A Simulation-Based Assessment of Mixed-Effects Cox Model Misspecification

• Activity Number: 42
• Type: Contributed
• Date/Time: Sunday, August 3, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Adel Elghafghuf*+ and Henrik Stryhn
• Companies: University of Prince Edward Island and University of Prince Edward Island
• Keywords: random effects distribution ; frailty models ; robustness ; misspecifying frailty distribution ; variance components ; simulations

Abstract

Mixed-effects Cox models can be fitted as Poisson generalized linear mixed models (GLMMs) after transforming time-to-event data to the Poisson GLMM framework. Estimation in these approximating models is based on Poisson maximum likelihood theory, assuming a specific distribution for random effects. However, the validity of the random-effects distribution assumption is often difficult to verify. In this study we assess, through simulations, the robustness of Poisson maximum likelihood estimation for a Cox model with normal random effects under misspecification of the random-effects distribution. The impact of misspecifying the distribution of random effects is assessed based on shared frailty, random slope, and nested frailty Cox models. Factors such as magnitude of the random effect variances, censoring rate, group size, and number of groups were accounted for in the assessment. In the simulations, the Poisson likelihood approach produced robust estimates under misspecification of the random-effects distribution for within-group fixed effects and in a wide range of situations for between-group fixed effects. Non-robust estimation of variance components was observed for a large magn