Considerations Regarding Choice of the Primary Analysis: FDA Perspective

• Activity Number: 157
• Type: Invited
• Date/Time: Monday, August 4, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Robert T. O'Neill*+
• Companies: Food and Drug Administration

Abstract

In addition to the accepted standard of specifying in a protocol key primary endpoints and the primary hypotheses to be tested in confirmatory clinical trails, it is also generally considered good statistical practice to specify the data analysis plan for a clinical trial prior to observing any unblinded data. The prespecification of the methods of analysis is intended to limit data-driven models and data-driven analyses that may bias overall conclusions derived from the trial. However, in longitudinal clinical trials that almost certainly will experience some subject dropouts, difficulties arise with anticipating the types, amount and patterns of subject dropouts for each treatment cohort so that an intention to treat analysis which preserves the initial randomization can be carried out. Regulators attempt to strike a balance between optimal statistical methods for the problem at hand, and inferences that are too dependent on the strength of the analysis assumptions and the untestability of these assumptions. The talk will consider a range of issues involved in reaching a fair balance in the subject dropout saga.