BAYESIAN MODEL AVERAGING IN PROPORTIONAL HAZARD MODELS - ASSESSING THE RISK OF A STROKE

In the context of the Cardiovascular Health Study, a comprehensive inv estigation into the risk factors for strokes, we apply Bayesian model averaging to the selection of variables in Cox proportional hazard mod els. We use an extension of the leaps-and-bounds algorithm for locatin g the models that are to be averaged over and make available S-PLUS so ftware to implement the methods. Bayesian model averaging provides a p osterior probability that each variable belongs in the model, a more d irectly interpretable measure of variable importance than a P-value. P -values from models preferred by stepwise methods tend to overstate th e evidence for the predictive value of a variable and do not account f or model uncertainty. We introduce the partial predictive score to eva luate predictive performance. For the Cardiovascular Health Study, Bay esian model averaging predictively outperforms standard model selectio n and does a better job of assessing who is at high risk for a stroke.