|
Abstract:
|
In clinical studies with co-primary hypotheses (H1) that need to be established simultaneously on multiple endpoints Y, correlation r among Y should be taken into account in overall power calculation. However, r is not always considered in practice, due to either no data for estimating r or no convenient software packages for incorporating r in power calculation. A naïve way to bypass the problem is to calculate power for individual H1, the product of which is then used as the lower bound (LB) of the overall power assuming nonnegative r. We show that positive r among Y do not necessarily lead to greater overall power compared to the Simple Product Of IndividuaL Power (Spoil-P), and that study designs and the relative locations among Y are also important factors to consider before using the Spoil-P as the LB for the overall power.
|
