Confidence intervals vs. point estimators [Regulatives / Guidelines]

Hi Dan,

❝ I am not an expert in statistics but as far as I know the 90% confidence interval for Cmax and AUC tells one that the true test/reference ratio lies within this interval with a probability of 90%.

If the experiment is repeated an infinite number of times, 90% of the point estimates would be within your bounds, that's what you'd expect, all other factors being equal and assumptions holding. This is interpreted to mean that the true PE with 90% certainty is within the bounds you first found.

❝ On the one side we will argue that for both primary bioequivalence parameters there is a overlapping of confidence intervals for all four studies. On the other side the assessor will argue that the point estimators of the second pilot study are not included by the confidence intervals of the other studies -> batch variability! Is this reasonable?

Yes, in principle. Do I get you correctly that there is no single point all four CI's have in common?
Here's a proposal, the guideline specifically reads: "The test product should usually originate from a batch of at least 1/10 of production scale or
100,000 units, whichever is greater, unless otherwise justified."
. This is because everybody knows that there is variability between batches and that performance of a pilot batch is not necessarily equal to a full scale batch. So, you will just with kind words remind the assessor that a bunch of his/her colleagues acknowledge that fact that pilot performance is not pivotal performance and cannot be expected to be. Yes, there is variability between batches. No, it is not a problem, you are handling it well by separating pivotal data from pilot data. This is what you can tell the assessor.

❝ As Pavidus explained BE studies are definitely not suitable to assess inter-batch variability of any product. Which arguments can support this idea?

Well... If you wanted to study everything you might do a mixed model with batch (within treatment some would say) as random factor. In that case one could take variability between batches into consideration. Noone does it, though, and the PK subgroup does not expect you to do this either ("The terms to be used in the ANOVA model are usually sequence, subject within sequence, period and formulation.").
So you can just explain back that batch is never a factor in ANOVAs according to recommendations published by the Efficacy Working Party, and that you see no compelling reasons to deviate from that document.
You might also speculate: The ref. product is probably 500 years old and chances are the production parameters and specifications for that process reflects to some extent reflects habits and traditions from those days, whereas your product accords with today's standards. Therefore, for all practical purposes, generics are often less variable. This is a generalisation and unfortunately impossible to know if it specifically holds for your product.

Honestly, I don't think you have wildly big a problem. I understand why the question is scary, but the task the assessor faces if he/she has to defend his/her views at CMD(h) is worse.

S.

Pass or fail!
ElMaestro