Estimating baseline distribution in proportional hazards cure models

Abstract

Semiparametric proportional hazards cure models have been proposed recently for survival data from some clinical trials where cure is a possibility. However, it is known that the estimated baseline distribution in the proposed methods will not be proper if there is no restriction on the tail of the distribution, which leads to the identifiability problem. In this work, we investigate several methods that impose restrictions on the tail of the baseline distribution. A simulation study shows that the proposed methods are useful in reducing estimation errors in the cure models. The impact of the different methods on the estimation of regression parameters and survival probabilities of patients who are not cured is detailed in the paper. The results provide useful guidelines for practitioners to select appropriate estimation methods for the semiparametric cure model. The application of the results is illustrated with a real data set from a clinical trial of breast cancer.