Missing covariates in mixture cure models: a multiple imputation approach

Medical advances have greatly improved cancer survival rates, making it essential to evaluate treatments in terms of both cure and survival prolongation. Mixture cure models have been introduced in survival analysis to differentiate between cured patients and those at risk. However, missing covariates can bias the estimation of such mod-els. We address this issue by applying multiple imputation techniques to mixture cure models. Specifically, we extend currently available proposals by considering different covariate vectors for modelling the probability of cure and survival of uncured patients. We introduce and compare exact and approximate imputation approaches, evaluating their performance through simulations and a case study in osteosarcoma patients from the BO06 clinical trial.