Controlling type I errors for two-stage bioequivalence study designs

Abstract

Recent publications have introduced sequential (two-stage) designs for investigation of bioequivalence in cross-over trials and the occurrence of type I errors have been investigated using trial simulations. They have so far focused on Test:Reference ratio of 0.95 and 0.9 and it has been observed that type I errors above 0.05 may occur. In this paper the behavior of two-stage designs for investigations of bioequivalence is being further investigated, and it is concluded that the existing methods do not universally control type I errors to an acceptable level, and they have the disadvantage of not using the observed Test:Reference from stage 1 in the dimensioning of stage 2. It is observed that type I errors and the alpha at the second stage follow an approximately linear relation in the region of type I errors of 0.05. This principle is used to propose a method that uses both the CV and Test:Reference ratio and which targets specifically a type I error level of 0.05. This is done by using simulation after stage 1 to identify an optimal value for the second alpha. An example is given which illustrates how the method may be associated with both an ethical and economical advantage.