A Multiple Imputation Approach to Distinguish Curative From Life-Prolonging Effects in the Presence of Missing Covariates

Medical advances have increased cancer survival rates and the possibility of finding a cure. Hence, it is crucial to evaluate the impact of treatments both in terms of cure and prolongation of survival. To achieve this, we may use a Cox proportional hazards cure model. However, a significant challenge in applying such a model is the potential presence of partially observed covariates. We aim to refine the methods for imputing partially observed covariates based on multiple imputation and fully conditional specification approaches. To be more specific, we consider a general case in which different covariate vectors are used to model the probability of cure and the survival of patients who are not cured. In a large-scale simulation experiment, we investigated the performance of the multiple imputation procedure based either on the exact conditional distribution or on an approximate imputation model, which helps to draw imputed values at a lower computational cost. To assess the effectiveness of these approaches, we compare them with a complete-case analysis and an analysis that includes all available covariates in modeling both cure probabilities and the survival of the uncured. We discuss the application of these techniques to a real-world dataset from the BO06 clinical trial on osteosarcoma.