At the recent JSM in Chicago, Cytel’s Sam Hsaio and Lingyun Liu alongside Genentech's Romeo Maciuca, presented a framework for inference in adaptive bioequivalence trials with unblinded sample size re-estimation.
In bioequivalence trials where the variance is often unknown, and the sample size small, using boundaries derived under the assumption of a normally distributed test statistic may lead to type I error inflation. This problem can be overcome with p-value combination methods, however these approaches generally do not directly provide confidence intervals for the geometric mean ratio on the scale of the original pharmacokinetic endpoint.
Hsaio and Liu propose a simple adjustment to the CHW method (Cui, Hung and Wang). This adjusted approach involves pre-specifying a range of final sample sizes to allow some flexibility in the SSR procedure, yet uses pre-defined constant boundaries based on a "piecewise t-distribution" to derive repeated confidence intervals (RCIs) for the treatment effect. The RCIs have guaranteed coverage, and can be used for inference and clinical interpretation in the same way that conventional two-sided confidence intervals are typically used when applying the two one-sided testing (TOST) procedure.
In the presentation, they work through a hypothetical example of a parallel two stage design, unblinded sample size re-estimation study designed to demonstrate average bioequivalence of treatment vs reference product.