Programming the statistical procedures from SAS

are marginal effects in PROC GLIMMIX possible?

Reply
New Contributor
Posts: 3

are marginal effects in PROC GLIMMIX possible?

Hello all

 

I am analyzing health care data, which is structured in three levels: patients, clinicians and practices.  All variables, including the outcome, are binary.  The model ran beautifully after much tweaking.  The final model is pasted below.  Now, I'd like to get some marginal effects to help with interpretation.  Is there an easy way to do this with this type of model?

 

Thank you!


PROC GLIMMIX data=out.gen_np_NPI_size_ge5 METHOD=LAPLACE NOCLPRINT;
CLASS  image_rfr_npi TIN;
MODEL  X_RAY_MRI_FLAG   (EVENT=LAST) = np NPI_female TIN Disable age AIDS ALCOHOL ANEMDEF ARTH BLDLOSS CHF CHRNLUNG COAG DEPRESS D M DM CX DRUG HTN HTNCX HYPOTHY LIVER LYMPH LYTES METS NEURO OBESE PARA PERIVASC PSYCH PULMCIRC RENLFAIL TUMOR ULCER VALVE WGHTLOSS Atlanta Bangor Boston Chattanooga Cleveland Dallas Des_Moines Elgin Greensboro Kingsport Madison Melrose_Park Milwaukee Morristown Nashville New_Brunswick New_Haven Omaha Orange_County Others Peoria Phoenix Pittsburgh Portland Raleigh San_Bernardino Seattle Springfield Sun_City Tampa Worcester / CL DIST=BINARY LINK=LOGIT SOLUTION ODDSRATIO (DIFF=FIRST LABEL);
random intercept / type=vc subject=image_rfr_npi;
COVTEST / WALD;
run;

 

Table 9. Estimates from Two-level Generalized Linear Dichotomous Models Predicting Image (N: 83,705 patients & 7,071 Clinicians)

 

Model 1

Model 2

Model 3

Model 4

Model 5

 

Naïve, random npi intercept

Model 1+ level 2 predictors

Model 2 + level 1 predictors

Model 3 + TIN

Model 4 + HRR

 

Fixed Effects (Solution for Fixed Effects)

   

Intercept

-0.952

-1.163

-1.598

-2.282

-2.365

 

***

***

***

***

***

Specialist

 

0.747

0.747

0.434

0.434

 

 

***

***

***

***

Odds Ratios

 

2.1

2.1

1.5

1.5

Degrees of Freedom

    

76,572

Confidence Intervals

 

2.0-2.23

2.0-2.23

1.45-1.65

1.44-1.65

Predicted Probability

0.278

0.168

0.093

0.093

0.086

Error Variance (Covariance Parameter Estimates)

Intercept (NPI)

0.604

0.492

0.481

0.252

0.252

Model Fit

(-2) Log Likelihood

97,134.74

96409.79

95,289.10

93,530.39

93,498.85

NPI ICC

15.51%

13.01%

12.76%

7.11%

7.11%

 
    
      
      
      
      
     
      
      
 
      
      
 
      
      
       
Ask a Question
Discussion stats
  • 0 replies
  • 140 views
  • 0 likes
  • 1 in conversation