Blind or not, Indecisions or Decisions [Two-Stage / GS Designs]

Dear Jack,

❝ the reason I call this a one-stage design is that I would not formally test for equivalence at the time were one does the sample size re-estimation. Hence one would not need to spend any alpha at this time (as no testing is done and hence the EMA guide is not relevant) and hence no risk of the awkward situation that one could have to stop the trial early.

I wonder if the EMA will seeing this the same way. Especially claiming that the EMA guideline is not relevant .

❝ This is also the reason why doing the sample size re-estimation in a blinded fashion can make sense as using unblinded data will question the one-stage design.

The questions is why? Can you explain why an unblinded evaluation will have any influence?
All involved personal in an open study is unblinded. But the statistician should have a blindfold (this or better this one )?

❝ As for the question: "Can we expect to prove BE with a Nmax of 120 subjects with some pre-specified power?"

❝ I dont really see the problem there. You are not arriving at a test decision about BE if you decide to stop because you would need more resources than you have and hence it is not impacting your type-I-error.

Do you mean that the stopping the study due to exceeding Nmax is not a decision about BE? I have thought up to now that this is equal to the decision: Given the data of the 'internal' pilot we can not expect to reject the Null: Bioinequivalence with reasonable resources. Thus we have to stay with the Null (aka accept H0). In that direction I had understood hitting futility bounds in group sequential trials.
Or am I here totally wrong?

This is crucial in implementing a Monte Carlo simulation in which we would count 'BE' / 'not BE' to establish the type I and type II errors. As what shall I count hitting the futility criterion? 'Not BE' or a third answer? If you prefer a third, how to count them with respect to type I and type II errors (nominator/denominator)?

Regards,

Detlew