Bioequivalence and Bioavailability Forum

Dear Imran!

❝ If the Intrasubject CV is more than 60%, we calculated the sample size of

❝ approximately 130 subjects.

Correct

❝ I understand that guidelines suggest for replicate design if Intrasubject

❝ CV is more than 30%.

❝ If I design replicate study it will take long time to complete (approx 4

❝ months).

What's the reason behind this time frame--the half life of the drug or the clinical capacity?

❝ Please suggest how to plan such studies.

If it's the long half life I would not suggest a replicate design, because the main assumption of bioequivalence is equality in clearances within treatments (and doses, of course):
 F1     AUC1 × D2 × CL2
---- = ----------------
 F2     AUC2 × D1 × CL1

therefore setting D1 = D2 and assuming (!) CL1 = CL2:
 F1     AUC1 × D2 × CL2
---- = ----------------
 F2     AUC2 × D1 × CL1

we finally get:
       F1    AUC1
Frel = --- = -----
       F2    AUC2

The assumption of equal clearances may be well justified in 2×2 cross-over studies of relatively short duration (actually many people are not even aware of this underlying nasty thing), but 4 months...
Maybe even by the conventional 2×2 study you are going to shoot yourself in the foot (you cannot differentiate between variability originating in treatment/drug or in clearance).
In a replicate design of such a long duration you will see a lot of drop outs (a simple flu would suffice). The statistical analyes will be awfully complicated (even for complete data sets they are not easy anyway).
What's the total variance (intra+inter) of your drug?
Perhaps you should consider going fo a parallel design (yes, I know, even more subjects, but no washout): it will keep your clinical site busy.

If you are limited by the clinical capacity (not the half life), you should opt for another CRO.
Regulators don't like 'blurred' treatment periods; generally it's acceptable to split treatment groups in two, separated by about one week, but logistics become quite nasty. Sometimes you even have to include an additional parameter 'group' in your statistical model, and don't ask me what to do if you get a significant effect...