Hello, I have a dataset of ~25000 patients from ~1300 hospitals. I would like to calculate the odds of receiving one of two treatments, then obtain the probability of treatment for each patient (for propensity score adjustment). 20 patients were selected from each hospital, with treatment=1 oversampled, so the data has been weighted (weight variable='selweight'). I would like to control for correlations between patients at the same hospitals , and let the correlations vary by hospital. I have tried proc glimmix, with treatment='x', hospital code='oldfaccode' and patient id='puf_id', and a number of covariates, with an exchangeable correlation matrix: proc glimmix data=raw_data;
class
oldfaccode puf_id dxyr (ref='2008') facility_type_cd (ref='2') facloc(ref='Northeast') RACEcat (ref='White') hispanic (ref='0')
MED_INC_QUAR_12 (ref='1') NO_HSD_QUAR_12 (ref='1') metro13 (ref='0') insured (ref='1') charlson(ref='1')
er1 (ref='Negative/normal, within normal limits') pr1 (ref='Negative/normal, within normal limits') grade1 (ref="Grade I")
comedonecrosis1(ref='1) Present') detect_dcis1 (ref='1) Mammogram')
MODEL x= age dxyr facility_type_cd facloc RACEcat hispanic MED_INC_QUAR_12 NO_HSD_QUAR_12 charlson
metro13 insured er1 pr1 grade1 detect_dcis1 comedonecrosis1 /solution ;
subject=puf_id(oldfaccode) type=cs ;
random intercept/subject=oldfaccode type=cs;
weight selweight;
run;
In order to let the correlations vary between hospitals, I am using the statement random intercept/subject=puf_id(oldfaccode) type=cs Is this correct? I'm not even sure if I am using the terminology correctly. Additionally, how do I output the predicted probabilities? Thank you so much for any help!
... View more