Presented by by Dr. Wei-Wen Hsu, PhD, Associate Professor, Department of Environmental and Public Health Sciences, University of Cincinnati College of Medicine

The cure rate models, also known as two-component mixture models, are widely utilized in survival data analysis, particularly in scenarios involving long-term survivors. Significant attention in the literature has been dedicated to assessing covariate effects on both the cure fraction and conditional hazard rate of these models. However, interpreting covariate effects on overall survival responses from a marginal perspective presents challenges, particularly when covariates are shared by the two components of the mixture model, which is further complicated by high-dimensional covariates. Motivated by the analysis of TCGA microarray data on breast cancer, our objective is to identify genes highly associated with survival outcomes marginally in the presence of long-term survivors. To achieve this, we propose a marginal cure rate model suitable for high-dimensional covariates. Our approach involves a novel reparameterization technique to directly relate covariates to the marginal mean hazard rate, with variable selection performed through a LASSO-type penalized likelihood function. Notably, parameter estimation within this framework is non-standard, necessitating the utilization of techniques such as quadratic approximation and cyclic coordinate descent algorithms to obtain estimates. To evaluate the performance of the proposed model, extensive simulations are conducted. Furthermore, the application of our model using TCGA microarray data on breast cancer is illustrated.