I am analyzing a BE study using parallel study design. In the study, the 80 subjects were divided in three cohorts as the clinical site was not able to dose all of the 80 subjects on the same day. To take this variable into consideration, I thought of adding a Cohort to the plasma data file. How should this be captured in the BE tab? Should it be under Fixed Variance as:
Treatment
or Treatment+Cohort
or should I consider the cohort effect as a random factor?
Hi Alex, I’m not personally sure what to advise, can you check with the statisticians who wrote the original SAP for your protocol? Were the practicalities of the large group size examined then?
Ccertainly I think for a regulatory submission you’ll have to check the relevant guidelines e.g. EMEA varies from FDA for Crossover designs;
page 15 of the EMEA “GUIDELINE ON THE INVESTIGATION OF BIOEQUIVALENCE”
So in Phoenix WinNonlin we deleting the default random effect on Subject(Sequence)and update the Fixed Effects tab to specify the model as ‘all fixed’; Sequence+Formulation+Period+Subject(Sequence).
As this a statistical methodology question, have you looked at/asked on the BEBAC forum?
Hello everyone, long time lurker here, and I wish to apologize for resurrecting this topic, but I am faced with the following question, and have found no answers on the bebac forum.
In a similar study as the original poster, I have a parallel study design with 2 groups dosed on different dates. I would like to ask what the best method of adding the group variable as a Fixed effect is?
Should I add the group variable purely as a fixed effect, or should it be Treatment * group?
EMA and most others
The BE-GL states “The statistical analysis should take into account sources of variation that can be reasonably assumed to have an effect on the response variable.”
Is it reasonable to assume an effect if subjects were treated in the same CRO, the same bioanalytical method was used, etc? No. With two exceptions in more than 40 years I never came across cases where people not simply pooled the data. All studies accepted…
FDA, Eurasian Economic Union (incl. Russia!), Gulf Cooperation Council
Give a justification and pool the data as usual (i.e., use the common model without cohorts as an effect).
A between-group comparison has terribly low power (the study was not planned for that). Hence, commonly the test is performed at the 10% level. However, a significant result might well be a statistical artifact – 10% false positives are expected (i.e., you get a significant effect though there is no true one). See this presentation for the background and examples (sorry, cross-overs but you’ll get the the idea).
Only if you want to fix your trousers with a belt plus suspenders, add cohort as a fixed effect and ignore an eventual significant (>0.1) outcome.
Should I add the group variable purely as a fixed effect, or should it be Treatment * group?
Don’t dive into the murky waters of interactions. See the EMA’s Q&A document in the context of Two-Stage designs: “If such an assumption were true there is no single formulation effect that can be applied to the general population, and the estimate from the study has no real meaning”.