B1730
Title: Classical and Bayesian approaches for the mixture cure model with high-dimensional covariates
Authors: Fatih Kizilaslan - University of Oslo (Norway) [presenting]
David Michael Swanson - Department of Biostatistics The University of Texas MD Anderson Cancer Center Houston TX (United States)
Valeria Vitelli - University of Oslo (Norway)
Abstract: In survival analysis, the presence of substantial censoring following a prolonged follow-up period often indicates the presence of cured individuals who may never experience the outcome of interest. This phenomenon can be observed in certain clinical investigations, such as cancer studies, and becomes more noticeable among patients diagnosed in the early stages of cancer. Furthermore, progress in cancer treatments has led to a substantial number of patients being classified as cured. In such circumstances, it is reasonable to consider a mixture cure model that combines cured and uncured fractions, rather than using traditional methods in survival analysis, which assumes that all individuals in the sample will eventually experience the outcome of interest. In the mixture cure model, the overall population lifetime is defined by weighting the survival time of susceptible and cured patients with the uncured and cured rates. A mixture cure model is explored, suitable to handle large high-dimensional covariates, such as molecular data. The proposed model can accommodate both parametric and non-parametric approaches for the distribution of susceptible patients, taking into consideration both classical (via a novel expectation-maximization) and Bayesian inference methods. Extensive simulation studies are conducted to evaluate the performance of the models. The results of these simulation studies are presented along with the analysis of real data from a breast cancer study.
