A Bayesian Model with Shrinkage Prior to Solve the Uncertainty of Modeled Hazard Air Pollutant Concentrations

• Activity Number: 183
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Yi Cai*+ and Michael Swartz and Wenyaw Chan and Philip J. Lupo
• Companies: University of Texas Health Science Center at Houston and University of Texas Health Science Center at Houston and University of Texas Health Science Center at Houston and Baylor College of Medicine
• Keywords: Uncertainity ; Shrinkage Prior ; Bayesian logistic regression ; Hazardous Air Pollutants

Abstract

Epidemiologic investigations evaluating the health effects of hazardous air pollutants (HAPs) often rely upon estimates generated as part of the Assessment System for Population Exposure Nationwide (ASPEN) model. The estimates have a high, medium and low confidence score, which provides a qualitative assessment about the uncertainty of the modeled concentrations. To our knowledge, no study has accounted for these confidence scores in their analyses. Because of that, we introduce a Bayesian logistic regression model using priors that vary according to the confidence score to induce more shrinkage towards the null for lower confidence. We have conducted a simulation study to assess our model, examining the performance of the shrinkage priors in logistic regression models using continuous or categorical covariates respectively. We reported the 95% credible interval of the regression coefficient to evaluate the effect of the shrinkage prior and the model's ability to identify the true association. Our results suggest that using a Bayesian approach to account for uncertainty in the ASPEN model estimates is a robust method for identifying true associations with adequate power.