1. Maintaining Elimination in an Environment of Persistent Importation
Kurt Frey April 15, 2019

2. Measles surveillance within an agent-based modeling framework
IDM measles aims to model the major components of an eradication strategy:
Routine Immunization
Scheduled catch-up and maintenance campaigns
Disease surveillance
Reactive campaigns
Today’s focus is on surveillance and response for agent-based simulations:
Incidence Focus
Susceptibility Focus
Surveillance Contexts
Model Calibration
Reporting, Extent, Promptness
Efficiency
Known / Inferred

3. Total susceptibility is the main factor in measles burden calculations
Timeseries data informs transmission dynamics, not total burden.
Endemic and near-elimination contexts assume nearly universal exposure to infection. Multi-year susceptible accumulation only occurs after interrupting transmission.
WHO Reported Cases
Country
Cases
MCV1
India 82
China
74 448
---
Mongolia
49 095 86
Nigeria
41 160 42
Indonesia
24 802 80
Ethiopia
24 258 55
Kyrgyzstan
17 784 96
D. R. Congo
13 578 72
Pakistan
10 703 61
Mongolia
Nigeria
16k
12k
Nigeria and Mongolia are examples of the contrast between endemic and re-introduction contexts. 8k
Reported Cases 4k
2011
2012
2013
2014
2015
2016
2017
2018

4. Current agent-based simulations of measles transmission in Nigeria
Parameters appropriate to Nigeria were determined from reported age-at-infection and timeseries case reports.
High Infectivity Season
160
120
Populations at the admin-2 level were well-mixed and networked using a gravity-type model.
Supplementary immunizations (SIAs) followed the calendar of recorded activities for the period
Duration (Days) 80 40
Sept
Oct
Nov
Dec
Jan
Start Date 2k
0.90
60/k 1k
0.45
40/k
0.00
20/k
Population Density (km-2)
Geographic Centers
Routine Vaccination - Mean Fraction
Crude Birth Rates (yr-1)

5. Simulations parameters selected for a focus on near-elimination
Calibrated infectivity parameters were combined with a hypothetical schedule of SIAs.
SIA coverage was adjusted so interrupting transmission was likely, but maintaining elimination was uncommon.
1/ year
1/ year
100
1/ 2 years
1/ 3 years 75
Months with Incidence (%) 50
Targeted mean SIA coverages interrupted transmission in >90% of simulations.
Median monthly burden was non-zero due to re-introduction.
Continuous Transmission 25
Increasing SIA Coverage

6. National Monthly Incidence
Total burden was stable; timeseries outcomes were highly variable
Mean cumulative burden was around 5M total cases and ranged narrowly.
Distribution in time varied substantially.
104
Frequency
102
1.0M
Selected Examples
100
0.8M 1M 2M 3M 4M
Annual Incidence
0.6M
National Monthly Incidence
Both small outbreaks (<1k cases) and multi-year intervals between large outbreaks were common.
0.4M
0.2M
0.0M 5 10 15 20
Simulation Year

7. Additional SIAs were introduced in response to observed incidence
Recently added capability for measles EMOD simulations allows interventions as response to symptomatic incidence.
Burden
Reduction (%)
70%
90%
250
Two consecutive two-week periods with observed incidence signaled an outbreak.
Response was implemented rapidly (<3 weeks) after an outbreak.
Above 250km radius is the scale of a regional campaign.
100
50%
200
150 50
Response Radius (km)
30%
100 50
10%
0.1
1.0 10
100
Surveillance Rate (%)

8. Delays after outbreak identification limits maximum impact
Increased delay between outbreak identification and intervention reduced the achievable burden reduction.
Burden
Reduction (%)
250
100
A delay of 10 weeks was added onto the previous response scenario.
Intervention coverage is unchanged from the base case.
200
150
90% 50
Response Radius (km)
70%
100
50%
30% 50
10%
0.1
1.0 10
100
Surveillance Rate (%)

9. Increased Vaccine Need (%)
Vaccine demand provides bounds on outbreak response strategies
Response increases vaccine need; at fixed burden reduction, improved surveillance tended toward more efficient vaccine usage. The tendency was stronger for greater effects. 16
-90%
Burden Reduction
Less surveillance at a given burden implies a larger response radius.
Vaccine need is relative to baseline SIAs usage: 18M doses / year.
As a limiting case, zero burden reduction has zero vaccine need.
-80% 12
-70%
-60% 8
Increased Vaccine Need (%)
-50%
-40% 4
-30%
-20%
-10%
0.1
1.0 10
100
Surveillance Rate (%)

10. Exploring proactively triggering SIAs based on known susceptibility
Locally focused SIAs based on levels of under-five susceptibility guarantee burden reduction, but tend to be less dose efficient than a rapid response to observed incidence.
100 20 80 15 60
Proactive
Burden Reduction (%) 10
Increased Vaccine Need (%) 40
Reactive 5 20 20 30 40 50 60 20 40 60 80
100
Under-Five Susceptibility Threshold (%)
Burden Reduction (%)

11. Embedded Python scripts allow highly customized response strategies
EMOD and Python run separate, concurrent processes, exchanging data as needed.
Response planning occurs in Python, using a subset of the EMOD simulation data.
Changing logic does not require rebuilding the EMOD executable.
Plans can leverage Python module libraries such as scikit-learn.
Start EMOD
Python
Start Python
Update EMOD Simulation
Update Python?
Update risk mapping model Y N
Plan reactive response

12. Acknowledgements
National Primary Health Development Agency and NCDC (Ministry of Health of Nigeria)
World Health Organization Nigeria office
World Health Organization Headquarters
Funding by Bill and Melinda Gates through the Global Good Fund
** Figure not to scale