Bioequivalence and Bioavailability Forum

Hi Kaoula,

❝ Hi Everybody , I have big problem to understand (...)

You're not alone - I think most of us have trouble getting the definitions right from time to time.
Here's how I think about it, but please be aware I am often wrong:
1. When something is highly variable in the context of BE it currently means it is associated with an intra-subject CV>30%.
2. When a drug product is a HVDP then it means it has been associated with a CV>30%, preferably measured with a replicate administration in order to qualify.
3. You can give a drug directly intra-vascular and still observe a within-subject CV>30% - this would, I guess, qualify it as a HVD. If the formulation of interest itself is an IV formulation then the drug is a HVD, and the formulation hence is a HVDP.
4. HVDP and HVD thus differ by the extra layer of complexity introduced by the formulation of the drug (drug here means the API).
5. A formulation of a HVD will be a HVDP unless you measure a mighty rare statistical fluctuation. Are there case stories, anyone knows?
6. Subject-by-Formulation interaction is when the GMR or similarity of test and Ref depends on what subject you measure it in. Bear in mind that when you measure something in practice there will be a lot of measurement uncertainty or error variance and in some cases and depending on study design you might say the this uncertainty is a measure somehow of the intrasubject variability. But this uncertainty itself has nothing to with the SxF interaction. An SxF interaction is an effect.

I'll state this a little differently. Imagine this:
- John, Jimmy and Jenny respond identically (plusminus some measurement uncertainty) to Fred and Fiona in terms of Cmax of the Test product.
- John, Jimmy and Jenny respond identically (plusminus some measurement uncertainty) in terms of Cmax of the reference product, but they do not respond identically to Fred and Fiona in term of Cmax of the reference product.
That's an SxF interaction.
If you run a study with N formulations and M subjects you'd introduce a max of NxM columns in the model matrix if you want to study it (minus the model's redundancy that arise from the other terms); their total will deduct from the error's df if speaking of one such is relevant.