11-24-2015 03:37 PM
I have a question regarding estimating hospital effect using random intercept logistic regression.
For example, I have crude mortality rate and would like to know hospital variation on mortality.
model death=x1-x5/dist=binary link=logit ddfm=bw s;
random intercept/subject=hospital solution type=vc;
output out=glimmixout pred(blup ilink)=predprob
How can I get predicted hospital mortality or adjusted hospital mortality?
Can I use the random intercept for each hospital (using the random effect plus the estimate of the intercept) to compare the adjusted mortaliy across hospitals? This would assume that all covariates x1-x5 are at centered values, right?
Or should I use the predprob for each patient and sum them for each hospital and get an average probablity for each hospital?
Any thoughts about this?
Thanks a lot.