GEE modelling using proc genmod interpretation

[lone0708]

Hi All, 

I am sorry for a novice question, but i am having trouble interpreting the output of my proc genmod...

I am modelling the probability of being a case (case = 1, non case = 0) and estimating the trend in cases through a 15 year period of non-independent data using proc genmod and GEE modelling. I have one covariate, that might explain the trend. 

ods output GEEEmpPEst=output;
proc genmod data = have descending;
class id year (ref="2005") covariate (ref="0");
model case =year / dist = bin
                              link = id
                              lrci;
repeated subject = id;
run;

 My output is an

intercept estimate, estimates for every year, estimates for the covariate.

But which estimate is the trend estimate? Meaning the trend in cases increases XX% pr year.

[SteveDenham · Accepted solution]

You will probably want to remove "year" from the CLASS statement.  This would fit year as a linear continuous variable (=trend).  If you need ORs at specific years, you can use an LSMEANS statement with AT= and ODDSRATIO options.

I don't see "covariate" in your MODEL statement, so I am at a bit of a loss to see how you got an estimate for it.  In any case, if you believe the covariate may affect the trend, try fitting the following 

ods output GEEEmpPEst=output;
proc genmod data = have descending;
class id year (ref="2005") covariate (ref="0");
model case =year covariate covariate*year / dist = bin
                              link = id
                              lrci;
repeated subject = id;
run;

The result for covariate will indicate change in the intercept, and covariate*year will give the difference in the slopes compared to the reference value of covariate.

SteveDenham

[lone0708]

Thank you for the answer,
When i remove year from the statement, I get one estimate for the intercept, one for year and one for the covariate.
Is the intercept = trend estimate ?

[StatDave]

No. The Year parameter estimate is the trend in the case probability over time.