All Times EDT
Keywords: missing data, multiple imputation, estimands
Missing data is a common problem in clinical trials that can arise for different reasons. It can lead to information loss and potentially biased estimates if not handled properly. Missing data nature is usually complex, with a mixture of different mechanisms: missing completely at random (MCAR), missing at random (MAR) and missing not at random (MNAR). As a result, different missing data handling approach may apply. For example, in typical Atopic Dermatitis clinical trials, non-responder imputation (NRI) is the standard approach to impute binary efficacy endpoints, where subjects with missing efficacy data due to any reason are considered as treatment failures. However, NRI is a conservative approach that usually underestimates treatment effect and provides attenuated variance estimates. In addition, ICH E9 (R1) addendum emphasized the need to clearly define estimands in clinical trials. The description of an estimand requires specification on how intercurrent events are reflected. Handling of missing data approach would need to align with the prespecified estimands. Multiple imputation has gained increasing popularity in randomized clinical trials, where the missing data are imputed multiple times and the results of multiple complete-data analysis are combined. Multiple imputation provides valid inferences with variability in imputed datasets accounted for. However, multiple imputation could lead to biased results when the MAR assumption is implausible. Hybrid approaches are also adopted in recent clinical trials to handle different types of missing data. In this roundtable session, the following questions will be discussed: 1). When should different imputation approaches be used in clinical trials? 2). What are the things to consider when implementing multiple imputation? 3). How will multiple imputation fit in the estimand framework?