Bioequivalence and Bioavailability Forum

Dear Rafimmc!

❝ 1. Do we have to calculate 95% only? or Calculate 90% first and if study fails, go for 95% and add on more subjects.

❝

❝ 2. We have two test products (i.e 3 way study). If the study fails, we want to take only one test product (with better profile) for the second stage with add on subjects. So the second stage would become two way. Now we have to combine two stages for final statistical analysis to calculate 95% CI. Please help me a way out!

I think that you are mixing two things up.
From your description of the design I would guess you are talking about an “Add-On design”, not a group sequential design.

Add-On designs are covered in some countries’ guidelines (Canada, Japan, South Africa), but IMHO are unacceptable for some other countries’ regulators (USA, Brazil).
Evaluation differs also slightly:
Canada:
If BE not shown (study 1), additional subjects are included (study 2).
F-test (equality of variances) p >0.05
‘Formulation-by-Study Interaction’ effect in ANOVA p >0.05
⇒ pooled analysis, no α-adjustment
Japan:
Sequential design not allowed; size of study 1 should be large enough to show BE based on assumptions (point estimate, residual variance), but:
If BE not shown (study 1), additional subjects are included (study 2), where size of study 2 ≥50% of study 1.
F-test (equality of variances) p >0.05
‘Study’ effect included in ANOVA, p >0.05
⇒ pooled analysis, no α-adjustment
South Africa:
No specific recommendation about statistical analysis, but:
“The provision for add-ons should be made in the protocol a priori clearly reflecting the maximum number of subjects to be included.”

To my knowledge currently only New Zealand (M$-DOC) allows for a sequential design (Section 15.5.5 Number of subjects):
“Sequential testing may also be acceptable for studies expected to require a large number of subjects, i.e. a study is conducted on a predetermined subset of the required sample and the intended statistical analysis is performed. If the acceptance criteria are met, no further subjects need to be tested. If the acceptance criteria are not met, the results from the first part of the study can be used to determine how many more subjects should be tested. Appropriate statistical tests (e.g. sequential t-test) which make allowance for this design should be used. The ethically justifiable maximum number of subjects should also be considered. The final statistical analysis then uses all of the data.”
Such an approach would call for an α-adjustment.

For details of sequential designs in the context of BE-testing see:

Gould LA. Group Sequential Extension of a Standard Bioequivalence Testing Procedure. J Pharmacokinet Biopharm. 1995;23(1):57–86. doi:10.1007/BF02353786

Maybe there’s light at the end of the Euro-tunnel, since in the current recommendation for an update of the NfG it’s stated at #3:
“Recommendations on BE in the current guidance will be updated with regard to […] under which circumstances a sequential design may be used”

Now to you questions:
#1: In an Add-On design you don’t adjust the α; in a sequential design you are penalized for the ‘first look’, the α-level at each level depends on the stagewise sample size ratio n₁:n₂, and on the strategy for implementing the design (93–94% nominal CIs are common to maintain the overall type I error 0.05).
#2: An Add-On should be possible, if you use a 6×3 Williams’ design (see this post); a group-sequential design would become a statistical nightmare. In any case you should definitely contact the regulator for advice first!