On the statistical model of the two-stage designs in bioequivalence assessment

Two-stage clinical designs are currently recommended by the regulatory authorities for the assessment of bioequivalence (BE). A specific statistical methodology was recently proposed by the European Medicines Agency (EMA). The aims of this article are to elaborate on the suggested statistical design from the EMA and to compare it with the existing statistical methods reported in the literature.

Monte Carlo simulations were used to simulate the conditions of a two-stage BE design. The starting sample size was either 24 or 48, whereas the coefficient of variation of the within-subject variability was equal to 20% and 40%. Several geometric mean ratio levels of the BE metric were considered. Under each condition, 1,000,000 studies were simulated.

The overall performance, in terms of percentage of BE acceptance, is identical. The additional term, 'sequence × stage', suggested in the EMA method is in most cases nonsignificant. The same results were obtained regardless of the type (fixed or random) of the effect applied to the 'subjects' term.

Any BE study either finished or in progress which relies on the existing literature methodology leads to the same percentage of BE acceptance as if it was analysed with the recently proposed EMA method.