Bioequivalence and Bioavailability Forum

Dear Helmut!

❝ yesterday EMA’s PK working party published Rev. 7 of the Q&A document. According to the new #14:

1. The expected analysis for the combined data in a two-stage design is ANOVA with terms for stage¹, sequence², sequence × stage³, subject(sequence × stage)⁴, period(stage)⁵, formulation⁶.
   

   ❝

2. This model can be fitted provided that in each stage, there is at least one subject randomised to each sequence. This does not supersede the requirement for at least 12 subjects overall.
   

   ❝

3. A term for a formulation × stage interaction should not be fitted.

If you have a Two-Stage study underway, amend your SAP.

Numbering by me

First: Thanx for this valuable information! Indeed I have a 2-stage study under way.

Second: From Potvin et.al.:
"If the individual ln-transformed data are to be used in the analysis instead of the differences, then the error term derived from the GLM ANOVA model including
sequence2, stage1, period(stage)5, treatment6, subject(sequence × stage)4
will give the appropriate s2 term."
Thus the emphasis should be at sequence × stage .

❝ Note 1: The CI in the pooled analysis is not affected if compared to a model without subject(sequence × stage)...

Typo? If you fit without subject(sequence × stage) you would omit totally the subjects effects! This will heavily affect the CI I guess.
If you meant sequence × stage you are totally right. Since stage and sequence are between-subjects effects the additional inclusion of the sequence × stage interaction is only a further breakdown of the subjects effect. The question is why the mighty oracle want us to include such a term .

BTW: Where did your p-value came from?