Hi heaven sent,
interesting nick.
❝ Do I understand correctly that Reference scaling is the same design as Partial Replicate:
❝ RRT, RTR, TRR ?
No, you don’t.
- Reference-scaling is a method assessing bioequivalence.
- The partial replicate is a design for it.
Though you need a replicate design for reference-scaling,
this partial replicate is only one of them. In general I don’t recommend partial replicates since the statistical model behind is tricky (especially in the case of
heteroscedasticity: CV
wT ≠ CV
wR). Better to opt for one of the fully replicate designs. Most commonly used are TRTR | RTRT (4 periods) and TRT | RTR (3 periods).
- 4 period (full) replicates
- TRTR | RTRT
- TRRT | RTTR
- TTRR | RRTT
- TRTR | RTRT | TRRT | RTTR 1
- TRRT | RTTR | TTRR | RRTT 1
2 period (full) replicate3 period (full) replicates3 period (partial) replicates- TRR | RTR | RRT
- TRR | RTR 3
Notes:
- The FDA requires at least 24 dosed subjects if the study is intended for reference-scaling.
The EMA requires at least 12 eligible subjects in the sequence repeating R of the 3 period full replicate designs.
- In any of the designs conventional (unscaled) average bioequivalence ABE can be evaluated as well (e.g., if CVwR <30% or reference-scaling for a particular PK metric is not acceptable).
- The statistical models differ between regulations.
- FDA (model RSABE: reference-scaled average bioequivalence).
- EMA (model ABEL: average bioequivalence with expanding limits) – also adopted by the WHO, Eurasian Economic Union, Australia, New Zealand, Brazil, Egypt, South Africa, ASEAN States.
Cmax and some PK metrics of MR-products. WHO: pilot phase for AUC; full replicate design mandatory.
- Health Canada (ABEL but potentially wider limits). Only AUC.
- Gulf States, Mexico: ABE but wider acceptance limits for Cmax (75–133%).
Only for jurisdictions which have adopted ABEL and the Gulf States / Mexico you have to provide a clinical justification that expanding the acceptance limits imposes no risk (safety/efficacy) on patients.
Conventional sample size estimation is not possible – you need to perform simulations. I recommend the package PowerTOST
for the statistical software R (open-source and free of costs).
- Confounded effects (design not recommended).
- Balaam’s design (not recommended due to poor power characteristics).
- Extra-reference design (biased in the presence of period effects; design not recommended).