Estimation For Accelerated Failure Time Model With Dependent Censored Data

Investigators often obtain the censoring data in clinical and epidemiologic studies.Traditional statistical methods and models for censored-data are almost based on the hypothesis that the failure time and observation time are independent.But in practice the hypothesis is always invalid.Dependent censoring has recently attracted a great deal of attention.In this article we model event and censoring time marginally through the AFT models,and their association by a known copula.The joint distribution of the failure and censoring times is assumed to be a function of their marginal distributions.This function is called a Copula.Copula function is widely used in dependent censoring.In right censored data,we will use Copula and the idea of ”redistribution of mass” to obtain the likelihood function,use copula to obtain the likelihood function in interval censored to the AFT model with dependent censoring.The proposed method can improve the estimators when there exist the association between the event and the censoring time.Simulation results show the improvement of the proposed method compared to the naive method under independent censoring. Sensitivity analysis gives the evidences that the proposed method can obtain reasonable estimates even when the form of copula are misspecified. We illustrate its application by analyzing prostate cancer data.