Semiparametric Extended Hazard Cure Models

Abstract

For right censored survival data in presence of cure, we propose a new class of cure models, which include the proportional hazards (PH) cure model, the accelerated failure time (AFT) cure model, and the accelerated hazards (AH) cure model as special cases. Mixture regression cure models have been developed to formulate the cure proportion and the conditional survival distribution of uncured subjects with a logistic regression and a semiparametric survival regression, respectively. In this study, we generalize them to the EH cure models via an alternating algorithm to obtain the maximum likelihood estimators of the model parameters by approximating the unspecified underlying hazard function with spline functions. We use the boundary hypothesis test of the parameter space to derive the asymptotic distribution of the deviance for selecting a suitable cure model among the PH cure, AFT cure, AH cure, or some other EH cure models in analyzing a dataset. The proposed approach to EH cure models is evaluated in simulation studies, and illustrated by analyzing some real datasets. This is a joint work with Dr. Chen-Hsin Chen.?