|
Abstract:
|
We consider the modeling of semi-competing risks data in which observation of certain outcome events (say nonfatal cardiac events) is terminated by observation of other types of events (say death) but not vice versa. We propose a bivariate Lehmann model for modeling the dependence of such bivariate outcome data on covariates, for example the treatment assignment in a randomized clinical trial. The model requires that the logarithms of the joint survivor functions of the times to the two types of event over the observable region be a covariate-dependent multiple of the logarithm of a baseline joint survivor function. The proposed model includes some well-known frailty models, including those due to Clayton (1978) and Hougaard (1986), but is more general. We investigate some parametric and nonparametric strategies for fitting the model and compare these with the common approach of analyzing the time to the first event (of either type) as a so-called composite outcome.
|
