Comparing a Bayesian Approach (BEST) with the Two One-Sided t-Tests (TOSTs) for Bioequivalence Studies

Abstract

The two one-sided t-tests (TOST) procedure has been used to evaluate average bioequivalence (BE). As a regulatory standard, it is crucial that TOST distinguish BE from not-BE (NBE) when BE data are not lognormal. TOST was compared with a Bayesian procedure (BEST by Kruschke) in simulated datasets of test/reference ratios (T/R) which were BE and NBE, wherein (1) log(T/R) or T-R were normally distributed, (2) sample sizes ranged 10–50, and (3) extreme log(T/R) or T-R values were randomly included in datasets. The 90% “credible interval” (CrI) from BEST is a Bayesian alternative of the 90% confidence interval (CI) of TOST and it can be derived from a posterior distribution that is more reflective of the observed mean log(T/R) distribution that often deviates from normality. In the absence of extreme T/R values, both methods agreed BE when observed T/R were lognormal. BEST more accurately concluded BE or NBE, while requiring fewer subjects, when observed log(T/R) or T-R were normal in the presence of extreme values. Overall, TOST and BEST perform comparably on lognormal T/R, while BEST is more accurate, requiring fewer subjects when datasets are normal for T-R or contain extreme values. Of note, the normally distributed datasets only rarely contain extreme values. Our results imply that when BEST and TOST yield different BE assessment results from bioequivalent products, TOST may disadvantage applicants when T/R are not lognormal and/or include extreme T/R values. Application of BEST can address the situation when T/R are not lognormal or include extreme data values. Application of BEST to BE data can be considered a useful alternative to TOST for evaluation of BE and for efficient development of BE formulations.