A method for incorporating historical information in non-inferiority trials

In non-inferiority trials a new experimental treatment is compared to an active control therapy. Often information on the current control therapy is available from previous studies. Exploiting historical evidence is useful both for reserving more resources to the arm of the new therapy and for improving the accuracy of es- timates, as long as current and historical control data are sufficiently homogeneous. In this article we propose a Bayesian method for incorporating past information based on dynamic power priors that tunes the degree of borrowing according to a posterior measure of compatibility between current and historical control data. Fre- quentist Type-I error and power properties of the methods are also discussed.