2-stage design with interim sample size estimation [Two-Stage / GS Designs]

posted by d_labes  – Berlin, Germany, 2010-10-26 13:28 (4105 d 09:42 ago) – Posting: # 6080
Views: 12,366

(edited by d_labes on 2010-10-26 13:45)

Dear Jack,

» this is quite a complex question ...

Sure! :yes: Otherwise I had not asked :-D.

» The first and most important question in my mind is why not consider a single stage design with a sample size review at interim? So basically not allowing for the study to stop early (besides the sponsors constraint).

This was the intention of the design described. But why do you call it single stage?

The reason behind it was that for some constituent (its a FDC) the variability is not known at all. Each assumption about is only a delusion. Not to waste resources the pilot, usually performed in such circumstances, should included in the study (sometimes called internal pilot).

The 'extreme' alpha to spend at stage 1 is only for compliance with the new EMA guidance which calls definitely for it, statistical sound or not. Just to cite page 16 to 17:
"Two-stage design
... If this approach is adopted appropriate steps must be taken to preserve the overall type I error of the experiment and the stopping criteria should be clearly defined prior to the study. The analysis of the first stage data should be treated as an interim analysis and both analyses conducted at adjusted significance levels (with the confidence intervals accordingly using an adjusted coverage probability which will be higher than 90%) ..."

» The second choice one needs to make is if the sample size review is going to be blinded or not (note that I am ignoring any knowledge about a possible shift in point estimates).

I'm not sure about this. Usually BE studies are performed with administration of the study medications in an open fashion. Why then make a blinded evaluation at interim?

The interim sample size estimation is highly influenced by a shift in the point estimator. Especially the futility bound: Can we expect to prove BE with a Nmax of 120 subjects with some pre-specified power?
An a-priori difference in Cmax is expected due to the nature of the study products. Not considering it is like "Mit offenen Augen gegen eine Wand fahren" (to drive against a wall with open eyes).

I'm on the same point of view as Helmut above. All I've read in the meantime is for group-sequential or adaptive designs for superiority trials with parallel groups, or at best some rare papers for non-inferiority.
Since the role of the Null and the Alternative are reversed in equivalence trials compared to superiority it is likely that the results obtained do not apply.

Nevertheless, thanks for your input.



Complete thread:

 Admin contact
21,835 posts in 4,569 threads, 1,554 registered users;
online 6 (0 registered, 6 guests [including 4 identified bots]).
Forum time: Friday 22:10 CET (Europe/Vienna)

No problem can stand the assault of sustained thinking.    Voltaire

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz