|
Abstract:
|
Replicated crossover designs (e.g., TRR, RTT) are sometimes used to demonstrate average bioequivalence of a test (T) and reference (R) treatment. Pharmacokinetic responses from such designs (e.g., log(AUC)) are commonly analyzed using a linear mixed effects modeling approach described in an FDA guidance document on bioequivalence analyses. For the "FDA approach", in PROC MIXED terminology, the covariance matrix for each subject (= ZGZ'+R) is modeled using both a RANDOM statement, with TYPE = FA0(2) for G, and a REPEATED statement, with TYPE = VC for R. A simpler approach is to only use a REPEATED statement, with TYPE = CS (with or without the EMPIRICAL option) or UN for R. The type I error rate and power properties of the different covariance structure-modeling approaches will be compared using simulations.
|
