All Times EDT
Keywords: Bioequivalence, multivariate bioequivalence, Two one-sided test
While generally accepted methods to statistically assess bioequivalence between a generic treatment and a reference medical product are laid out rigorously by the European Medicines Agency (EMA), they are typically aimed at problems in the univariate framework. They generally consist of assessing whether a confidence interval of the difference between the two treatments is contained within a predetermined bioequivalence region. In the dermatological field, Quartier et al. 2019 recently developed a novel methodology which jointly monitors a drug’s concentration in many targeted areas of the skin. In this higher dimensional framework, an application of the EMA’s guidelines designed for the univariate framework at each of the monitored areas is problematic, as the likelihood under the null of exiting bioequivalence region at least once solely due to sampling (type-I error) increases with the number of dimensions, a known consequence of multiplicity. Correcting the latter with conventional methods such as the Holm-Bonferroni or the Šidák correction is insufficient given the usually complex underlying dependence structure in this high dimensional framework. In this paper, we develop a new methodology that aims at expanding bioequivalence limits to control for the desired type-I error rate using a matching principle, generalising the state-of-the-art method in the statistical assessment of bioequivalence in the univariate setting, known as the Two One-Sided Test, to the multivariate framework. Besides coverage precision, our methodology also provides the best compromise between power and type-I error, largely dominating other existing multivariate methods consistently for any sample size and number of dimensions.