LSMeans and SE from MMRM

[Sandrapharma]

I am using proc mixed for MMRM to analyze change from baseline from different visits using the code below. TRT is the treatment variable, which is either Active1, Active2 or Placebo. I noticed that when I used the same Placebo data vs Active1 or Active2 and run the MMRM model independently, the LSMean and SE for Placebo (TRT) is different. Why does the LSMean and SE estimate for Placebo changes even it is the same placebo data? For the placebo estimate, is it based on data from placebo only or is it based on data from both Active and placebo?

proc mixed data=datasets;
class  TRT(ref='Placebo') AVISIT(ref='Week 3') SUBJID;
model CHG = Base TRT AVISIT TRT*AVISIT base*AVISIT/solution ddfm=kr;
lsmeans TRT TRT0*AVISIT/diff=all cl;
repeated AVISIT/type=UN subject=SUBJID;
ods output Tests3=ES LSMeans=LS Diffs=DF;
run;

[jiltao · Accepted solution]

LSMEANS are model based and are computed as L*beta. When your data is different for Active drugs, the beta estimates are different, and therefore the LSMEANS are different. This affects not only the lsmeans for active drugs, but also for lsmeans for placebo, where the data values might remain unchanged. Remember, the model is fit and estimated using the entire data, not separately by different treatments.

Hope this helps,

Jill

[Sandrapharma]

Thank you for the explanation!