Cox Model Selection Based on Information Criterion

• Activity Number: 353
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Yu-Mei Chang*+
• Companies: Tunghai University
• Keywords: adaptive penalty parameter ; data perturbation ; generalized degrees of freedom ; Kullback-Leibler loss ; right-censored survival data

Abstract

In medical studies, Cox proportional hazards model is the most commonly used method to analyze the survivor function of patients when the right-censored survival data are accompany with covariates which are associated with patients' physiology and conditions. For model selection, Akiake's information criterion (AIC) and Bayesian information criterion (BIC) are the most commonly used methods. However, the two information criteria may generally perform differently under different circumstances, so that neither the AIC nor the BIC is superior in all cases. In practice, the underlying true model is unknown and model selection is only based on the specific selection criteria. Therefore, we propose a data-driven method to choose the optimal penalty parameter based on a concept of generalized degrees of freedom, resulting in an approximately unbiased estimator of the Kullback-Leibler loss via a data perturbation technique. The effectiveness of the proposed method is justified by a simulation study and an asymptotic analysis. The application of the method is also illustrated by a real data set.