Keywords: Missing Data, Causal Effects
Treatment noncompliance and missing data are common problems in clinical trials. They are especially important for testing equivalence between a generic drug and an innovator drug in clinical endpoint bioequivalence (BE) studies. This is because the current primary equivalence analysis is based on the per-protocol (PP) population, which is mostly composed of completers and compliers. According to Frangakis and Rubin (2009), however, PP analysis is subject to selection bias because treatment non-compliance and drop out are post-treatment variables, and a crude comparison within the PP population is generally not a causal effect. In 2015, generic drugs accounted for 86% of the market share in U.S. However, statistical research in bioequivalence tests, especially how to handle complications such as missing data and non-compliance data, has been lacking. The missing data working group of the FDA (Lavange & Permutt 2016, Permutt 2016) recommended to use causal inference to handle missing data in clinical trials. In this presentation, we aim to propose causal effect estimands for testing equivalence and to propose potential primary and sensitivity statistical approaches for estimating these causal estimands in clinical endpoint bioequivalence studies in the presence of missing data and non-compliance. Simulation and real data analyses will be provided to illustrate the proposed causal inference approach to evaluate equivalence.