboratory testing (section 7) (30).
Estimating the burden of dengue presents an opportunity
to also estimate the burden of other arboviral infections, We recommend using these clinical and laboratory test-
such as Zika virus disease, chikungunya and yellow fever. ing algorithms when performing the over and under di-
Indeed, co-estimating the burden of Zika virus disease agnosis analyses (1.3 and 1.4) and febrile cohort (1.5)
and chikungunya at the same time as that for dengue may studies in areas of high prevalence of dengue, Zika and
improve dengue burden estimates due to the additional chikungunya.
insight into misdiagnosis among arboviral diseases. This If Zika virus disease is co-circulating, the inclusion crite-
will be most relevant for countries that have experienced ria for arbovirus testing in the UFI study (under-diagnosis
large-scale outbreaks of these diseases and have contin- analysis, 1.4) and febrile cohort (1.5) should also include
ued circulation of more than one arbovirus. rash, even if the participants still attend school or work.
There is a growing recognition that clinical guidelines for Self-reporting of rash to study organizers should be en-
diagnosis and management of arboviral infection need couraged in all cohort participants.
to change when dengue, chikungunya and Zika virus dis- The remaining stages of the burden toolkit can be com-
ease co-circulate due to the high potential of misdiagno- pleted as normal, but with separate entries for each ar-
sis and hence improper management. The Pan American bovirus.

32
3.4 REDUCING UNCERTAINTY IN BURDEN 2. Method uncertainty, where two different methods for
ESTIMATES calculating the overall burden of a particular level of
dengue severity give different answers
e.g. calculating apparent dengue incidence using (i) ap-
The value of burden estimation is not just to generate a
parent cases per notified case times total notified cases or
single estimate of a country’s burden but rather to calcu-
(ii) apparent cases per cases that sought treatment times
late the uncertainty around this figure, identify the source
total estimated clinical burden and the two methods give
of this uncertainty and target future data collection activ-
different answers
ities to improve the accuracy of future burden estimates.
In the burden calculator tool, the clinical burden of den-
Uncertainty in burden estimates from the burden calcula-
gue uses only one method, so is only affected by pa-
tor can come from two sources:
rameter uncertainty. By default, a pie chart of sources of
1. Parameter uncertainty, where two data sources of the parameter uncertainty is produced on the FINAL BURDEN
same type suggest different values ESTIMATION tab:
e.g. two different reporting fidelity assessments suggest
two different values for reporting fidelity

33
 This pie chart can be used to prioritize future data col- der-diagnoses analyses in the same catchment area as
lection efforts to reduce uncertainty in the clinical burden the febrile cohort to see if routine surveillance is more or
of dengue. In this example from Sri Lanka, the greatest less complete in that area.
reduction in uncertainty would be gained by conduct- The total number of dengue infections (which is also used
ing further fidelity assessments, particularly into the no- to estimate the number of asymptomatic infections) can
tification fidelity of non-severe dengue cases. This may also either be estimated by serological febrile cohorts or
include stratifying such assessments across urban versus seroprevalence surveys. The strengths and weaknesses
rural healthcare provision environments to provide more of each approach are outlined in Table 13; however,
locally-specific measures of notification fidelity. both add value to burden estimation. If serological cohort
Reducing estimates of the community burden of dengue studies were performed in years with abnormally high
is also important as it accounts for a larger proportion of or low dengue incidence (as determined by comparison
total burden than clinical burden and is often measured with routine surveillance data), this may explain why ob-
using more limited data. As well as the parameter uncer- served incidence is higher or lower than estimated by
tainty in each of the measurements extracted from febrile seroprevalence surveys that estimate long-term average
cohorts and seroprevalence surveys, there is additional incidence. The limitations of calculating force of infection
method uncertainty around how different data types are from seroprevalence surveys should also be considered,
combined. and seroprevalence surveys that suggest implausibly high
Estimates of the number of self-managed apparent den- or low infection rates relative to notified cases should be
gue cases can be obtained by directly measuring the in- examined.
cidence of self-managed dengue in the febrile cohort, or
by subtracting the estimated clinical incidence calculated
using the burden calculator from the incidence of all ap-
parent infection in the cohort. The former method has the 3.5 EVALUATING CHANGE IN BURDEN
advantage of direct measurement of self-managed den- OVER TIME
gue, but may not be generalizable to the whole country
and treatment seeking behaviour may be very different One of the ultimate aims of wide-scale burden estima-
in the area of the febrile cohort than in other parts of tion is to quantify the impact of interventions on disease
the country. The latter method is more generalizable, but morbidity and mortality at the national and international
assumes that all gaps in routine passive surveillance have scales. Measuring such effects is often complicated by
been correctly enumerated in the burden calculator. parallel improvements in disease surveillance that lead
If these two methods produce conflicting results, the dif- to more cases being reported. Measuring changes in
ference may be explained by varying treatment-seeking surveillance systems is, therefore, important in assessing
rates. Demographic and Health Surveys (DHS) routine- burden change over time (Table 17). Such changes can
ly collect information on treatment seeking for fever and include, but are not limited to, changes in dengue testing
these should be examined to test the hypothesis of differ- procedures and methods, physician training for arbovi-
ent treatment rates in the area of the febrile cohort. If this rus diagnosis and greater provision of government subsi-
shows no difference, countries should attempt to conduct dized healthcare. Measuring these changes will require
completeness analyses, over-diagnosis analyses and un- re-applying the surveys detailed in this dengue burden

34
estimation toolkit (1.1–4. and 2.1–2) at periodic time in- sis analysis should be re-performed if other major causes
tervals. Depending on the surveillance system, however, of febrile illness increase, e.g. Zika virus disease.
some of these surveys may be required to be completed In the absence of any major nationwide intervention
more or less frequently to achieve reliable dengue burden programmes or invasion of new dengue serotypes, the
estimates. burden of DENV infection in the community is unlikely
The most common increases in disease surveillance efforts to change significantly. As a result, febrile illness cohorts
are to (i) increase the completeness of routine surveillance and seroprevalence survey measures can be updated at
systems (1.2) and (ii) increase the availability of dengue less frequent intervals in many settings. Treatment-seeking
testing to support clinical diagnosis, which decreases behaviours (which are measured as part of the febrile
over-diagnosis (1.3). If surveillance system completeness illness cohorts) may change at more frequent intervals,
is already high and a high proportion of notified cas- particularly if major new sources of care become avail-
es are already tested using appropriate dengue-specific able, such as new private hospitals, or longer-term so-
diagnostic tests, updating the completeness assessment cioeconomic changes occur that increase utilization of
and over-diagnosis analysis are probably unnecessary; healthcare services.
however, updating will be required if the national case Follow-up seroprevalence surveys are likely to require a
definition for dengue changes. different experimental design and different data analysis
The over-diagnosis analysis may also require reassess- techniques to inform changes in burden over time. Young-
ment at semi-frequent intervals, particularly if there are big er children must be enrolled in follow up seroprevalence
increases in treatment or point-of-care diagnostic capac- surveys as observing changes in the past 5 years, for ex-
ities that allow greater detection of early-stage clinically ample, requires enrolling children aged below 5 years.
non-specific dengue cases. Furthermore, the over-diagno- Data must also be analysed using modelling methods that
incorporate time-varying force of infection estimates.

Table 17. Frequency of burden estimation activity repeat measures to monitor changing dengue burden over time

Activity Effort Frequency Considerations

1.1 Data assembly Low Annually Changes in case definition will require repeating all of sections 1.2–4

1.2 Completeness assessment Low Annually Unless already high

1.3 Over-diagnosis analysis Low Annually Unless a high proportion of notified cases are already test-confirmed

1.4 Under-diagnosis analysis Medium Semi-frequently May be required more frequently if new causes of febrile illness
(~2 years) emerge, e.g. Zika virus disease

1.5 Febrile illness cohorts High Infrequently Surveys of treatment-seeking behaviour may require more frequent
(~5 years) updates

1.6 Seroprevalence surveys High Infrequently Survey protocol and analysis requires modification to detect changes
(~5 years)

35
 REFERENCES

1. Global strategy for dengue prevention and con- 8 . Dengue haemorrhagic fever: diagnosis, treatment, prevention
trol, 2012–2020. Geneva: World Health Organ- and control, 2nd edition. Geneva: World Health Organiza-
ization; 2012 (http://apps.who.int/iris/bitstre tion; 1997 (http://www.who.int/csr/resources/publica-
am/10665/75303/1/9789241504034_eng.pdf). tions/dengue/itoviii.pdf).

2. Dengue guidelines for diagnosis, treatment, prevention 9. Huy R, Buchy P, Conan A, Ngan C, Ong S, Ali R, et al.
and control, 3rd edition. Geneva: World Health Or- National dengue surveillance in Cambodia 1980–2008: ep-
ganization; 2009 (http://whqlibdoc.who.int/publica- idemiological and virological trends and the impact of vector
tions/2009/9789241547871_eng.pdf). control. Bull World Health Organ. 2010;88(9):650–7.

3 . Technical handbook for dengue surveillance, dengue out- 10 . Waggoner JJ, Pinsky BA. Zika virus: diagnostics for an emerg-
break prediction/detection and outbreak response. Gene- ing pandemic threat. J Clin Microbiol. 2016;54(4):860–7.
va: World Health Organization; 2016 (http://www.who. 11 . Simmons C, Farrar J, Chau N, Wills B. Dengue. N Engl J
int/tdr/publications/year/2016/tech_handbook_dengue/ Med. 2012;366(15):1423–32.
en/). 12. Thai KT, Nishiura H, Hoang PL, Tran NTT, Phan GT, Le HQ, et
4. Olkowski S, Stoddard ST, Halsey ES, Morrisson AC, Barker al. Age-specificity of clinical dengue during primary and sec-
CM, Scott TW. Sentinel versus passive surveillance for meas- ondary infections. PLoS Negl Trop Dis. 2011;5(6):e1180.
uring changes in dengue incidence: evidence from three con- 13 . Suttinont C, Losuwanaluk K, Niwatayakul K, Hoontrakul S,
current surveillance systems in Iquitos, Peru. bioRxiv. 2016. Intaranongpai W, Silpasakorn S, et al. Causes of acute,
doi: http://dx.doi.org/10.1101/040220. undifferentiated, febrile illness in rural Thailand: results of a
5. Stevens GA, Alkema L, Black RE, Boerma JT, Collins GS, prospective observational study. Ann Trop Med Parasitol.
Ezzati M, et al. Guidelines for accurate and transparent 2006;100(4):363–70.
health estimates reporting: the GATHER statement. PLoS Med. 14 . Manock SR, Jacobsen KH, de Bravo NB, Russell KL, Negrete
2016;13(6):e1002056. M, Olson JG, et al. Etiology of Acute Undifferentiated Febrile
6 . Barniol J, Gaczkowski R, Barbato EV, da Cunha RV, Salgado Illness in the Amazon Basin of Ecuador. Am J Trop Med Hyg.
D, Martinez E, et al. Usefulness and applicability of the re- 2009;81(1):146–51.
vised dengue case classification by disease: multi-centre study 15 . Abhilash KPP, Jeevan JA, Mitra S, Paul N, Murugan TP, Ran-
in 18 countries. BMC Infect Dis. 2011;11:106. garaj A, et al. Acute undifferentiated febrile illness in patients
7. Horstick O, Jaenisch T, Martinez E, Kroeger A, See LLC, Far- presenting to a Tertiary Care Hospital in South India: clinical
rar J, et al. Comparing the usefulness of the 1997 and 2009 spectrum and outcome. J Glob Infect Dis. 2016;8(4):147.
WHO dengue case classification: a systematic literature re- 16 . Burattini MN, Lopez LF, Coutinho FA, Siqueira-Jr JB, Homsani
view. Am J Trop Med Hyg. 2014;91(3):621–34. S, Sarti E, et al. Age and regional differences in clinical
presentation and risk of hospitalization for dengue in Brazil,
2000-2014. Clinics. 2016;71(8):455–63.

36
17 . Bhatt S, Gething PW, Brady OJ, Messina JP, Farlow AW, 26. Undurraga EA, Halasa YA, Shepard DS. Use of expansion
Moyes CL, et al. The global distribution and burden of den- factors to estimate the burden of dengue in Southeast Asia: a
gue. Nature. 2013;496(7446):504. systematic analysis. PLoS Negl Trop Dis. 2013;7(2):e2056.
18 . Capeding MR, Tran NH, Hadinegoro SRS, Ismail HIHM, 27 . Stanaway JD, Shepard DS, Undurraga EA, Halasa YA, Cof-
Chotpitayasunondh T, Chua MN, et al. Clinical efficacy and feng LE, Brady OJ, et al. The global burden of dengue: an
safety of a novel tetravalent dengue vaccine in healthy chil- analysis from the Global Burden of Disease Study 2013. Lan-
dren in Asia: a phase 3, randomised, observer-masked, pla- cet Infect Dis. 2016.
cebo-controlled trial. Lancet. 2014;384(9951):1358–65. 28. Shepard DS, Undurraga EA, Halasa YA, Stanaway JD. The
19. Villar L, Dayan GH, Arredondo-García JL, Rivera DM, Cunha global economic burden of dengue: a systematic analysis.
R, Deseda C, et al. Efficacy of a tetravalent dengue vaccine in Lancet Infect Dis. 2016;16(8):935-41.
children in Latin America. N Engl J Med. 2015;372(2):113– 29. Suaya JA, Shepard DS, Siqueira JB, Martelli CT, Lum LC,
23. Tan LH, et al. Cost of dengue cases in eight countries in the
20. Morrison AC, Minnick SL, Rocha C, Forshey BM, Stoddard Americas and Asia: a prospective study. Am J Trop Med Hyg.
ST, Getis A, et al. Epidemiology of dengue virus in Iquitos, 2009;80(5):846–55.
Peru 1999 to 2005: interepidemic and epidemic patterns of 30. Tool for the diagnosis and care of patients with suspected
transmission. PLoS Negl Trop Dis. 2010;4(5):e670. arboviral diseases. Washington. D.C.: Pan American Health
21. Porter KR, Beckett CG, Kosasih H, Tan RI, Alisjahbana B, Organization; 2017 (http://iris.paho.org/xmlui/han-
Rudiman PIF, et al. Epidemiology of dengue and dengue hem- dle/123456789/33895).
orrhagic fever in a cohort of adults living in Bandung, west
Java, Indonesia. Am J Trop Med Hyg. 2005;72(1):60–6.
22. Informing dengue vaccination programs: best practices for
conducting a serosurvey. Geneva: World Health Organiza-
tion; 2017.
23 . Vong S, Khieu V, Glass O, Ly S, Duong V, Huy R, et al. Den-
gue incidence in urban and rural Cambodia: results from pop-
ulation-based active fever surveillance, 2006–2008. PLoS
Negl Trop Dis. 2010;4(11):e903.
24 . Balmaseda A, Hammond SN, Tellez Y, Imhoff L, Rodriguez Y,
Saborío SI, et al. High seroprevalence of antibodies against
dengue virus in a prospective study of schoolchildren in Mana-
gua, Nicaragua. Trop Med Int Health. 2006;11(6):935–42.
25. Tuntaprasart W, Barbazan P, Nitatpattana N, Rongsriyam Y,
Yoksan S, Gonzalez JP. Seroepidemiological survey among
schoolchildren during the 2000-2001 dengue outbreak of
Ratchaburi Province, Thailand. Southeast Asian J Trop Med
Public Health. 2003;34(3):564–8.

37
Dengue is a viral disease vectored by Aedes
mosquitoes that has spread throughout the
tropical world since the mid twentieth century.
Existing reactive control efforts have failed to
stop the expansion of dengue virus transmission
and many areas now have endemic circulation
of all four dengue virus (DENV) serotypes. New
strategies are needed to reverse this trend; and
to be effective, they must be based on accurate
quantitative information about the burden of
dengue.
The aim of this toolkit is to estimate the national
annual burden of dengue when applied in
a given country or subnational area. This is
achieved through a series of six sub-objectives
that: assemble existing data, amend gaps in
surveillance completenes, and correct for both
over and under diagnosis to estimate the true
clinical burden of dengue. These clinical burden
estimates should then be combined with new or
existing community-based surveys to estimate the
symptomatic and inapparent community-based
burden of dengue.