|
Abstract:
|
For pivotal trials, regulatory authorities request pre-specified primary and sensitivity analyses for assessing the robustness of the primary analysis result when substantial missing data is anticipated. The National Research Council (NRC) report (2010) questioned the reasonableness of the missing at random setting, and encouraged use of a not missing at random (NMAR) setting as the primary analysis. This presentation follows the NRC report with a focus on design strategies for primary and sensitivity analyses based on the NMAR assumption. We propose a process to investigate the mean-shift model with a non-future dependence missing data mechanism. The goal is to provide a method for finding an appropriate shift parameter to specify for the primary analysis in the protocol or SAP, based on the criteria of maintaining the type-I error rate for late phase trials using simulations based on early phase data or information from interim data from the current trial. Several components of the shift parameter, such as the metric, magnitude, and the algorithm are considered when examining the type-I error rate; and the robustness of the results are presented and discussed.
|
