A Frailty Model for Bivariate Interval-Censored Data Allowing Weak Dependence and Independence

• Activity Number: 457
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Naichen Wang*+ and Lianming Wang
• Companies: University of South Carolina and University of South Carolina
• Keywords: Interval-censored ; Bivariate ; Frailty ; PH model ; MCMC

Abstract

Interval-censored data commonly arise in real-life epidemiologic, social, and medical studies, in which participants undergo multiple examinations at different times. The failure time of interest is never observed exactly but is known to fall within some examination times. For multivariate interval-censored failure times, the gamma frailty PH model is widely used but is known to produce a large estimation bias when the events are weakly correlated or independent. In this paper, we propose a mixture of frailty model to solve this issue. Our approach allows one to test independence among the events of interest. A Gibbs sampler is proposed based on a data augmentation and is straightforward to implement. Our method is evaluated by simulation studies and illustrated by a real-life medical data application.