Title: Simulating survival data for cure models in overall or net survival framework
Authors: Juste Goungounga - University of Burgundy, INSERM, UMR1231 (France) [presenting]
Olayide Boussari - University of Burgundy (France)
Valerie Jooste - University of Burgundy (France)
Abstract: Simulation studies are pivotal in survival analysis to evaluate the performances of new and existing models. When the cure assumption holds, cure models allow estimating either overall survival or net survival (survival that would be observed if the studied disease -say cancer- were the only possible cause of death) and its asymptotic value, the cure fraction. Net survival is usually estimated by splitting the observed mortality into two forces: one due to cancer (excess mortality) and one due to other causes (expected mortality). Hence, a mechanism to generate survival data for the cure model in a net survival framework must include two independent time variables: time until death by cancer and time until death by other causes. To reflect plausible real data, a specified cure model for data generation could also require flexible and complex functions. We present methods for using different distributions when generating survival times by varying time-to-cure and cure fractions. We illustrate a numerical integration approach that could be used when a closed-form of cumulative hazard due to cancer does not exist and root finding techniques as an alternative to simulate survival times from mixture and non-mixture cure models. A user-friendly R package is also provided.