Keeping subjects in mixed-effects models [General Statistics]
❝ I do regard the fasted leg as the reference. The treatment lost was […] one of the tests (fed) that did not show up for Period 3.
OK, so no problem with PHX’s REML. Just run the analysis on your entire data set.
❝ The protocol was written and provided an overview of the data analysis done in Phoenix. It did not discuss dropout subjects and what to do if subjects dropped out. Perhaps it is a good thing if the protocol is vague on this issue … suitably vague? do you agree?
I don’t agree. IMHO it is better to invest your intellectual horsepower in the protocol, than to state sumfink ambiguous and have to come up with “creative” ideas in the analysis. If the protocol is written by a genius, the analysis can be done by a dummy. I’m not sure whether it always / easily can be done the other way ’round. Creative solutions might be interpreted by assessors as the first step of cherry-picking:

“Oh, the study passed by applying Method X. But there are alternatives. Would the study have failed with Method Y? Let’s ask the applicant for a justification, …”.
Such a request can turn out to be a show-stopper. Methods are based on different assumptions, which are difficult – if not impossible – to proof given the limited sample size in BE studies. So you might be forced to come up with a lot of
To call the statistician after the experiment is done
may be no more than asking him to perform a postmortem examination:
he may be able to say what the experiment died of. R.A. Fisher
You can’t fix by analysis
what you bungled by design. R.J. Light, J.D. Singer, J.B. Willett
100% of all disasters are failures of design, not analysis. R.G. Marks
❝ Often I think it is mistake to give too much detail in a protocol, since you can create protocol violations. The FDA did not comment on this issue during their protocol review […]
I would rather call them protocol deviations; violation is a little bit harsh. See also ICH’s Q&A on E3. However, if you stated contingency plans in the protocol (aka “be prepared fo the unexpected”) you rarely have to deviate from the original method(s).
❝ All of the subjects completed the study. I showed that the result were identical in SAS (GLM) and WinNonlin and Kinetica.
Note that
Proc GLM
= WinNonlin only for complete data sets (balanced or imbalanced, but subjects completed all periods). For imbalanced data sets Proc Mixed
= WinNonlin; if you want to get the same results as Proc GLM
you have to exclude incomplete subjects in WinNonlin.Dif-tor heh smusma 🖖🏼 Довге життя Україна!
![[image]](https://static.bebac.at/pics/Blue_and_yellow_ribbon_UA.png)
Helmut Schütz
![[image]](https://static.bebac.at/img/CC by.png)
The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
Complete thread:
- BE and drop out subjects Phoenix AngusMcLean 2014-01-24 00:02
- Keeping subjects in mixed-effects models Helmut 2014-01-24 04:18
- Keeping subjects in mixed-effects models AngusMcLean 2014-01-24 14:26
- Keeping subjects in mixed-effects modelsHelmut 2014-01-24 15:28
- Keeping subjects in mixed-effects models AngusMcLean 2014-01-24 16:28
- Keeping subjects in mixed-effects models Helmut 2014-01-24 16:59
- Keeping subjects in mixed-effects models AngusMcLean 2014-01-26 15:58
- SAP Helmut 2014-01-26 16:25
- Keeping subjects in mixed-effects models AngusMcLean 2014-01-24 16:28
- Keeping subjects in mixed-effects modelsHelmut 2014-01-24 15:28
- Keeping subjects in mixed-effects models AngusMcLean 2014-01-24 14:26
- Keeping subjects in mixed-effects models Helmut 2014-01-24 04:18