Model Selection Under the Proportional Hazards Mixed Effects Model (PHMM)

• Activity Number: 187
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Michael Donohue*+ and Ronghui Xu and Anthony Gamst and Florin Vaida and David P. Harrington
• Companies: University of California, San Diego and University of California, San Diego and University of California, San Diego and University of California, San Diego and Dana-Farber Cancer Institute
• Keywords: model selection ; proportional hazards ; mixed effects ; Akaike information criterion

Abstract

In this talk we discuss model selection under the proportional hazards mixed effects model (PHMM). Recently established asymptotic properties of the nonparametric maximum likelihood estimator allow us to use the profile likelihood for selection of both nested and non-nested PHMMs. We define a profile Akaike information for general models with nuisance parameters. Asymptotic quadratic expansion of the log profile likelihood allows unbiased estimation of the Akaike information by a profile Akaike information criterion (pAIC). The pAIC focuses on the population parameters, such as the fixed effects and variance components. We will also discuss conditional AIC (cAIC), which, in turn, incorporates the estimated random effects. Computation of both pAIC and cAIC under PHMM will be addressed, and examples will be given to show their applications.