Abstract:
|
Borrowing historical data may lead to higher power and better estimation precision when comparing a new treatment to a control in a clinical trial (Viele et al 2014, Statistics in Medicine). However, not all clinical trials will benefit from historical data borrowing especially under stronger type I error control. The decision of borrowing historical data should be contingent on multiple factors, whether study populations are similar between historical trials and the new trial, whether the magnitude of power gain is meaningful under a certain degree of type I error control, and under which scenarios does borrowing overturn a non-significant result into a significant result. In this presentation, we propose a framework for the decision making of whether to adopt historical data borrowing methods for a clinical trial and a closed form for calculating type I error rates under historical data borrowing for binary endpoints.
|