Risk Factors Assessment and Bayesian Model Averaging with Hyperprior Structures in Matched Case-Control Studies

• Activity Number: 29
• Type: Contributed
• Date/Time: Sunday, August 9, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Yi Mu*
• Companies: CDC
• Keywords: Bayesian ; Bayesian model average ; hyperprior ; matched case control ; MRSA

Abstract

Risk factors assessment in case-control studies are often carried out using statistically variable selection methods, such as stepwise method. Inference is then carried out conditionally on the selected model, but this ignores the model uncertainty implicit in the variable selection process. This limitation may be addressed by adopting a Bayesian model averaging approach, which selects a number of all possible such models, and uses the posterior probabilities of these models to perform all inference and predictions. Traditionally, the regression coefficient ?j for every predictor was estimated in isolation from the others, for example, a prior distribution in which each of the ?j is independent and normal. In this study, we treated all the regression coefficients as coming from a shared t distribution, with hyperprior on slope coefficients across predictors. A desirable effect of this prior structure is that the estimates of the regression coefficient experience shrinkage and it helps control for "false alarms". The methods are applied and compared in the context of a matched case-control study of risk factors for invasive Methicillin-resistant Staphylococcus aureus infection.