Multiple Imputation Approaches for Missing Time-to-Event Outcomes with Informative Censoring: Practical Considerations from a Simulation Study Based on Real Data

Abstract

Missing outcomes data represent a common threat to the validity and robustness of clinical trials with time-to-event outcomes. Recent extensions of multiple imputations (MI), namely controlled-MI, have been introduced as a viable approach for sensitivity analysis in the presence of informative censoring, yet they lack validation based on real data. In this study we used data from a randomized trial to generate realistic scenarios of censoring mechanisms and compare several imputation approaches for missing outcome data. Our results confirm the relevance of multiple imputations especially in studies with long follow-up and higher proportion of potentially informative censoring.