2014.ISDS.Abstracts.Final.pdf


ISDS Annual Conference Proceedings 2014. This is an Open Access article distributed under the terms of the Creative Commons Attribution-
Noncommercial 3.0 Unported License (http://creativecommons.org/licenses/by-nc/3.0/), permitting all non-commercial use, distribution, 
and reproduction in any medium, provided the original work is properly cited.

ISDS 2014 Conference Abstracts

Computational Method for Epidemic Detection in 
Multiple populations
Ekaterina Shatskikh* and Michael Ludkovski

Statistics, UC Santa Barbara, Goleta, CA, USA

Objective
Detect epidemics over multiple Populations using computational 

methods

Introduction
Currently Centers for Disease Control and Prevention (CDC) 

employ threshold rules to declare epidemic outbreaks, such as 
influenza, separately in each population. However each year influenza 
starts in one population and spreads population-to-population 
throughout the country. Therefore there is a need for an algorithm to 
declare the epidemic that uses information from multiple populations.

Methods
The standard stochastic SIR model can be used to simulate the 

epidemic behavior in one population1. We use the extended model by 
coupling populations with each other, where the infection can spread 
from one population to another.

We apply the Quickest Detection method2 to announce the 
epidemic. To qualify the detection accuracy, we balance the costs 
of false alarms and detection delays. Thus, we will announce an 
epidemic if the future cost is greater than the immediate cost. Since 
epidemic evolution is stochastic, we generate epidemic scenarios and 
estimate the expected costs.

We use Sequential Regression Monte Carlo3 (SRMC) to efficiently 
approximate the decision rules. This consists of judiciously picking 
epidemic scenarios using sequential design and then applying 
regression techniques to construct the entire surface of expected 
future costs. Moreover SRMC allows the algorithm to self-correct by 
simulating additional epidemic scenarios for the model states where 
the algorithm is not sure about announcing the epidemic.

Results
The figure below illustrates our approach on a simulated dataset 

involving two populations of size 2000. We constructed a detection 
map based on the number of infected individuals in the first 
population It

(1) and posterior probability of epidemic in the second 
population Pt. The blue region corresponds to the regions where the 
epidemic should be announced; while the red region corresponds to 
the regions where the epidemic should not be announced. The small 
white regions (which can be seen on the boundary of the previous two 
regions) represent the regions where the algorithm is not sure about 
announcing the epidemic.

Our algorithm is better than the CDC approach because it uses the 
additional information about the number of infected individuals in 
other populations. This additional information helps the algorithm to 
see the whole picture about the state of epidemic.

Conclusions
We developed an algorithm for detecting an epidemic in a multiple 

population scenario. We are working on an extension of this algorithm 
to cover the case where we observe only partial information (for 
example, only diagnosed infected individuals) and we use particle 
filtering to estimate It

(1) and Pt.

Figure: Detection map for the second population in a two-population model 
constructed based on 2000 adaptively generated scenarios. Decisions are based 
on a loess metamodel.

Keywords
Simulation; Stochastic compartmental models; Early outbreak 
detection

Acknowledgments

This material is based upon work partially supported by the National 
Science Foundation under Grant No. ATD - 1222262.

References

1. Andersson H., Britton, T. Stochastic Epidemic Models and Their
Statistical Analysis. Lecture Notes in Statistics. Springer, 2000.

2. Poor H.V., Hadjiliadis O. Quickest Detection. Cambridge University
Press, 2008

3. Gramacy R. Ludkovski M. Sequential Regression for Optimal Stopping 
Problems. arXiv:1309.3832, 2013.

*Ekaterina Shatskikh
E-mail: shatskikh@umail.ucsb.edu    

Online Journal of Public Health Informatics * ISSN 1947-2579 * http://ojphi.org * 7(1):e158, 2015