Parameterizing Individual-Level Models of Infectious Disease Spread Using Sampling-Based Likelihood Approximations

• Activity Number: 134
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Rajat Malik*+ and Rob Deardon and Grace Pui Sze Kwong
• Companies: Department of Mathematics & Statistics, University of Guelph and University of Guelph and Department of Population Medicine, Ontario Veterinary College, University of Guelph
• Keywords: individual-level models ; infectious diseases ; spatial models ; MCMC ; sampling ; Bayesian inference

Abstract

Individual-level models (ILMs), fitted in a Bayesian MCMC framework, are a class of discrete time models used to model the spread of infectious diseases. They can account for spatial and temporal disease dynamics by modelling the infectious pressure exerted by infected individuals of a population on each susceptible individual. Unfortunately, for large populations, quantifying this infectious pressure can be computationally burdensome leading to a time-consuming likelihood calculation. Here, we introduce sampling methods to speed the calculation of the likelihood function while accounting for the accuracy of model parameterization and compare the performances of these methods via simulation studies. Decreasing likelihood computational time is accomplished by obtaining a subset of the infected set of individuals within the population at each time point. Using a simple random sampling approach, the results show that as the sampling proportion of the infected set increases, ILM parameter estimates become closer to their true values. However, our results suggest substantial computational savings can be attained with acceptable information loss.