A Finite Mixture Model for Sample size calculation for Survival Endpoints – the ‘Pseudo Cure Model’

Most patient populations consist of a heterogenous mix of patients with different underlying risk levels. As patients at higher risk tend to have events earlier, the average risk of the remaining at-risk patients drops. This undermines the constant hazard assumption common to many survival modelling techniques, and leads to bias in the estimation of hazard ratios associated with important covariates of interest, especially treatment effects. A novel approach to counteracting this bias is to model the hazard using a 2-component mixture model counteracts this bias. This work will develop the modelling tools to do this, and compare to current methods in the literature.