Bayesian Analysis of Population Bioequivalence Using the Independence Chain Algorithm in PROC MIXED

The statistical test for population bioequivalence given in the current FDA guidance document on this matter ignores the dependence between summary statistics. It is very conservative in cases where there is a high degree of correlation of subject responses between test and reference formulations, which is usually the case in bioequivalence studies. Adapting the ideas of Kass and Wolfinger (2000), we use the PRIOR statement in PROC MIXED to approximate the distribution of the population bioequivalence parameter ΘPBE, from which we can construct tests based either on the probability of the upper tail region (i.e., Pr(ΘPBE>Θ0)), or on the one-sided upper 95% confidence bound of ΘPBE,. We show that this test can conclude population bioequivalence when the FDA method does not. We also examine the statistical properties of this approach. The Kass and Wolfinger approach is also compared to another alternative given by McNally et al which is also more powerful than the FDA method, but is slightly anti-conservative.