A Multi-fidelity Approach for Efficient Network-based Simulation of Disease Outbreaks

Brian M. Adams
Optimization and Uncertainty Estimation Department
Computation, Computers, Information, and Mathematics Center
Sandia National Laboratories

Effective public health interventions during a disease outbreak depend on rapid characterization of the initial outbreak and pathways for pathogen spread. Epidemiologically-based modeling and simulation can characterize both and enable practitioners to test intervention strategies. While compartmental differential equation models are often used to represent bulk epidemic properties, they are unsuitable for early time simulations (first few days) when a small number of people are infected (and even fewer symptomatic), nor are they capable of representing spatial disease spread.

In this talk I will present social contact network-based models for early epoch simulation of disease outbreaks and bioterror incidents. They generate detailed predictions of disease spread, including tracking symptomatic individuals, for use in a Bayesian inverse problem context to determine the initial location, time, and size of an outbreak. Our implementation allows for high-fidelity predictive simulations, made efficient though scalable parallelism, as well as reduced-order simulations through population and/or location sampling, and static (time-projected) graph-based simulation, which are practical for inverse problem solution. Efficiently inverting data on early symptomatic presenters characterizes the original source, and the forward simulation can then suggest others who may be at risk

Back to the program