Adaptive Nonparametric Variable Selection for Survival Data

• Activity Number: 505
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Lisha Chen*+
• Companies: Yale University
• Keywords: variable selection ; survival analysis ; nonparametrics

Abstract

We consider the variable selection problem for survival data, that is, to identify the subset of covariates that are associated with survival time. Most existing methods adopted the widely used proportional hazards model and select variables by penalizing the covariate coefficients. We take the route of nonparametric testing and develop an algorithm to test the significance of each variable conditioning on the presence of all other variables. More specifically we exam the significance level of each variable by applying popular univariate testing methods such as log-rank test and logit-rank test after conditioning on all other variables. The state-of-art of our algorithm though lies in constructing the optimal conditioning operator by an adaptive iterative procedure.