Flexible Regression Model Selection for Survival Probabilities of an Absorbing Event

• Activity Number: 122
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Rui Wang*+ and Gregory DiRienzo
• Companies: The State University of New York at Albany and The State University of New York at Albany
• Keywords: Survival Analysis ; Cox model ; Cross-validation ; Missing data

Abstract

We propose a strategy to flexibly model survival probabilities as a function of baseline covariates for an absorbing survival event such as death. This strategy requires modeling the censoring random variable, which is right-censored in this situation. We propose a flexible approach to modeling the censoring variable that re-estimates its distribution as a function of covariates at each event time among only those subjects at risk. This setting can be cast as a missing-data problem where augmented inverse probability weighted complete-case estimators can be calculated. Specifically, we obtain consistent and asymptotically normal estimators of regression coefficients and average prediction error for each working survival model at the given time points of interest, that are free from the nuisance censoring variable. The model selection strategy uses multiple hypotheses testing procedures that control a given error rate when comparing working models based on estimates of average prediction error. An extensive simulation study shows the methods generally perform well and they are illustrated with analysis of the popular Mayo Clinic trial of Primary Biliary Cirrhosis (PBC)