Bayesian Spatially Varying Coefficient Models for Estimating the Toxicity of the Chemical Components of Fine Particulate Matter

• Activity Number: 597
• Type: Topic Contributed
• Date/Time: Thursday, August 2, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Yeonseung Chung*+ and Francesca Dominici and Brent Coull and Michelle Bell
• Companies: Korea Advanced Institute of Science and Technology and Harvard School of Public Health and Harvard School of Public Health and Yale University
• Keywords: Bayesian spatial process ; Fine particulate matter ; Hierarchical model ; Missing imputation ; Spatially varying coefficients

Abstract

Several studies have reported associations between long-term exposure to ambient fine particulate matter (PM2.5) and mortality. However, it is not much explored which chemical constituents determine the toxicity of PM2.5. The health effects of long-term exposure to PM2.5 vary across different locations and such spatial heterogeneity in health responses to long-term PM2.5 may be explained by the chemical composition of PM2.5. In this research, we propose a statistical model to investigate the spatially-varying (SV) health effects of long-term PM2.5 and the effect modification by the chemical components simultaneously. The model consists of two parts; (1) we use a Bayesian SV coefficient Poisson regression to quantify the spatially-heterogeneous toxicity of long-term PM2.5 on mortality; (2) we regress the chemical component levels on the SV coefficients to identify the components that modify the PM2.5 toxicity. Applying the proposed model to the US Medicare Cohort Air Pollution Study (MCAPS) data, we encounter a missing value problem for the chemical components and incorporate a Gaussian spatial process as an imputation procedure in the model.