Bioequivalence and Bioavailability Forum

Hi John!

❝ ❝ Well, if the study was planned for 80% power, CV <50%, and a GMR of 1 you would need 21. With a lot of drop-outs… In a full replicate (CV 30, GMR 1) you need only 16.

❝

❝ 16 to finish I assume?

No, 16 planned. In the actual study you might pass with fewer by luck.

I tried to track the sample size issue down. ACPS Meeting, October 6, 2006,
Topic 3: ACPS Questions – Highly Variable Drugs presented by Barbara Davit:

• Question 1
  Does the committee agree with the use of a point estimate constraint when applying scaled bioequivalence? If yes, is the 80 – 125% limit on the point estimate appropriate?
  
• Question 2:
  We propose a minimum sample size of 36 subjects when evaluating the BE of highly variable drugs. Does the Committee concur?

Voting members of the ACPS were Charles Cooney, Carol Gloff, Meryl Karol, Arthur H. Kibbe, Melvin Koch, Marvin C. Meyer, Kenneth Moris, Cynthia Selassie, Marc Swadener. Outcome:
Question 1 (PE constraint): yes (8), no (0), abstain (1)
80 – 125% constraint: yes (2), no (3), abstain (4)
The Committee asked the FDA to rephrase Question 2 to:

We propose a minimum sample size of subjects when evaluating the BE of highly variable drugs. Does the Committee concur? If yes, what would be the minimum sample size?

yes (6), no (1), abstain (2)
Four members voted for a minimum sample size of 24, three members voted for a minimum sample size of 36, and two members abstained.
For the discussion see this part of the transcript.

At the August 5, 2009 meeting FDA’s Dale Conner presented an update. See slides 18–19:

Protocol for Reference-Scaled ABE Approach

• BE study uses a three-period, reference-replicated, crossover design with sequences of TRR, RTR, & RRT
  
• A four-period design is also acceptable (sequences of TRTR and RTRT)
  
• Usual pharmacokinetic sampling to determine Cmax, AUC(0-t), and AUC(0-inf)
  
• At least 24 subjects should be enrolled
  
• Reference replicate data analyzed for determination of σ_wr
  
• If σ_wr < σ_w0 then data analyzed using unscaled average BE method
  
• If σ_wr ≥ σ_w0 then data analyzed using scaled average BE and point-estimate criteria

(my emphasis)
Fascinating: Scaled BE limits given on slide 20.

Also interesting that in none of the API-specific guidances (first one progesterone, April 2010) a minimum sample size is mentioned.