Bioequivalence and Bioavailability Forum

Dear rafimmc,

why didn't you give us the sample size of the pilot study? OK, I make an educated guess: 18?

❝ Cmax 101.38%(78.76-130.49); Intrasubject CV 45.5%; Power 42.10%

❝ AUC0-t 94.66% (76.05-117.84);Intrasubject CV 39.0%; Power 51.31%

❝ AUC0-inf 95.16% (78.44-115.43);Intrasubject CV 34.1%; Power 60.50%

I don't think a-posteriori power does make any sense; but that's another story; see e.g.,

Hoenig JM, Heisey DM. The Abuse of Power. The Pervasive Fallacy of Power Calculations for Data Analysis. Am Stat. 2001;55(1):19–24. doi:10.1198/000313001300339897. free resource.

❝ Sample size for pivotal study is found to be 64.

Hhm, based on which expected point estimate?
For C_max I get 64 subjects for GMRs of (100+2)%, acceptance range 80%-125%, 80% power.
Which such a high variability, you must not forget that both values gained form the pilot study are uncertain:
- the point estimates
- the intra-subject CVs

Conventionally in sample size estimation we would use a deviation of +5% from the reference as a minimum.
But it should be noted, that the small deviation of test from reference observed in the pilot may only be due to simple chance!

But even if you strongly belief in your formulation, the high variability remains problematic.

Calculating the upper (one-sided) 95% confidence limit of the CV of 45.5% for C_max one gets 67.8% calling for 142 subjects (staying with your +2% expectation. For +5% n=196!
For details of the calculation based on the Chi²-distribution see Chow & Liu (2000),* but use a one-sided 95% CL instead (they give a calculation for the two-sided CI).
Here we see one nasty property of power curves:
The upper CL (67.8%) of the CV is 149% of the estimate (45.5%), but the sample size (142) is 222% of the value (64) we would calculate from the estimate (z-axis: upper 95% CL as percentage of the estimated CV from the pilot study, x-axis: sample size, and y-axis: CV from the pilot study).

❝ Can we expect the pivotal study would clear the product or to recommend to

❝ reformulate the product?

Both answers are negative. I think that your product is fine, your difficulties will come from the high variability making it quite difficult to demonstrate BE.
You may opt for a reference scaled average bioequivalence approach (SABE); which has some chances of acceptance in the US, but is almost futile according to recent developments in the EU.
I wouldn't go any further without a scientific advisory meeting with the regulator.

* S-C Chow and J-p Liu
  Design and Analysis of Bioavailability and Bioequivalence Studies
  Marcel Dekker, New York (2^nd ed. 2000), pp 187-219