Bioequivalence and Bioavailability Forum

Dear Cmax!

❝ Can any one suggest how we can calculate sample size for a partial replicate study with scaled approach... considering CV 45% in normal cross over we may have to do with 100+ subjects...

Sounds like you are aiming at T/R 0.95 with 90% power? Would need 110 subjects (and 82 for unscaled partial replicate).

❝ what would be the sample size if we propose a partial replicate...

In theory RSABE should adjust the acceptance range according to the intra-subject variability of the reference. In other words, power (and therefore the sample size) is independent from CV. For details see:

Tóthfalusi L, Endrényi L and A García-Arieta
Evaluation of Bioequivalence for Highly Variable Drugs with Scaled Average Bioequivalence
Clin Pharmacokinet 48/11, 725-743 (2009)
online Abstract

Fig. 3. Power curves of simulated three-period crossover studies evaluated by average bioequivalence (ABE) and by scaled ABE (SABE) with two different regulatory constants. Three-period test-reference-test/reference-test-reference bioequivalence trials were simulated with 36 subjects and different within-subject variabilities. (a) Coefficient of variation (CV) = 35%. (b) CV= 45%. (c) CV= 55%. The simulated bioequivalence studies were evaluated using ABE with the conventional criterion and by SABE with two regulatory cutoff values. The regulatory cutoff sigma_S = 0.760 was proposed by Tothfalusi and Endrenyi, whereas the US FDA suggestion is 0.893. In both cases, an additional criterion was applied, which constrained the point estimate of the geometric mean ratio to values between 0.80 and 1.25. These figures show the percentage of crossover studies that successfully declare bioequivalence as the true geometric mean ratio gradually increases.

Notice that for the same sample size power drops in the conventional (unscaled 2×2 cross-over) ABE design, but is only fairly affected in the scaled methods. The sample size for a unscaled partial replicate is 75 % of a 2×2 Xover.

Now for the pitfalls:

• You can use any sample size in the scaled approach. There was a discussion at the FDA, whether a minimum sample size should be given in guidances. FDA suggested 32, but was voted down. According to a presentation by Barbara Davit given last May in Ljubljana FDA requires 24 subjects.
  
• At higher CVs (> ≈50 %) the restriction of the GMR (80%-125%) takes over - scaling according to the sigma_WR might not save you.
  
• In the EU, scaling is only allowed for C_max and up to a CV of 50 %.
  
• Again in the EU you have to demonstrate that a CV >30 % is not a result of outliers. Since there is no consensus which method of outlier testing should be used you are skating on thin ice. The larger the group is, the higher the chance of outliers. So think it over.