Bioequivalence and Bioavailability Forum

Hi John!

❝ Does FDA (or EMA) really care about the power of a study?

I don’t think so. FDA says “The study should have 80 or 90% power to conclude BE between these two formulations.” – not “must have”. EMA requires only an “appropriate sample size estimation” – whatever this might be. Since lower power translates into higher producer’s risk agencies shouldn’t care (their job is to deal with the patient’s risk). It might be that the IEC/IRB will be reluctant to approve a study given an expected 30% chance of failure. The IEC/IRB is concerned about the safety of subjects in the study – you can only try to convince them. Might be difficult for a drug with a nasty safety profile – which on the other hand is unlikely for an HVD/HVDP.

❝ Example, hypothetically speaking, lets say I set up a study with a sample size of xx (yes n>12) based on a power of 70% and the study passes. Does it matter that my power is less than 80%?

No; a posteriori power is irrelevant. BTW, AFAIK not stated anywhere explicitly IMHO FDA wants at least 24 subjects.

❝ My recent pilot study (3-way partial rep, n=60, withinsubject CV of ref = 55-60%) ended up with a (Highest) ratio of 1.2 and 95% upper confidence limit of -0.12456 ⇒ Tada, it passed BE.

Hhm, why don’t you save the money (or wire somefink to my bank account instead) and submit the pilot study as pivotal evidence of BE? See III.A.2. Pilot study

“A pilot study that documents BE can be appropriate, provided its design and execution are suitable and a sufficient number of subjects (e.g., 12) have completed the study.”

❝ We will move to a pivotal study, the formulation will be identical with no potential variabilities (So if there is then it's not my fault!).

Or the GMR and/or CV go loony and jump around between studies. Wouldn’t surprise me with a CV of 60%…

❝ If I follow Endrenyi et al's paper on sample size for HVD then I will be looking at n~200 subjects for a 3-way partial replicate study at 80% power. For a 4-way full replicate study at 80% power, n~120.

Yep. PowerTOST comes up with 201 and 134. I would really forget the partial replicate – both for pilots and pivotal studies. In many cases CV_WT < CV_WR. If this is the case (but from a partial replicate you don’t get the information) you will be rewarded. Example: 2×2×4 replicate, CV_WT=CV_WR=60%, n=134. For CV_WT 40%, CV_WR 60% you need only 100 (~25% less subjects). Try:
sampleN.RSABE(targetpower=0.8, theta0=1.2, CV=c(0.4, 0.6), design="2x2x4", details=F)

❝ Management said no go.

Power. That which statisticians are always calculating
but never have. Stephen Senn

❝ For a compromise I think we should be able to go with n=80. Based on R, with a ratio=1.2 & CV=0.6, the power will be ~70%.

Yessir.
power.RSABE(theta0=1.2, CV=0.6, n=81, design="2x3x3", details=F)
power.RSABE(theta0=1.2, CV=0.6, n=80, design="2x2x4", details=F)

❝ I am curious if the agency would care about this power business...

Me too. I’m not a regulator. Some thoughts above.