Posting: # 18625
I need some suggestion regarding full replicate study statistical data analysis for Europe scope. Study conducted in four period full replicate design. Majority of the subjects completed all four period of the study and some subjects completed at least two periods with one test and one reference treatments.
The subjects with at lease two reference treatments included in calculation of ISCVR.
The study showing more than 30% ISCVR. The study eligible for scaled average bio equivalence approach. (widening criteria for Cmax). The subjects who completed all four periods will be included in PK and Stats. Here my question is whether to include the subjects who completed at least one test and one reference for bio-equivalence calculation?
Thanks and regards
Posting: # 18626
» Here my question is whether to include the subjects who completed at least one test and one reference for bio-equivalence calculation?
I do not work with EMA so I am unable to provide any empirical evidence. However, you may check the EMA guidance
here. Page 14 of the guidance states "...subjects in a
crossover trial who do not provide evaluable data for both of the test and reference products ...should not be included. "
With FDA, the safe route would be to provide statistical analysis with and without exclusions (especially if they both pass ).
I have seen some discussions around regarding drop-outs in replicate studies. Perhaps this implies that subjects aren't evaluable if they don't complete all 4.
Follow up question for the pros: if you include partially completed subjects, does this unbalancing have an effect on the mixed model?
I guess you lose a df if you have a subject that only had RR or TT, so I understand that. But what about a subject that has TR (instead of TRTR).
Posting: # 18627
both the fixed effects model (EMA) and the mixed effects model (FDA) are stone cold. You can fit whatever you like and you can get a qualified likelihood-based answer.
EMA's sentence about at least one T and R for each subjects should be interpreted exactly as such (but behind the scenes, this does not imply that every subject contributes with a (one) data point which is fed into a grandiose pot of T/R ratios).
» Follow up question for the pros: if you include partially completed subjects, does this unbalancing have an effect on the mixed model?
"have an effect" is a bit vague. The only thing that gets funky is the calculation of denominator df's. In case of data gaps this is not so straightforward and that is one of the topics that cause statisticians to attack each other with some very pugnacious remarks.
"(...) targeted cancer therapies will benefit fewer than 2 percent of the cancer patients they’re aimed at. That reality is often lost on consumers, who are being fed a steady diet of winning anecdotes about miracle cures." New York Times (ed.), June 9, 2018.