Bioequivalence and Bioavailability Forum

Hi Shuanghe!

❝ Assessment of linearity will consider whether differences in dose-adjusted AUC meet a criterion of ± 25%.

❝ However, for this product, only 2 strengths exist. x and 2x mg. I guess power model doesn't make any sense here.

Two values, two parameters, zero dfs.

❝ The CRO send us a protocol based on standard 2x2 BE design with standard 80-125% BE criteria for 90% confidence interval of Cmax, AUCt and AUCinfinity to conclude dose proportionality.

❝ 1. what does +-25% means in the guideline mentioned above? in term of log-transformed or normal scale?

❝ In normal scale, if one dose gives AUC= 100, then the AUC of the other dose (adjusted by dose) should be within 100 +- 100*0.25 = 75 to 125?

I think, that’s what EMA means. They are talking about reliable sources (articles in peer-review journals, maybe the SmPC, …), not actually studies, IMHO. In the literature you have only means (quite often arithmetic ones).

❝ In log-scale as in BE, 80-125% criteria corresponds log-transformed difference +-0.22314... (so 100*exp(+-0.22314...) will give 80-125%. so does this mean we can use 0.25 instead and back-transform to normal value as interval of 100*exp(+-0.25) which is 77.88 - 128.4%? That would be weird.

Let’s start with the usual δ = 0.2; θ₁ = 1 – δ and θ₂ = θ₁^-1 which gives 80–125%. OK, but why not setting θ₂ = 1 + δ and θ₁ = θ₂^-1 which would give 83.33–120%?
Answer: History/convention/simplicity. In the transition period (where log-transformation became the method of choice) some people suggested a δ of 0.1802, which would give a symmetric acceptance range of 81.98–121.98% which is as 40% wide as the ‘old’ 80–120%.^*

I don’t think that EMA seriously considered the log-story (-δ 75-133.33%, +δ 80–125%). Furthermore this ±25% in connection with literature data IMHO calls only for a comparison of means (no CI).

❝ 2. I think AUCt alone should be used as criteria (difference of Cmax may obtained for additional information purpose but not as a criterion, AUCinf is not a required parameter in Europe.) as indicated by guideline and also suggested by Martin in other post.

Agree.

❝ 3. In general, how would you design the study if I may ask?

I would design it as the CRO suggested and assess dose-normalized AUC_t. I’m not sure whether EMA accepts this sloppy approach (±25% of means) in an actual Xover study. My gut feeling tells me that they want to see the 90% CI with 90-125%. I never performed a DP study with less than three strengths; in one case (before the current GL!) I was asked to provide Bonferroni-corrected CIs. BTW, in my studies I employed the stepwise approach given in by Chow/Liu (Ch. 18.3).

---

• Actually this is a delusion. p[-∞,+∞] of the normal distribution = p]0,+∞] = 1 of the lognormal.

---

P.S.: Have you noticed this goodie?