Inflation of the type I error: investigations on regulatory recommendations for bioequivalence of highly variable drugs.

PURPOSE: We investigated different evaluation strategies for bioequivalence trials with highly variable drugs on their resulting empirical type I error and empirical power. The classical 'unscaled' crossover design with average bioequivalence evaluation, the Add-on concept of the Japanese guideline, and the current 'scaling' approach of EMA were compared.
METHODS: Simulation studies were performed based on the assumption of a single dose drug administration while changing the underlying intra-individual variability.
RESULTS: Inclusion of Add-on subjects following the Japanese concept led to slight increases of the empirical α-error (≈7.5%). For the approach of EMA we noted an unexpected tremendous increase of the rejection rate at a geometric mean ratio of 1.25. Moreover, we detected error rates slightly above the pre-set limit of 5% even at the proposed 'scaled' bioequivalence limits.
CONCLUSIONS: With the classical 'unscaled' approach and the Japanese guideline concept the goal of reduced subject numbers in bioequivalence trials of HVDs cannot be achieved. On the other hand, widening the acceptance range comes at the price that quite a number of products will be accepted bioequivalent that had not been accepted in the past. A two-stage design with control of the global α therefore seems the better alternative.