Data-Driven Infectious Disease Control with Uncertain Resources

TL;DR

This paper introduces a data-driven, two-stage optimization model for resource allocation in infectious disease control under uncertainty, demonstrating significant reductions in infections using real epidemic data.

Contribution

It develops a flexible, data-driven modeling approach that transforms epidemic control into a tractable mixed-integer linear program for online decision making.

Findings

Reduced infections by approximately 400 cases in Sierra Leone.

Validated model with real Ebola epidemic data.

Provided insights into optimal regional resource allocation.

Abstract

We study a resource allocation problem for containing an infectious disease in a metapopulation subject to resource uncertainty. We propose a two-stage model where the policy maker seeks to allocate resources in both stages where the second stage resource is random. Instead of a system of nonlinear differential equations that governs the epidemic trajectories in the constraints of the optimization model, we use a data-driven functional form to model the cumulative number of infected individuals. This flexible data-driven modeling choice allows us to transform the optimization problem to a tractable mixed integer linear program. Our flexible approach can handle an online decision making process, where the decision makers update their decisions for opening treatment units and allocating beds utilizing the new information about the epidemic progress. We utilize a detailed simulation model, validated by real data from the 2014 Ebola epidemic in Sierra Leone. Our results show that our policies produce about 400 fewer number of infected individuals in Sierra Leone compared to the policies applied during the actual epidemic. We also provide a detailed comparison of allocation policies generated by our optimization framework which sheds light on the optimal resource allocation in different regions.