How to perfrom mixed effect model with poisson regression with robust estimator?

[Diels_O]

Dear all, I have solve using proc genmod to perform modified poisson regression to generate Risk ratio,

here, how can I adjusted the site effect, would the following program appropriate?

proc genmod data = stroke;
class enroll_order group(desc) score(desc) IVT sites/PARAM=GLM;
model score = group ivt sites/ dist=poisson link=log covb corrb;
repeated subject = enroll_order/type = un;
estimate "beta" group 1 -1/exp;
lsmeans group/diff exp cl;
run;

Besides, how can I use mixed model to get risk ratio with sites as random effect?

I'm confuse to get the RR ratio, and the class group(desc) seems to wrong in proc glimmix,

How can I calculate RR with robust estimator as Zou.et(2004)

proc glimmix data=stroke EMPIRICAL;
class site group;
model Score= group/ dist=poisson link=log SOLUTION;
random site/ subject= enroll_order;
run;

[SteveDenham]

For your GLIMMIX code, I would add enroll_order to the CLASS statement. I am somewhat baffled by the GENMOD code, though.  The CLASS statement there looks to me as if it treats your response variable (score) as a multinomial.  Given that, I am not sure that the data follow a Poisson distribution, but rather follow an ordinal multinomial distribution.

SteveDenham 

[Diels_O]

Thank you very much! Now I add it into a class statement, and how can I estimate the RR ratio using the output of glimmix?  Thank you again!

[Diels_O]

Steve, The code in the Genmod was introduced by Zou.et (2004) using the genmod to generate the modified poisson regression with robust estimator. The score was a binary variable , not a multinomial. Really thank you for your careful look.

[StatDave]

See this note for Zou's Poisson method. Note that the response must be binary, not multi-level to do this - a relative risk is a ratio of probabilities of the event level of a binary response. In the note, it is shown using GENMOD, but GLIMMIX could also be used.

[Diels_O]

Thank you Dave! I met problem with the lsmeans and the estimate, How can I get RR and it's CL in proc glimmix?

The estimate seems to not provide exp function. Thank you again for your kind help!

[StatDave · Accepted solution]

Just omit the EXP option and then exponentiate the estimate and confidence limits manually.

[hongjie76]

Try: estimate "RR group" group 1 -1/cl ilink;