CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Emulation based likelihood approximation for spatial infectious disease models Authors:  Gyanendra Pokharel - University of Winnipeg (Canada) [presenting]
Abstract: Mechanistic models for spatio-temporal infectious disease offer a great advantage in capturing heterogeneity in populations during an epidemic. These models quantify probabilistic outcomes regarding the risk of infection of susceptible individuals due to various susceptibility and transmissibility covariates, including their spatial distance from infectious individuals. However, such models are generally fitted in a Bayesian Markov chain Monte Carlo (MCMC) framework, which requires multiple calculations of what is often a computationally expensive likelihood function; thus, computationally prohibitive MCMC-based analysis. We propose an alternative approach, the so-called emulation technique. The model is again fitted in a Bayesian MCMC framework but replaces the computationally expensive true likelihood by the Gaussian process approximation of the likelihood function built over the design matrix constructed on a pre-defined parameter grid. We show that such method can be used to infer the model parameters and underlying characteristics of spatial disease systems and that this can be done in much more computationally efficient manner compared to the full Bayesian MCMC approach.