[Application of multilevel models in the evaluation of bioequivalence (II).]

The main purpose of this paper is to explore the applicability of multivariate multilevel models for bioequivalence evaluation. Using an example of a 4 x 4 cross-over test design in evaluating bioequivalence of homemade and imported rosiglitazone maleate tablets, this paper illustrated the multivariate-model-based method for partitioning total variances of ln(AUC) and ln(C(max)) in the framework of multilevel models. It examined the feasibility of multivariate multilevel models in directly evaluating average bioequivalence (ABE), population bioequivalence (PBE) and individual bioequivalence (IBE). Taking into account the correlation between ln(AUC) and ln(C(max)) of rosiglitazone maleate tablets, the proposed models suggested no statistical difference between the two effect measures in their ABE bioequivalence via joint tests, whilst a contradictive conclusion was derived based on univariate multilevel models. Furthermore, the PBE and IBE for both ln(AUC) and ln(C(max)) of the two types of tablets were assessed with no statistical difference based on estimates of variance components from the proposed models. Multivariate multilevel models could be used to analyze bioequivalence of multiple effect measures simultaneously and they provided a new way of statistical analysis to evaluate bioequivalence.