All Times EDT
Keywords: Historical Borrowing, Control Arm, Real World Evidence
The world is awash in data. In almost every clinical trial we have access to information about the control or treatment arm in the form of real world evidence or historical randomized trials. Borrowing this data creates opportunities and risks. To the extent the historical data is "on point" and matches the current trial, borrowing can decrease type I error, increase power, and improve estimation. To the extent the historical data is discordant with the current trial, we may inflate type I error, reduce power, and creates biased estimates of treatment effect. In this talk we will relate Bayesian understanding of the magnitude of "drift" (differences between historical data and current parameters) and the "sweet spot", or region of the parameter space where borrowing is beneficial. We will illustrate how this region can change based on differing forms of borrowing, and illustrate how modern methods such as clustering borrowing can increase both the range and magnitude of benefit.