Divide-Recombine Prediction for Virginia Lyme Disease Emergence Based on Spatio-Temporal Count Data

• 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): Yuanyuan Duan*+ and Jie Li and Korine Kolivras and Stephen Prisley and James Campbell and David Gaines and Yili Hong
• Companies: Abbvie and Virginia Tech and Virginia Tech and Virginia Tech and Virginia Tech and Virginia Department of Health and Virginia Tech
• Keywords: Divide-Recombine ; INLA ; Markov random field ; Spatio-temporal ; Intrinsic CAR

Abstract

The increasing demand for modeling spatio-temporal data is computationally challenging due to the large scale dimensions involved. The traditional Markov Chain Monte Carlo (MCMC) method suffers from slow convergence and is computationally expensive. The Integrated Nested Laplace Approximation (INLA) has been proposed as an alternative to speed up the computation process by avoiding the extensive sampling process required by MCMC. However, even with INLA, handling large-scale spatio-temporal datasets remains difficult, if not infeasible, in many cases. We propose a new Divide-Recombine (DR) prediction method for dealing with spatio-temporal data. A large spatial region is divided into smaller subregions and then INLA is applied to each subregion. To recover the spatial dependence, an iterative procedure has been developed to recombine the model fitting and prediction results. Stable estimation/prediction results are obtained after several updating iterations. Simulations are used to validate the accuracy of the new method in model fitting and prediction. The method is then applied to the areal (census tract level) count data for Lyme disease cases in Virginia from 2003 to 2010.