GLIMMIX Predicted probabilities and 95% confidence interval for new person

[ks17]

Hi, I am using proc glimmix for a mixed effects model with patient-level data and a random intercept for site. I am trying to get the predicted probabilities for a new patient at each of the sites by appending a data set with an empty response column to the full data set and using the following code to obtain the predicted probabilities and CI for the subset of new patients: 

output out=predOut predicted(blup ilink)=predProbs stderr(blup ilink)=stderr lcl(blup ilink)=lower ucl(blup ilink)=upper; 

But I am unclear what the confidence intervals and standard errors represent. From the documentation I found: "The distinction between prediction and confidence limits in Table 38.11 stems from the involvement of the predictors of the random effects. If the random-effect solutions (BLUPs, EBES) are involved, then the associated standard error used in computing the limits are standard errors of prediction rather than standard errors of estimation. The prediction limits are not limits for the prediction of a new observation." 

So does that mean that I cannot use the computed CIs as the 95% CI about my predicted probability for this new patient. Instead should I bootstrap, by refitting the model for each bootstrapped sample and getting the predictions for this new patient and then using that distribution for my 95% CI? And what exactly does the outputted standard error and confidence interval represent. 

[Rick_SAS]

Please provide the complete GLIMMIX statement you are using. Are you including an intercept term for each patient?

The different types of predictions are discussed in the article "Visualize a mixed model that has repeated measures or random coefficients," although that article does not discuss confidence limits.

[ks17]

There is a random intercept per "site" not per subject. The "fulldata" data set contains the data to which I want to fit the model, and an appended data set of new patients, which are patients with the same "var1","var2","var3" values but one for each site and their outcome variable is missing. From this code, I can obtain predicted probabilities for this same patient as if they had been treated at each site. This code also outputs a standard error and a 95% CI. 

In general, I want to know what the interpretation of the standard error and the 95% CI for this prediction are, and if based on the documentation these should not be used for a new patient that was not in the original data set. 

Proc glimmix data=fulldata;
class site var1 var2 var3;
model outcome(event='1')=var1 var2 var3 /link=logit dist=binary;
random intercept/subject = site; 
output out=predOut predicted(blup ilink)=predProbs stderr(blup ilink)=stderr lcl(blup ilink)=lower ucl(blup ilink)=upper; 
run;