This paper introduces a two-stage bioequivalence design involving the selection of one out of two candidate formulations at an initial stage and quantifies the overall power (chance of ultimately showing bioequivalence) in a range of scenarios with CVs ranging from 0.1 to 1. The methods introduced are derivates of the methods invented in 2008 by Diane Potvin and co-workers (Pharm Stat. 7(4): 245-262, 2008). The idea is to test the two candidate formulations independently in an initial stage, making a selection of one of these formulations basis of the observed point estimates, and to run, when necessary, a second stage of the trial with pooling of data. Alpha levels are identified which are shown to control the maximum type I error at 5%. Results, expressed as powers and sample sizes, are also published for scenarios where the two formulations are far apart in terms of the match against the reference (one GMR being 0.80, the other GMR being 0.95) and in scenarios where the two test formulations have an actual better match (one GMR being 0.90, the other GMR being 0.95). The methods seem to be compliant with wording of present guidelines from EMA, FDA, WHO, and Health Canada. Therefore the work presented here may be useful for companies developing drugs whose approval hinges on in vivo proof of bioequivalence and where traditional in vitro screening methods (such as dissolution trials) may have poor ability to predict the best candidate.
Keywords: bioequivalence; power; screening; type I error.