Bioequivalence and Bioavailability Forum

Hi cher0424 & Tarak,

I agree with what ElMaestro wrote about study planning etc.

❝ […] but the different between two batch should be within 5%, so particularly no much influence in trial result. Unless the test product quality happen on the edge of 80%-125%. In that case the differences between two batch are essential to pass or fail.

Let’s play the devil’s advocate. CV_intra 25%, CV_total 50%, expected T/R 95%, 2×2 crossover study planned for ≥80% power. We would design the study with 28 subjects (expected power 80.7%). Let’s further assume that both the measured content of the first batch of the reference (R1) and of the test (T) were 100%. The true – though unknown – bioavailability of T is 95%. You try hard to find a second batch of the reference which matches R1 but the only one you could get had a measured content of 95% (within ±5% of both R1 and T).
If you evaluate only the first period as a parallel design (bad idea, since Freeman* showed that you don’t get unbiased estimates if the study was planned a s a crossover):

T/R1  95.00% (90% CI: 60.57–148.99%)

Glorious failure since you are extremely underpowered with 14 subjects for a CV_total of 50%.
Same story if you look only at the second period:

T/R2 100.00% (90% CI: 63.76–156.84%)

If you pool both periods you pass easily…

T/R   97.47% (90% CI: 87.12–109.05%)

…but through the backdoor (i.e., by introducing R2) shifted the result towards nicer results. If we would have had enough R1 for the entire study we would have got:

T/R1  95.00% (90% CI: 84.91–106.28%)

The CV employed in study planning was an assumption. What if it would be larger, say 40%? T/R1 (as planned) would fail with a lower CL of 79.70% but pass with T/R (lower CL 81.77%). Very bad idea. No regulator would buy that – for good reasons.

---

• Freeman P. The performance of the two-stage analysis of two-treatment, two-period cross-over trials. Stat Med. 1989;8(12):1421–32.