Marginal proportional hazards models for multiple event-time data

Abstract

The Wei et al. (1989) semiparametric approach to the analysis of multiple event-time data assumes that each event time is related to covariates through a proportional hazards model with a completely unspecified baseline hazard function, but does not impose any constraint on the joint distribution of different event times. As a result of the order restriction, it is not clear whether or not event times can simultaneously satisfy their respective marginal proportional hazards assumption, while having continuous joint distribution. This leads to inability to conduct simulation studies. We resolve this issue by constructing parametric models for multiple event times with proper joint density functions and marginal proportional hazards. Simulation studies are reported that compare efficiencies of the method of Wei et al. (1989) and of the maximum likelihood approach.