Modeling Disease Surveillance and Assessing Its
Effectiveness for Detection of Acute Respiratory
Outbreaks in Resource-Limited Settings
L. Ramac-Thomas1, T. Philip1, H. Burkom1, J. Coberly1,
S. Happel Lewis1, J.P. Chretien2
1The
Johns Hopkins University Applied Physics Laboratory
2Walter Reed Army Institute of Research, Global Emerging Infections System
7th Annual
Ann al Conference of the
International Society for Disease Surveillance
Track 1: Surveillance Innovations (2) – Evaluation of Detection Approaches
Raleigh NC
Raleigh,
December 4
4, 2008
This presentation was supported by funding from DoD Global Emerging Infections System (GEIS).
Outline
• Objectives
• Background
• Approach
A
h and
dM
Methods
h d
• Validation and Study Design
• Results
• Conclusions
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
2
Acute Respiratory Infection
Epidemic
p
Simulator (ARIES)
(
)
Develop model to
–
–
Evaluate acute respiratory illness surveillance in resourceresource
limited settings
Measure potential benefit of policy decisions and
countermeasures
Required features
–
–
–
–
Focus on early outbreak stages
Restrict demographic modeling to features relevant to disease
spread
Include knowledge of existing surveillance capability
Portability
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
3
Background
A U.S. Department of Defense program is underway to assess health
surveillance in resource-poor settings and to evaluate the Early
Warning Outbreak Reporting System (EWORS)
This program has included
several informationgathering trips, including a
trip to Lao
PDR in September 2006
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
4
Challenges of Surveillance
in Resource-Poor Regions
Healthcare access is limited byy
– Available transportation
– Lack of trained care providers or insurance/ability to pay for care
– Preference
f
for
f herbal, spiritual healers
– Lack of modern communication technology
Magnified Threat of Major Epidemics
– Infectious disease outbreaks not uncommon
– Many workers at human-animal interface
– Vaccine, antiviral supplies scarce if available
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
5
Application to EWORS Systems and Lao PDR
EWORS Systems
– 35 sites in 4 countries in SE Asia and in 2 sites in Peru
– Daily transfer of patient records from network hospitals to national
hub
– Popular
p
at each installation
Initial Application in Lao PDR
– IInvestigative
ti ti trip
t i by
b 6-member
6
b EWORS International
I t
ti
l Working
W ki
Group
– Interviews at hospitals, National Center for Laboratory and
Epidemiolog Ministr
Epidemiology,
Ministry of Health
Health, and other go
government
ernment agencies
– Collection of healthcare-seeking behavior information
– Discussions of both theoretical and actual surveillance practices
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
6
Key Features of ARIES
• Model is individual-based,
individual based but includes only infected and
exposed
• Attention limited to the stages of the event before
population behavior radically changes
• Assumptions of near-instantaneous detection in published
research are unrealistic, especially in resource-limited
settings
g
• Goal is to implement realistic surveillance modeling
Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
7
Components of ARIES
Demographic model generates features relevant to outbreak
spread.
Disease model simulates progression of disease in infected
agents.
Travel model simulates agent travel patterns to mimic
geographic spread.
y in detection, data entry,
y
Surveillance model simulates delays
data transmission, and epidemiologic investigation.
Information basis
census data, population survey reports, site-visit interviews,
acquired data from EWORS, area geography, disease model
Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
8
Household Model
Accurately estimate household makeup of specified province
in Lao
Considerations
–
–
Household size
Age group and sex
Preserve census
dependency ratios
Sex of household head for
comparison to census
distributions
P
Pregnancy
status
t t
1600
1400
Number of Ho
ouseholds
–
–
–
sim total
sim urban
sim rural w/roads
sim rural w/o roads
actual total
actual urban
actual rural w/roads
actual rural w/o roads
1800
1200
1000
800
600
400
200
0
1
2
3
4
5
6
7
8
9
Members per Household
10
11
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
9
Disease Model
Disease stages model [Feighner]
– Stage
S
lengths
l
h
– Probability of complications
– Percentage of asymptomatic
Disease transmission model [Glass]
[
]
– Susceptibility and infectivity are functions of disease stage and
age
– Transmission is also dependent on type of contact (household,
peer, random)
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
10
Disease Stage Model
STAGE 1:
Infected, NonSymptomatic,
Non-Contagious
Non
Contagious
STAGE 2:
Infected, NonSymptomatic,
Contagious
Stage 2
> 2 days?
YES
NO
STAGE 3a:
Infected,
Symptomatic
Symptomatic,
Contagious
Person Dies?
Recovered
NO
YES
YES
Complications?
NO
Dead
STAGE 3b: Infected,
Symptomatic,
Contagious
w/Complications
YES
Person Dies?
NO
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
11
Travel Model
Establish location of an infected individual throughout
course of infection
C
Considerations
id ti
–
–
–
–
–
Subdivide province into rural and urban districts
Update agent locations on time scale of days
Movement is district-to-district
Occupation and age influence agent itinerary
For each new agent location,
location recompute probabilities of travel to
other districts
– Multi-day trips are allowed
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
12
Surveillance Model
• To include surveillance in a disease model need to consider both
surveillance lag and the effect of surveillance system on surveillance
lag
• Model both traditional surveillance and EWORS
• Investigate advantage of proposed EWORS expansion
– Additional provincial EWORS hospitals
– EWORS systems in chosen district hospitals
– Other interventions
• EWORS hospital currently in Luang Prabang District
• Simulations will also be run with an additional EWORS system at Nam
Bak
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
13
Surveillance Model
Time for Patients to
Seek Care
Apx.
1-14 days
Onset of Illness
Time for patient to seek care
depends on:
1. Health beliefs & attitudes
2. Income
3. Availability of health care
& transportation to it
4. Relationship with HCP
5. Severity of illness
6. Status of patient within
household
Effect of
Surveillance
System
Surveillance
Lag
Time for Dr to ID
O tbreak
Outbreak
Up to
1 week
Without Rapid Test
1-3 days
Time
for PH response
Effect of
Surveillance
systemdepends
dependson:
on:
1.
How
soon
investigation
is initiated
1. Design of the system
2.pHow
until LPH acts or contacts RPH
a. Paper
or long
electronic
3.
How
long
until
RPH or
investigates
b. Data pre-computerized
specially entered
4.
How
long
until
RPH
actslag
or contacts NPH
c. Data source & its reporting
5.
How
long
until
NPH
investigates
d. System validity (IT & Epi)
6. How long
NPH
acts
e. Integration
of until
system
into
surveillance
f. Nodal differences
2. PH Belief in system
3 PH U
3.
Use off system
t
Flu A
0-1 days H5N1
Components of Surveillance System
Outbreak identification depends on:
1. Time from visit to entry into system
1. Availability of rapid tests
2. Time from entry to analysis
2. No. (+) tests
3. Time from analysis to review of results
3. No. patients with complications
4. Time from review to investigation
4. No. &/or rapidity of deaths
5 Time from investigation to action
5.
5. Age patients with
6. Other, e.g. time variation by site
complications/death
0-1
0-1
days days
0-2
days
1-3
days
Time for Public Health Response to Start
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
14
Target vs. Simulated Values
•
•
•
•
Serial Interval: 2.6 days (for an effective reproductive rate of about 1.6)
Household Attack Rate: 0.25
Fraction of symptomatic cases: 0.67
Overall case fatality rate: 6%
Histogram of Serial Intervals
18
Histogram of Household Attack Rates
16
30
Histogram of Symptomatic Ratios
CFR Histogram
25
2.45
More
2.44
2.43
10
0.29 0.295 0.3
More
5
Household
0.63
0.635 0.64Attack
0.645 Rate
0.65 0.655 0.66 0.665 0.67 0.675 0.68 0.685 0.69 More
0
M
or
7.
5
7
6.
5
6
5.
5
Ratio of Symptomatic Cases
e
0.285
11
0.28
.5
0.275
10
0
10
0.27
15
9.
5
0.265
20
9
0.26
5
25
8.
5
Serial0 Interval (days)
2.42
2.4
2.39
2.38
2.37
10
8
F re
quency
2.41
10
5
30
15
2.36
2.3
2.31
2.29
2.28
2.27
2.26
0
15
2.35
2
20
20
2.34
4
Frequency
y
6
2.33
8
2.32
10
Fre
equency
12
2.25
Fre
equency
14
Case Fatality
y Rate ((%))
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
15
Study Design
Three scenarios
• No EWORS system
• EWORS in Luang
Prabang
• EWORS in Luang
Prabang and Nam Bak
Each scenario - 5 seed
cases in single province
• Luang Prabang
• Chomphet
• Nam Bak
• Vieng Kham
100 runs for each
province
scenario/seed p
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
16
Epidemic Curve Variability
• Curves show number of new symptomatic cases by day for one of the
simulated epidemics.
• Runs
R
start
t t with
ith 5 seed
d cases iin L
Luang P
Prabang
b
tto iincrease lik
likelihood
lih d
of spread.
Nr. Ne
ew Symptomattic Cases
Epidemic Curve Early Variability
1200
1000
800
600
400
200
0
4
5
6
7
8
9
10
11
12
13
14
Outbreak Day
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
17
Rate of Epidemic Spread
D is tr i b u t io n o f D a y s U n til a n In fe c tio n o u t o f D is tr i c t
50
S e e d D is t ric t
45
L u a n g P ra b an g
Cho m phe t
N a m B ak
V ie n g K h a m
40
Numbe
er of Runs ( out of 99 )
35
30
25
20
15
10
5
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
D a y o f O u tb re a k
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
18
Time to Outbreak Identification
No EWORS
Mean= 23.2, Med = 18
N = 56
EWORS in LP
Mean= 15.9, Med = 13
N = 70
EWORS in LP & NB
Mean= 8.4, Med = 8
N = 99
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
19
Media
an Days to Outbre
eak ID
Outbreak Identification & Reporting
35
No EWORS
30
EWORS in Luang Prabang
25
EWORS in Luang Prabang & Nam Bak
20
15
10
5
0
Luang Prabang
Chomphet
Nam Bak
Vieng Kham
Seed Province
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
20
Conclusions
• Outbreak identification time
– up to 2-week improvement in median identification time with
EWORS system
– Additional few days’ advantage with district-level system in Nam
Bak
• Rate off epidemic spread
– Probability of out-of-district infection within 3 days > 50% in every
scenario
– Infection reaches town of Luang Prabang within 4 days, regardless
of seed district
• Variability among runs
• Effect of rapid test capability at provincial hospital: the median dropped
below 6 days, even without EWORS, in each scenario
• Modeling
g surveillance capability
p
y is important
p
 Objectives  Background  Approach & Methods  Validation & Study Design  Results  Conclusions
21
References
•
•
•
•
Lao PDR, M. (2001). Report on National Health Survey, health status
in Lao PDR. Vientiane.
F
Ferguson,
N.,
N ett al.
l (2005)
(2005). "St
"Strategies
t i ffor containing
t i i an emerging
i
influenza pandemic in Southeast Asia." Nature 437: 209-14
Glass, R., et al. (2006). "Targeted Social Distancing Design for
P d i IInfluenza
Pandemic
fl
E
Emerging
i IInfectious
f ti
Di
Diseases."" E
Emerging
i
Infectious Diseases 12(11).
Feighner, B. (2007). Pandemic influenza policy model, 2007
comprehensive
h
i project
j t report.
t L
Laurel,
l MD
MD, JJohns
h H
Hopkins
ki U
University
i
it
Applied Physics Laboratory and the Department of Defense Global
Emerging Infections System (DoD-GEIS).
22
BACKUPS
23
Included Modeling Elements
Population details for
“agents”
• Age group and sex
• Statistics for rural/urban,
north/south/central
• Occupation category from
census data
• Travel survey information
• Provincial geography
• SES surrogates
– Occupation
– Literacy, education
– Clean water availability
– Household income
Disease spread
Health-care-seeking
care seeking
• Health
behavior
• Travel patterns (interregional spread)
• Immunocompetence,
based on statistics for:
– Age
– Pregnancy
– Nutrition
– Vitamin A intake
24
Infected Agent Attributes
– Static attributes (drawn from population data )
•
•
•
•
•
•
•
•
Age
Sex/pregnancy status
Family size
Peer group size
Region / Location (Province)
Rural Access to Road
Road, Rural No Access to Road
Road, Urban
Occupation
Type of Health Care Access
– Dynamic Attributes
• Disease State
• Ambulatory
A b l t
• Infectiousness
25
Modeling Person-to-Person
Transmission
• D
Draw IInfector’s
f
’ Disease
Di
Stage
S
Times,
Ti
TD(stage)
• Compare time to infection TI to TD at
each stage
• If TI > TD for all stages,
stages contact not
infected.
Adapted from: Glass RJ, Glass LM, Beyeler WE, Min HJ. Targeted social distancing design for
pandemic influenza. Emerg Infect Dis. 2006 Nov;12(11):1671-81.
26
Lao PDR Statistics
Related to SES
Key
y Survey
y Percentages
g
Literacy
Safe water
Ru
ra
l
Ur
ba
n
na
l
Na
t io
So
ut
h
Nearest hospital <4 km
away
Medical practitioner in
village
Villages with 4 essential
drugs
g
Ce
nt
ra
l
No
r th
100
90
80
70
60
50
40
30
20
10
0
Ministry of Health,
Health National Institute of Public Health,
Health Lao State Planning Committee,
Committee
National Statistical Center, Report of National Health Survey: Health Status of the People In
Lao P.D.R., Vientiane, 2001.
27
Inference of Reporting Delays
from Data
Comparison by year:
Ratio of (Visits sent by day)/All Visits
Days Late:
< 40%
40-60%
60-80%
60
80%
> 80%
28