Antoine Croxo 1,2, Julie Bertrand 1, Florence Loingeville 2
1 Iame, Inserm (Paris, France), 2 ULR2694 METRICS, University of Lille (Lille, France)
Introduction: In bioequivalence trials, regulatory authorities recommend testing average bioequivalence (ABE) between a test (T) and a reference (R) product with two one-sided tests from non-compartmental analysis (NCA-TOST) estimates of the area under the curve (AUC) and the maximal concentration (Cmax) obtained from rich pharmacokinetic (PK) sampling collected in a crossover design study [1,2].
However, in the case of highly variable drugs (HVDs), i.e. drugs in which the within-subject coefficient of variability (CV) is over 30% for at least one of the BE endpoints, these recommendations present significant challenges due to the widened confidence intervals around the estimates. To overcome the constraints of standard BE limits, reference-scaled average bioequivalence (RSABE) and NCA-TOST (RS-NCA-TOST) have been developed [3] to assess BE conditionally to the CV of the reference drug product and are now recommended by regulatory agencies [2,4].
On the other hand, a model-based two one-sided tests (MB-TOST) using nonlinear mixed-effects models was introduced in 2011 [5] which has been shown to address sparse data and difficult study settings [6]. Here, we propose and evaluate an RSABE extension of MB-TOST for HVD generics (MB-RS-TOST).
Objectives: To implement MB-TOST for RSABE, compare with NCA counterparts through simulations its statistical properties on rich PK data and evaluate through simulations its feasibility on sparse PK data.
Methods: Both RS-TOST and RS-MB-TOST following FDA (US) and EMA
(EU) guidelines were evaluated through a simulation study inspired from a BE trial for dasatinib, a HVD in oncology with 60% WSV on C[max]. The simulation follows a two-sequence, four-period fully replicated crossover study design with both rich (n = 19) and sparse (n = 6) sampling schemes.
Four simulation scenarios were implemented, one with C[max] T/R geometric mean ratio = 95% to estimate the power, and three scenarios with C[max] T/R geometric mean ratio = 80, 70 and 60% to assess the type I error for the classic ABE, and the RSABE following the EU and US guidelines, respectively.
Necessary sample sizes to achieve 80% power for BE testing in all four scenarios were computed with an iterative process using PFIM4.0 software for bioequivalence evaluation [7] (except for NCA- and MB-TOST).
Results: RSABE testing enabled around 3-fold reduction in sample sizes compared to classical ABE testing.
Under rich sampling, all tests controlled their type I error in their respective scenarios. In terms of power, RS-NCA-TOST and RS-MB-TOST showed comparable performance, with 82% and 83% estimates following the US guidelines and 76% and 84% estimates following the EU guidelines, respectively.
Under sparse sampling, all MB-TOSTs controlled their type I error with powers at 82% and 83% following the US and EU guidelines, respectively.
Conclusion: This work demonstrates that RS-MB-TOST may serve as a viable alternative for BE assessment of HVDs when only sparse sampling can be obtained.
References:
References:
[1] US FDA (2022), Statistical Approaches to Establishing Bioequivalence
[2] EMA (2010), Guideline on the investigation of bioequivalence
[3] B. M. Davit et al. (2012), The AAPS Journal
[4] US FDA (2011), Draft Guidance for industry: bioequivalence recommendations
for progesterone oral capsules
[5] A. Dubois et al. (2011), Statistics in Medicine
[6] F. Loingeville et al. (2025), Statistical Methods in Medical Research
[7] C. Dumont et al. (2018), Computer Methods and Programs in Biomedicine
Reference: PAGE 34 (2026) Abstr 11989 [www.page-meeting.org/?abstract=11989]
Poster: Drug/Disease Modelling - Absorption & PBPK