Bioequivalence of ANDA Data using a Non-Informative Bayesian Procedure (BEST) Compared with the Two One‑Sided t‑Tests (TOST).

Abstract

The regulatory statistical standard for evaluating average bioequivalence (BE) in generic drug testing, formulation bridging, and 505b2 drug comparisons has traditionally employed the two one-sided t-tests (TOST) procedure. A comparison of BE determinations of TOST and a t-distribution-based, non-informative Bayesian procedure (Bayes_T) was conducted on 2341 pharmacokinetic parameter datasets in 678 anonymized abbreviated new drug applications (ANDAs) to assess the influence of deviations from lognormality and the presence of extreme values. This research has been motivated to assess reliability of statistical inference procedures for accurate and fair regulatory assessments of BE and non-BE (NBE). The BE criterion of 90% confidence (CI) or Bayesian credible (CrI) intervals of log test/reference ratios for TOST and Bayes_T was 0.80-1.25. TOST. Bayes_T agreed on BE conclusions in 98.9% of cases. There were 20 disagreed cases in which TOST rejected BE and Bayes_T concluded BE, wherein all cases failed the lognormality test and 17 of which contained extreme values (4.2% of 409 cases that contained extreme values). In this circumstance, TOST can be overly conservative in the presence of extreme values. There were 7 cases in which TOST concluded BE at outer BE bounds, while Bayes_T marginally rejected BE, despite these cases successfully passing the lognormality test. While TOST remains a widely accepted method for BE assessment, the presence of extreme values and deviations from lognormality may lead to faulty inference of BE. The Bayes_T methodology offers an alternative approach to TOST that can be prespecified to assess BE in such scenarios. Pre-specified application of the Bayes_T procedure may ensure more reliable outcomes in regulatory assessments of BE.