This message has been cross posted to the following eGroups: Statistical Consulting Section and Statistical Computing Section .
-------------------------------------------
Dear All,
I have 63 clusters (physicians) and each cluster has 1-41 patients (610 patients in total but 176 have missing values).There are two repeated measurements (of response and some predictors) on each of these patients.
I tried to fit the GEE model (using several correlation structure as well as ALR regression) as well as mixed effect regression model for multi level data using the following codes but either they don't converge or the estimate of the covariance parameter (in case of GLIMMIX) for Physician level is negative which basically means it is very small or zero.
I would really appreciate any comments or suggestions from the group.
/********************************************************************************************************/
proc genmod data=xyz descending;
class gender grage gr_revenu SM cardiac_related_condition PHYSICIAN PATIENT;
MODEL y= gender gr_revenu SM CMB SE AQ AC/dist=bin link=logit TYPE3;
/*OR TRY THIS*/
/*REPEATED SUBJECT=PATIENT / SUBCLUST=PHYSICIAN logor=NEST1; */
/*OR TRY THIS*/
/*repeated subject=PATIENT(PHYSICIAN) / logor=fullclust;*/
/*OR TRY THIS*/
Repeated subject=PATIENT(PHYSICIAN) / corr=EXCH corrw;
run; /********************************************************************************************************/
/
*QUASI-LIKELIHOOD*/
/*PROC GLIMMIX data=xyz NOCLPRINT;*/
/*or this? FULL LIKELIHOOD*/
/*PROC GLIMMIX data=xyz METHOD=QUAD(QPOINTS=1);*/
/*or this USING THE OPTION method=rspl nobound WILL GIVE PSEUDO LIKELIHOOD BASED ESTIMATES AND NOBOUND ALLOWS NEGATIVE VARIANCE ESTIMATES*/
PROC GLIMMIX data=xyz METHOD=RSPL NOBOUND;
class gender grage gr_revenu SM cmd PHYSICIAN PATIENT;
MODEL y(event='1')= gender gr_revenu SM0 cmb SE AQ AC/ DIST=BINARY SOLUTION ODDSRATIO ;
/*try this? It is recommended to model unstructured covariance matrices in terms of their Cholesky parameterization (TYPE=CHOL) rather than TYPE=UN. */
/*RANDOM INTERCEPT / SUBJECT = patient(physician) TYPE=CHOL;
RANDOM INTERCEPT / SUBJECT = physician TYPE=CHOL;*/
/*try this ? Variance Component*/
/*RANDOM INTERCEPT / SUBJECT = patient(physician) TYPE=VC;
RANDOM INTERCEPT / SUBJECT = physician TYPE=VC;*/
/*Unstructured*/
RANDOM INTERCEPT / SUBJECT = patient(physician) TYPE=UN;
RANDOM INTERCEPT / SUBJECT = physician TYPE=UN;
RUN;
Results: GLIMMIX:
Patient(physician) =0.982
Physician=-0.059
Results GENMOD:
alpha1=1.99 Alpha2=-0.05
Thank you
Best Regards,
Tasneem
-------------------------------------------
[Tasneem] [Zaihra]
[Post Doctoral Fellow]
[McGill University]
-------------------------------------------