I have this project that in the original method, the person suggested to use a generalized linear mixed model with a logit link, and I need to follow this approach.
The response variable is a binary variable for Remission Response (y/n); All subjects get the same treatment at Week 0. And the project wants to look at the effectiveness of the treatment in achieving the Remission response in a real world evidence setting in three different stratum (bionaive, 1 prior biologic, 2 or more prior biologic)
The Remission Response measured repeatedly at week Weeks 6, 16, 26, 36, and 52 for each subject
Another variable Stratum, which is a categorical variable with three levels (biologic naive, 1 prior biologic, 2 or more prior biologic)
Timepoint(Weeks) and Stratum will be included as fixed effect. Least square means odds ratio and 95% CI need to be calculated for each stratum.
My model would be:
PROC GENMOD DATA=Test;
CLASS Remission Weeks Stratum USUBJID;
MODEL Remission=Weeks Stratum Weeks*Stratum /DIST=BIN LINK=LOGIT;
LSMEANS Stratum*Weeks;
Repeated USUBJID;
or
PROC GLIMMIX DATA=Test;
CLASS Remission Weeks Stratum USUBJID;
MODEL Remission=Weeks Stratum Weeks*Stratum /DIST=BIN LINK=LOGIT;
LSMEANS Stratum*Weeks;
Random USUBJID;
RUN;
My questions are:
1. Can you use PROC GENMOD for a generalized linear mixed model? Would both model work? (I don't have the data yet)
2. Can someone explain if we need a mixed model for such repeated measures? I don't understand why the person originally suggested use a generalized linear mixed model where I did not see any random effect to be included.
I am pretty sure @StatDave will have some comments that clarify the issue, but until then:
If you use PROC GENMOD to analyze the repeated measures, you end up looking at marginal differences between levels, using generalized estimating equations.
If you use PROC GLIMMIX, and fit both the repeated factor and the individual subject, you can get conditional differences (conditioned on the random effects).
Think about your research question and goals, and the choice becomes much easier.
A note to consider: Check your syntax for both PROCs. I think there may be minor changes needed in the REPEATED statement in GENMOD, and there needs to be a RANDOM weeks/type= {pick from CS, CSH, or CHOL) subject=USUBJID; statement in the GLIMMIX code.
SteveDenham
PROC GENMOD does not fit the generalized linear mixed model which is a subject-specific model. That type of model is available, as you note, in PROC GLIMMIX. GENMOD fits a population-averaged sort of model using the Generalized Estimating Equations method. The choice is more a matter of what type of inference you want to make. If you want to make statements about individual subjects, you would use the GLMM in GLIMMIX. If you want to make statements at the overall population level, then use the GEE model in GENMOD. Also, in GENMOD, proper syntax is: REPEATED SUBJECT=USUBJID; to request the default independence structure of the correlation matrix. If you want to specify a different structure, add the TYPE= option - for example, REPEATED SUBJECT=USUBJID / TYPE=EXCH;
Registration is now open for SAS Innovate 2025 , our biggest and most exciting global event of the year! Join us in Orlando, FL, May 6-9.
Sign up by Dec. 31 to get the 2024 rate of just $495.
Register now!
ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.
Find more tutorials on the SAS Users YouTube channel.