Bioequivalence and Bioavailability Forum

Hi ElMaestro!

Thanks for the reply.

❝ confusing that they ask for means and not medians.

Yes! Mean score does not make much sense, working with medians would be better. I found, yesterday, this Public Assessment Report from EMA where the Wilcoxon Signed Rank Test was used (Pages 13 and 14), but seems like no CI was assessed as FDA requests. I'm not really sure on how to construct a 95% CI for MSt-1.25*MSr (where MSt= Mean Score for test and MSr= Mean Score for reference).

❝ A potential way forward to requirement #2 is the construction of confidence intervals based on proportions of score 1 and above from binomial testing.

You mean classify the scores as meaningful (i.e. 2 or higher) and not meaningful (i.e. 1 or lower)?

❝ As I see it this means several tests must be done which can be rather nasty.

That worries me very much also.

I think that the Wilcoxon SR test, as it was used in the Actavis study linked above, would be a good way do compare T vs R, since it also evaluates (indirectly) the adhesion through time. What do you think?

Thanks in advance.

Lucas

---

UPDATE

I couldn't reproduce the p-value described in the Actavis' study (0.0882) in R, using wilcox.test(), can anybody help me with that?