Dynamic Space-Time Model for Syndromic Surveillance with Particle Filters and Dirichlet Process

Abstract

Massive spatio-temporal data are challenging for statistical analysis due to their low signal-to-noise ratios and high-dimensional spatio-temporal structure. To resolve these issues, we propose a novel Dirichlet process particle filter (DPPF) model. The Dirichlet process models a set of stochastic functions as probability distributions for dimension reduction, and the particle filter is used to solve the nonlinear filtering problem with sequential Monte Carlo steps where the data has a low signal-to-noise ratio. Our data set is derived from surveillance data on emergency visits for influenza-like and respiratory illness (from 2008 to 2010) from the Indiana Public Health Emergency Surveillance System. The DPPF develops a dynamic data-driven applications system (DDDAS) methodology for disease outbreak detection. Numerical results show that our model significantly improves the outbreak detection performance in real data analysis.