Bioequivalence and Bioavailability Forum

Hi Achiewin,


I second what Helmut said. We don't know much about the distribution of e.g. Tmax (or Tmax difference) so it can be argued that whatever test you apply, it has to be non-parametric.
Most stats packages allow a Wilcoxon (rank-based) test optionally garnished with a CI of sorts.
This gives you a p-value. There is some controversy, re. the null hypothesis that the p-value corresponds to and how p should be derived. An added complexity is that in BE we often have paired designs, and we often (I'd think almost always) have subjects with identical values for the metric being compared (like 6 subjects has Tmax of 2 hrs in one group, and 8 subjects had Tmax of 2 hrs in the other group) and such cases may further complicate the derivation of p.

So, the very simple question of "how do I compare?" quickly spirals into something that has no simple answer. If you are in a situation where you have to send in "something" along the lines of a comparison of something like Tmax etc, then I think it is ok to simply submit what your stats package offers in its most simple form with a clear description of what exactly it is you are doing. Notably be clear if you present data for the median difference or the location of the median of your metric.
If the regulators don't happen to like what you give them, they must tell so and ask for something else. Their request will usually give a decent hint as to what exactly they are then looking for.