In evaluating bioequivalence, a sparse crossover design is used for the pharmacokinetic (PK) bioequivalence study. In such studies, the concentration sample is collected at a single time point from each subject. We can obtain only one value (averaged over subjects) for Area under the Curve (AUC) at time t from the study. Hence, a traditional PK analysis cannot be used.
We examined several non-parametric bootstrap methods and a parametric method for sparse crossover PK study. The 90% confidence interval for the ratios of mean AUC for these sparse crossover designs is constructed differently from traditional PK studies. We did a series of simulations under different scenarios to compare several bootstrap methods - including the bias corrected and accelerated (BCa) method and different stratified bootstrap strategies, and a parametric approach that incorporates Fieller's method.
The simulation studies show that the power and coverage probability are similar for different bootstrap methods and the parametric method. We recommend using the parametric approach with Fieller's method to construct the confidence interval for AUC.
|