Statistical Modeling of Epidemics and Contact Networks That Transmit Them

• Activity Number: 513
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Jun Yin*+ and Brian J. Smith
• Companies: and University of Iowa
• Keywords: Bayesian Survival Analysis ; Epidemics ; Networks ; Latent Space ; Influenza ; Statistical Software

Abstract

Different attempts have been made to modify traditional epidemic models to relax homogeneity assumptions, so as to handle more complex and realistic situations. We propose a network-based approach to the modeling and prediction of infectious disease outbreaks.

Our focus is on heterogeneous populations where there is variation in individual susceptibility, infectivity, and person-to-person contact patterns. To address the complexity of disease propagation over a contact network, we develop a Bayesian survival model that maps the network onto a latent space and uses latent positions to predict disease transmission.

We present an R package ('epinet') implementation of our methods and an application to a high school contact network. The package uses C code to implement an MCMC algorithm to efficiently estimate parameters and predict disease outcomes. Our application involves contact data collected by mobile sensors distributed to individuals, and provides estimates of influenza transmission in line with the network structure. In it, we address issues that are of direct interest to public health professionals, such as prediction of future outbreaks and their expected length.