The analysis of bioequivalence with respect to TMAX under a 2 x 2 crossover design

We develop population-averaged and subject-specific continuation ratio logit models for assessing bioequivalence of two formulations with respect to TMAX under a 2 x 2 crossover design. We propose generalized estimating equations and mixed-effects estimators for obtaining confidence intervals for odds ratios representing population-averaged and subject-specific formulation effects. We show analytically that the population-averaged approach leads to asymptotic intervals that are narrower than the subject-specific intervals when the intrasubject correlation is positive. Hence, asymptotically, it is easier to show bioequivalence with respect to TMAX under the population-averaged approach than under the subject-specific approach. We illustrate these results with an analysis of a multiple-dose bioequivalence study.