How to test the interaction between strata and treatment groups in a mmrm model?

[Robin_moon]

I constructed a mmrm model as below:

 proc mixed data = have;
 class id treatment week strata;
 model chg = base treatment week treatment*week treatment*strata/ ddfm = KR;
 repeated week/ subject = id type = UN;
 lsmeans treatment*week/ cl alpha = 0.05 diff
 ods output Tests3=tests3 lsmeans=lsmeans diffs=diffs;
 run;

(1) id: represents the patient id;

(2) treatment: contains 2 group, "drug" and "placebo";

(3) chg: change from baseline of hba1c;

(4) base: the value of hba1c at baseline;

(5) week: contains 3 levels: "week8", "week16" and "week24"

(6 )strata: stratification based on the median of baseline hba1c, ie

      group1=patients of those who have baseline hba1c <= median of baseline hba1c;

      group2=patients of those who have baseline hba1c > median of baseline hba1c.

In general, I'd like to know how to the test the interaction between strata and treatment groups. I am not sure if I constructed the model in a right way because I looked at the results (as below), the values of Diffs in LS means were pretty close between strata, but p for interaction was significant. (The Diffs in LS means were got from the output file diffs, and the p for interaction were got from the output file test3.)

I also checked the Diffs in LS means of week8 and week16, the diffs in LS means were all pretty close between 2 strata, but the p for interaction were all significant (p-value < 0.0001). So I am not sure if I interpret the result in the right way. What data should look quite different if the p for interaction is statistically significant? Thanks.

[SteveDenham]

You may need to simplify your model, as I see 3 variables that are functions of one another:

Chg = observed value - base

Base

Strata = median function determined by Base,

so the induced collinearity between base and strata may well be driving the similarity of the estimates between the strata groups. What happens if you remove strata from the model, and (perhaps) post-process the predicted values obtained from an OUTP= option in your MODEL statement into quintiles, terciles or two groups? Does that address your research question?

SteveDenham