Multiple Imputation‐Based Sensitivity Analysis

Abstract In this article, we introduce the concepts of sensitivity analyses for missing data, and outline how these may be conveniently performed using multiple imputation (MI). Motivated by data from a trial and a cohort study, we describe how multiple imputation may be used for sensitivity analysis with both pattern mixture and selection models. We then discuss eliciting expert opinion for sensitivity analysis, reference based sensitivity analysis and extensions to longitudinal data. Our aim is to equip readers with the concepts, examples and references needed to apply these methods to their own data.