Bioequivalence and Bioavailability Forum

Hi Detleffff and David,

❝ Post-hoc power is useless! Search the forum to find numerous discussions on that topic.

❝ Have also a look at Helmut's lectures, especially about power and sample size estimation.

❝ For instance this one: "Sample Size Challenges in BE Studies and the Myth of Power"

Haha, this forum is sometimes a venue for extremists.

Upon careful deliberation together with my crew, I have decided to award you both a full point and none of you are lined up for keelhauling or flogging at the moment.
a. Detlefff, it is very true that posthoc power is often useless. It is too often used in a way that is not particularly informative and where clinical managers just seem to present the PHP in order to cling on to some dumb argument why they have conducted a good trial (even if it failed, of course).
b. David, I think you are trying to make a point which was not picked up by Detleffff: If the trial fails it could be due to chance, and it might suggest that some of our assumptions were not met. In such a situation we could plug in the observed CV and observed GMR in the power equation and I can see your point, it may actually be one little step towards getting a better understanding. We might achieve the same by looking separately at CV and observed GMR, but certainly PHP is better than nothing in this situation.

Having said this I, too, am not in too much favour of PHP; I am only writing this post because PHP exceptionally could have some limited use as David pointed out, even though alternatives to PHP in that situation would at least give the same info, if not more. But the more ways we can look at failures, the better we will understand them.

A good day to you both

Amen.

Dear Detlew,

If post-hoc power is useless why power calculations are used in Potvin methods for sequential designs?

There is nothing wrong with post-hoc power calculations imo. What is wrong is the conclusions you take from it and if you look at post-hoc power alone to justify a failure. As I stated, for studies that fail to demonstrate bioequivalence a lot other methods to "dig" into the failure could be used. As El Maestro stated GMRs and CVs may be more informative alone, but in fact I believe they all provide a similar conclusion. When I look at post-hoc power, GMRs and CVs what I'm trying to check is if the problem was in the formulation or in the number of subjects included in the study.

That being said, power calculations or CVs alone are not the way I use to check data for explanation of failures. As I stated, I like to check individual ratios and to perform sensitivity analysis by removing "extreme" subjects and check how the overall conclusion is affected. Sometimes, the explanation for those failures may be in those extreme subjects and discussions with clinical operations may unleash some problems not evident from the data collected. Pooling data from several studies in the same unit is also useful to check for potential "trends" in outlying values, although I never had the opportunity to do that.

Regards,

David