Bioequivalence and Bioavailability Forum

Hi Manu,

❝ Since other H-A guidelines don't clearly specify the same, would it be recommanded to have a sensitivity analysis based on all data (complete and incomplete cases) with a random effect model?

For a 2x2x2 design you have no random effect in a standard evaluation with Proc GLM in SAS, even when the bogus random statement is used. The algo still fits the entire set of factors as fixed.

❝ The concern might be if your initial analysis shows BE but not the sensitivity analysis....

I myself, and this might be because I have a walnut-sized brain and no insight into statistics, find it difficult to get my head around sensitivity analyses for BE. After all, we have a 90% CI to tell us how well we know the similarity of T and R, and from the applicant's perspective hunting for outliers and excluding them is discouraged discouraged anyway.
Perhaps in territories where the BE assessment for one of the primary endpoints occurs without construction of a 90% CI would I think a s.a. could make sense.