Bioequivalence and Bioavailability Forum

Hi NewInPK (nice nick...)!

❝ Thank you for maintaining this forum, it is very helpful.

Thanks.

❝ After a few sort or regular courses and a lot of googling, [...]

Consider the old-fashioned way: reading! Some textbooks are given in this post.

❝ 1) Text in WinNonlin help menu: "Measures for each individual should be

❝ displayed in parallel for the formulations tested. In particular, for each

❝ BE measure the ratio of the individual geometric mean of the T product to

❝ the individual geometric mean of the R product should be tabulated side by

❝ side for each subject."

Your quote from the online-help starts with: "This is in response to the FDA guidance"...
Pharsight refers to FDA's guideline from 2001, which states this sentence in http://www.fda.gov/cder/guidance/3616fnl.pdf Section 4.A.2; but it is nonsense anyhow. So far about guidelines.

❝ 2) When using the BE wizard in WinNonlin, in the "Bioequivalence Fixed

❝ Effects" window, what does "Fixed Effect Confidence Level" refer to? For a

❝ crossover (1 treatment and reference) what are the fixed effects?

You see them below 'Model Specification'; for a 2×2×2 cross-over they are: sequence, treatment, and period. They are fixed, because they are given by the design of the study.
Since we want to infer from our study (a random sample) to the general population, in the next window you find the random effect: subject(sequence).

❝ 3) [...] but when looking at the WinNonlin output, I don't find a

❝ confidence interval of the difference, instead I assume the CI of the

❝ ratio is listed.

Look at 'Depent Variables Transformation' in the BE wizard which defaults in WinNonlin to ln(x). Since many biological metrics are skewed to the right this transformation makes sense (apart from this empiric observation there are pharmacokinetic and bioanalytical reasons as well). Therefore the ANOVA (or LME or any other parametric statistical method) is done on the differences of the logs (which are the ratios of untransformed values, since ln(x)-ln(y)=x/y).

❝ My guess is that I interpret to have bioequivalence if the value of the

❝ Ratio[% Ref] is included in the CI_90Lower and CI_90_upper. Is this right?

No. The point estimate is always within it's own confidence interval.
BE is demonstrated if the confidence interval is included in the acceptance range. The acceptance range has to be stated in the statistical protocol and agreed upon with the competent authority. See the third window of the BE wizard: 'Percent of Reference to Detect' which in WinNonlin defaults to 20%.
For this example the lower acceptance limit is given with 100%-20% and the upper with 1/(100%-20%)=125%. 20% are conventionally applied, but may be smaller for NTIDs in some legislations (Canada, Denmark), where e.g., 10% is required (lower: 90%, upper: 111.1...%), or wider (EU and many other countries for C_max of HVDs) as well (lower: 75%, upper: 133.3...%). If you are testing more than one formulation simultaneously (e.g., for dose proportionality) Bonferroni-adjusted alphas may be needed.
If at least one confidence limit lies outside the acceptance range, BE is not demonstrated (deviation of test from reference and/or CV_intra higher than expected in study planning, more drop-outs than expected,...). Only if the entire CI lies outside the acceptance range BioINequivalence is shown.

❝ 4) I am not clear if the 95% confidence level from the "Bioequivalence

❝ Fixed Effects" window of the BE wizard is related to the 90% confidence

❝ interval from the "Bioequivalence Options" window.

Forget it, nobody will require these CIs.

---

Edit: Link corrected for FDA’s new site. [Helmut]