|
Abstract:
|
Test-then-pool (TP) is a simple approach that borrows historical information to improve efficiency of drug development. The original approach examines the difference between the historical and the current information and suggests pooling if no significant difference. One drawback is that no significant difference may not imply consistency between the historical and the current information. An insignificant p-value may result from a small sample size in the current control, rather than no difference between the historical and the current control. Therefore, the current trial may often borrow the historical information using the original approach. Statistically, it is more natural to use an equivalence test for examining the consistency. In this talk, we will discuss the equivalence-based approach for a continuous endpoint, explain the relationship between the two TP approaches, explore the choice of an equivalence margin through the overlap probability, and propose an adjustment to the nominal testing level for controlling type I error under the true consistency. Statistical properties of the approaches and their applications to a real clinical trial will be presented.
|
