|
Abstract:
|
In a registration clinical trial, when the ignorability of missing data is considered an appropriate assumption in a context of the likelihood-based approach, a sensitivity analysis may provide an inferential statement for a scenario of non-ignorable missing data, which serves to make a comparison with an inferential conclusion obtained from the primary analysis under the ignorable missing data assumption. This comparison is considered to provide a tool for checking the robustness of the primary analysis result to possible violation of the ignorability assumption. Several sensitivity analysis methods have been used in the NDA clinical trials. The list includes a delta-adjusted Pattern Mixture Model, a control-based Pattern Mixture Model, Pattern Mixture Model with Non-future dependency, etc. In this presentation, we assess the statistical properties of commonly used sensitivity analysis methods using repeated measure trials data (data simulated from a few example NDA trial data). We also examine the usefulness of the sensitivity analysis methods in drawing a conclusion on the robustness of the primary efficacy analysis result.
|
