0
$\begingroup$

I am creating multiple GEEs with the same covariates, but I am testing different clustering variables. The outcome is a binary yes/no variable and both VAR1 and VAR2 are binary, as well. Patients can be seen at 1 of 7 clinics, and each clinic has multiple doctors. Thus, there is an argument to be made that we should cluster based on clinic or doctor (at this time, we are not considering multiple random effect models). I want to create the ROC curve for both the GEE clustered by clinic and clustered by doctor, but when I do so using this SAS note, I am getting the same ROC curve. I have checked the predicted probabilities, and they are different for the two models. Anyone know what is going on or why this is happening? Appreciate the help and insights.

Here is my code and the outputted ROC curve for both.

*CLUSTERED BY DOCTOR;
PROC GLIMMIX data = DATA ORDER=INTERNAL empirical=root ;
    class DOCTOR VAR1 VAR2;
    model OUTCOME(event='1')= VAR1|VAR2    
            / dist=bin link=logit covb  solution ;
    output out=info1 pred(ilink)=phat  lcl(ilink)=low ucl(ilink)=up resid=res student=student;
RANDOM _residual_ / SUBJECT=DOCTOR  TYPE=cs  gcorr solution  ;
Run;
      proc logistic data=INFO1;
         model OUTCOME(event='1') = / nofit; 
         roc "GLIMMIX model" PRED=phat;
         run;

*CLUSTERED BY CLINIC;
PROC GLIMMIX data = DATA ORDER=INTERNAL empirical=root ;
    class CLINIC VAR1 VAR2;
    model OUTCOME(event='1')= VAR1|VAR2    
            / dist=bin link=logit covb  solution ;
    output out=info1 pred(ilink)=phat  lcl(ilink)=low ucl(ilink)=up resid=res student=student;
RANDOM _residual_ / SUBJECT=CLINIC TYPE=cs  gcorr solution  ;
Run;
      proc logistic data=INFO1;
         model OUTCOME(event='1') = / nofit; 
         roc "GLIMMIX model" PRED=phat;
         run;

enter image description here

$\endgroup$

1 Answer 1

0
$\begingroup$

From my post on SAS Communities

Kathleen Kiernan. https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2018/2179-2018.pdf "The choice of the marginal (population-averaged) model or conditional (subject-specific) model often depends on the goal of your analysis: whether you are interested in population-averaged effects or subject-specific effects.

The GEE model is a marginal, or population-averaged model. If you are interested in making predictions about individuals, then you would use GLIMMIX to the fit the conditional model using G-side random effects and obtain the subject specific estimates."

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.