Developing Mitigational Strategies for Pandemic Outbreaks
Research Abstract
Most
epidemiologists believe that in the event of a human-to-human
transmittable avian influenza pandemic, as many as 90 million people
in many geographically dispersed cities in the U.S. are expected to
become ill. At present, scalable mitigation strategies do not exist
for such pandemic outbreaks. To fill this vacuum, we propose to
develop a multi-threaded large scale simulation model that would
mimic the real-time social dynamics in cross-regional pandemic
outbreaks. Using the simulation as the underlying model, we will
develop a two-tier optimization framework to generate mitigation
strategies at both national and local levels. The national strategy
will focus on optimal allocation of limited resources including
vaccines, prophylactic drugs, and healthcare personnel reserve. The
local strategy will aim at maximizing utilization of the allocated
resources, which will serve as an input for the national model. We
will address the following research objectives.
1. Build a highly parallelizable simulation model that will utilize
available high performance computing capabilities to assess societal
impact of large cascading pandemic outbreaks (e.g., avian
influenza).
2. Build
simulation-based optimization models to generate dynamic resource
allocation strategies at the national level and resource utilization
strategies at the local level.
3. Develop a computing framework to support synthesis of an
efficient computational strategy for simulation and optimization
models.
4. Demonstrate how strategic and tactical guidelines can be
developed for healthcare providers using sample avian influenza
pandemic scenarios involving multiple cities with millions of
people.
5. Develop educational and dissemination plans encompassing K-12,
undergraduate and graduate levels, and professional ranks.
Our research
approach will benefit from our recently developed prototype for a
large-scale multi-agent discrete event simulation system. It has an
architecture that is expandable to a multi-threaded framework, which
will facilitate a simultaneous simulation of thousands of subzones
of a cross regional outbreak involving multiple cities.
Implementation of such a multi-threaded simulation model is suitable
for our existing terascale computing facilities (e.g., TeraGrid).
The national level model will be invoked at each time epoch when the
outbreak spreads to an unaffected city, and will determine for the
remaining resources a strategy consisting of actual allocation to
the newly affected city and virtual allocation to the cities likely
to be affected. The resource utilization model at the local level
will be formulated as a Markov decision problem (MDP) and will find
an optimal multidimensional decision vector consisting of target
population groups for vaccine and antiviral drug distribution,
hospitalization eligibility, and quarantine policy. Due to the vast
nature of the state and action spaces, the MDP will be solved using
a reinforcement learning approach.

A Spread Model with Two-Tier Mitigation Approach
Publications
Conference Presentations
-
Das, T. K.,
Savachkin, A, Prieto, D., and Uribe, A.
A Simulation-based Optimization
Framework for Pandemic Impact Mitigation. INFORMS Annual Meeting 2007, Seattle,
WA.
-
Savachkin, A, Das, T. K., Prieto, D., and Uribe, A.
A
Simulation-Based Optimization Model for Dynamic Resource
Utilization at Federal and Local Levels During Cross-Regional
Pandemic Outbreaks.
INFORMS Annual Meeting 2007, Seattle, WA.
|