Two-Stage Design and EMA’s fixed effects

Phoenix Certara Forums Phoenix WNL basics
1

Hi,

a goody from the BE-wizard. I used to setup Potvin’s model of pooled data (stage 1&2) as:
Fixed: Sequence+Stage+Period(Stage)+Treatment
Random: Subject(Sequence*Stage)

Results agree with reported ones (Example 2) of the paper:
94.12% CI: 88.45–116.38%

Unfortunately EMA wants all effects fixed (which works in the conventioanl setup of stage 1) but in the pooled analysis I got:
“ERROR 11050: Parsing error: The containing term must be in model.”
The core output stops after the partial test; no LSMeans are calculated.

I could run the model in LME – but only if I didn’t ask for the LSMeans (otherwise I got the same error like from the BE wizard).
Results validated against R and SAS (http://forum.bebac.at/mix_entry.php?id=9462).

Do you think it will be possible to catch the error (I guess LME passes the results to BE)? Most users are not familiar with LME and are lost when they try to use the BE wizard.

Example project attached.
[file name=Potvin2.phxproj size=2567353]Certara | Drug Development Solutions

P.S.: Of course LSMeans are not necessary for the BE analysis. Some regulators will be surprised if the are missing anyhow.Potvin2.phxproj (2.45 MB)

2

Hi Helmut, I’m just getting a chance to look at this. I changed the last BE model in your project to: Sequence+Stage+Period(Stage)+Treatment+SequenceStage+Subject(SequenceStage) so that the “containing term” is in the model, and the error message is no longer given. The CI_User_Lower and CI_User_Upper in the BE output are now: 88.445196, 116.37701 The Ratio_%REf is: 101.45436 These are the same as the point estimate and back-transformed CI’s in the “pooled fixed backtransform” object. Does that resolve your problem? Linda

3

Hi Linda! [quote]Sequence+Stage+Period(Stage)+Treatment+SequenceStage+Subject(SequenceStage) so that the “containing term” is in the model, and the error message is no longer given.[/quote] I see. You have to introduce "SequenceStage" in order to use the nested subject-term. [quote]The CI_User_Lower and CI_User_Upper in the BE output are now […] These are the same as the point estimate and back-transformed CI’s in the “pooled fixed backtransform” object.[/quote] Fine. Agrees with SAS and R. :wink: [quote]Does that resolve your problem?[/quote] Yes and no. I wouldn’t have guessed this setup. The paper reads: “… model including sequence, stage, period(stage), treatment, subject(sequence × stage) …” The SAS-code is: treatment period(stage) sequence stage subject(stagesequence) The R code is: 0+treatment+period%in%stage+sequence+stage+subjectj%in%(stage*sequence) Maybe you post this code at the support site as well. Two-stage designs are acceptable in Canada, the US, and the EU. Russia is expected to follow soon. THX for the solution; I’ll post it at my forum.