Nonproportional hazards regression models for survival analysis.

摘要

Censoring arises naturally in clinical trials, biological research, reliability studies and many other fields. The two most commonly encountered censoring mechanisms are right censoring and interval censoring. Regression analysis of censored survival data has attracted a great deal of attention and for this, many methods have been proposed. However, most of these methods employ the proportional hazards model which may not fit the data well and assume that the censoring mechanism is independent of the survival time of interest.;In the first part of this dissertation, we discuss fitting the additive hazards regression model to right-censored data when covariates are subject to measurement errors. We propose a nonparametric-correction approach for inference about regression parameters and the baseline cumulative hazard function. The second part of this dissertation considers the analysis of case I interval-censored data when the observation time may be related to the underlying survival time. Inference procedures are presented for estimating regression parameters under, again, the additive hazards regression model. In the third part of this dissertation, we investigate fitting a class of generalized proportional hazards models, the linear transformation models, to case II interval-censored data. For the estimation of regression parameters and the prediction of survival probabilities, estimating equation approaches are presented. Extensive simulation studies are conducted and suggest that the proposed methodologies work well. The approaches are applied to four real datasets that motivated this study and some remarks are given.