Bioequivalence and Bioavailability Forum

Hi J M,

❝ I have theoretical questions about computing BE and making the link with statistical model. Considering the statistical linear model for cross-over design :

❝ Xijlk = µ + Tl + Pk + SEj + SUi|SEj + eijkl

❝

❝ With:

❝ Xijlk: ln-transformed response


❝ µ: overall mean


❝ Tl: fixed formulation effect


❝ Pk: fixed period effect


❝ SEj: fixed sequence effect


❝ SUi|SEj: random effect of the ith subject in the jth sequence (nested)


❝ eijkl: independent random effects


❝

❝ I have two questions:

❝ 1/ did I write this statistical model correctly for a 2x2x2 cross-over design? I assume T + P + SE + SU|SE are the CVinter and eijkl the CVintra. I am right?

For EU you will need to treat the subject effect as fixed.
Depending on your coding your may or may not want/need to speak of the nesting.
"ijlk".... do you use another version of the alphabet in France?

❝ Why in the bebac presentations there is multiplicative model roughly the same with “*” instead of “+” and withoud the nested part “SUi|SEj” I have found in other references? Any relevant links?

Yeah, even oracles make mistakes.
"Xijk: ln-transformed response of j-th subject" he meant un-transformed as long as we have * signs between factors.

❝ 2/ in PowerTOST (nice package!), I enter the T/R ratio when I use the power.TOST() function (I suppose that’s T in my model) and CVintra that corresponds to eijkl, did it suppose that there is no effect of P (period), SE (sequence) and SU|SE (subject nested in sequence)?

This is the wrong question. Sequence, Period, Subjects effects do not influence our decision about BE. Period and Sequence effects are nuissance effects. Subjects differ generally - in most studies you will see a P-value that is low.