Bioequivalence and Bioavailability Forum

Ahoy!

❝ As an alternative to scaling we might in stead relax the alpha and keep accepting and rejecting the same products…

I disagree. The BE-problem is originally stated in terms of the maximum acceptable difference (Δ), where Δ is considered to be clinically not relevant. This Δ leads to the acceptance range (see the history lesson above).

If a a drug/drug product is highly variable it will be very difficult to demonstrate BE of the reference to itself given the con­ven­tio­nal Δ of 20%. On the other hand, practically all HVDs/HVDPs have a flat dose response curve (i.e., even a large Δ is not expected to be clinically relevant). Hence, it makes sense to widen the accep­tance range (either to a fixed value in some jurisdictions or scaled based on the s_wR in others).
If we modify the AR, we can still keep the patient’s risk. I know of only one regulation where the α is mo­di­fied. ANVISA requires for NTIDs the conventional AR but assessed by a 95% CI. IMHO, that’s a flawed approach. Yes, the risk will be lower. But which risk? To have 2.5% of the patient-population with a BA of <80% and 2.5% with >125%. Can be nasty for NTIDs. I don’t get the “idea”.

❝ I think the only reason not to formulate the problem (=acceptance vs rejection) in terms of alpha relaxation but rather in terms of widening of CI limits is that if we formulate it as a widening of alpha level then we need to admit explicitly what scaling involves: An occasional increase in type I error.

An inflated TIE could easily (!) be avoided by either pre-specifying a lower α (0.025 or 0.0304 depen­dent on the design) or adjusting α based on s_wR. Time to publish sumfink.
Even then, one big problem – common to all RSABE-methods – remains open. Products are approved according to different standards. Furthermore, the procedure is not reversible. In ABE we can recode T with R and the decision (pass/fail) will always be the same. This is not the case in RSABE if s_wR ≠ s_wT (of course we need a fully replicated design to show that).

I think that the two Lászlós in one of their papers mentioned that it would be desirable that agencies pool s_wR-data and publish a fixed widened acceptance range in product-specific guidelines. In such a case the Null-Hypothesis is no more modified “in face of the data” and no inflation of the TIE is pos­si­ble. No replicate designs, no fiddling around with crappy models. A conventional 2×2 would do the job (back to the roots).
This is what I hope for (ha-ha): RSABE as a kind of “transition state” until enough data are collected allowing regulators to come up with justified recommendations.
Will never happen. I know, I know.