For highly variable drugs (within-subject CV of roughly 30% or more), fixed 80.00–125.00% limits can demand impractically large studies, even when test and reference are equally good. Reference-scaled approaches address this by scaling the acceptance limit to the variability of the reference product itself.
When the FDA approach applies
Under the FDA's RSABE method, scaling is used when the within-subject standard deviation of the reference, sWR, is at least 0.294 (about 30% CV). Below that, standard average bioequivalence applies.
The design it needs
Estimating sWR requires the reference to be given at least twice to each subject, so RSABE studies use a replicate design:
- Partial replicate (three-period, e.g. TRR / RTR / RRT)
- Full replicate (four-period, e.g. TRTR / RTRT)
The criterion
The test is based on the linearised criterion
(μT − μR)² − θ · σ²WR ≤ 0, whereθ = (ln 1.25)² / 0.25² ≈ 0.797
Bioequivalence is concluded when both conditions hold:
- The 95% upper confidence bound of the criterion (Howe's approximation) is ≤ 0.
- The point estimate of the geometric mean ratio lies within 80.00–125.00%.
How EMA differs (ABEL)
| FDA RSABE | EMA ABEL | |
|---|---|---|
| Trigger | sWR ≥ 0.294 | CVWR > 30%, clinically justified |
| Parameters | Cmax and AUC | Cmax only (AUC by exception) |
| Limits | Implied by scaled criterion | Widened 90% CI limits, capped at 69.84–143.19% (CV 50%) |
| Point estimate | 80.00–125.00% | 80.00–125.00% |
Practical tips
- Pre-specify the method, the software and handling of outliers in the SAP before unblinding.
- Check that the reference variability is not inflated by a few subjects; report it with and without them.
- Validate the mixed-model or within-subject computation against a published reference dataset.