proc genmod. Do I use distribution = Poisson? Or dist=binary?

[Rubye]

Hi – This is my first post so please forgive any initial transgressions.

Question: In proc genmod, do I use distribution = Poisson? Or dist=binary?

I am modeling risk of death from my exposure “treatment” (trt).
The study subjects come from 35 study sites (e.g. multiple hospitals). This study was not a clinical trial and had no randomization.
I am concerned about potential correlation in the data (due to clustering by hospital site). Total N=829.

I think of Poisson distribution as modeling COUNT data (e.g. # of deaths For example …79 deaths).
I think of binary distribution as modeling Binary data (e.g. death = 0 / 1 ).

If I use dist=Poisson, I believe SAS will account for the correlated data. But I am not seeking to model the count of deaths.
If I use dist=binary, I believe SAS will ~not~ account for the correlated data. However, dist=binary seems logical for modeling the risk of death where the outcome = 0 / 1. (dead / not dead.)

Below are my two sets of code. Clearly I am confused!
Most appreciative of guidance. Thank you-
Rubye

proc genmod data= example descending ; 
class trt ; 
model death = trt / dist=bin link=log ;
estimate "Beta TRT"  trt  -1 1 / exp ;
title "distribution=Binary link=log" ;
run;

/***********************************************************/
proc genmod data= example descending ;
class patient_id death ;
model death =	trt / dist=poisson link=log ; 
       			repeated subject = patient_id / type = unstr ; 
estimate "Beta trt"	trt	-1 1 / exp ;
title 	"distribution=Poisson link=log" ;
*ods output Genmod.Estimates=Peds_Beta ;
run ;

[hlsmith]

Seems to be more binary (yes/no 0/1). I think of count data a little bit more as number of speeding tickets (positive integers). Also, if you have different centers and you think they may explain your dependent - investigate using a random effects variable.

[Rubye]

"...investigate using a random effects variable" I'll take a look at that. Thank you-

[Dason]

Note that you would need to change to proc glimmix to add random effects into your model. (At least I don't think genmod allows random effects... it's been a while since I've used SAS)