Zero Inflated Poisson using Proc Genmod

[Jason]

Hello,

  I have a question regarding an outcomes analysis we are conducting on an HIV intervention project.  Using Genmod and trying to see if there is a difference at sixmonth follow-up (totwocnds) by group assigned (group) controlling for pretest value (totwocnd), age (yrs), duration to follow-up (fu3), and number of intervention sessions they attended (sessioncount). 

The distribution appears skewed towards zero, and therfore we decided to use a zero inflated poisson (ZIP) distribution for the outcomes count variable (totwocnds).  Everything seems to work well in the model.  However additionally we are required to control for possible clustering effects of the subjects within their cohorts.  I attempted to use the repeated statement, however I receive the following error that it can not be used with a ZIP distribution: 

ERROR: Zero-inflated distributions are not available for models fit with GEEs.

title 'Maales outcomes -- ITT Analysis - Full set those enrolled';

Below is the code I used:

proc genmod data=maales.itt;

    class subject group (ref='1') cohort/ param=ref;

      model totwocnds= group totwocnd yrs fu3 sessioncount / dist = zip link = log;

      zeromodel malep90s tgp_90s fmalp90s /link = logit;

      *repeated subject = subject(cohort);

       estimate 'group log(RR)' group 1/ exp;

        run;

Is there a way around this?  I tried using Proc Glimmix as well that has a random effect statement to control for clustering of subjects within cohort, however this procedure does not have an option for the ZIP distribution.

Any help would be greatly appreciated. 

Jason

[SteveDenham]

I don't have a good answer, but I know who does and how to find their answers: Head over to the archives for the SAS listserv at

http://www.listserv.uga.edu/cgi-bin/wa?S1=sas-l

and search on ZIP and NLMIXED.  You should find a ton of entries, many of which address your problem.

Steve Denham

[StatDave · Accepted solution]

You would need to do this in PROC NLMIXED by specifying the log likelihood for the ZIP model and then including a RANDOM statement to deal with your clustering.

[Jason]

Thank you for your responses Steve and Dave.  Both were very helpful!!  I was able to pull together the full model for nlmixed ZIP regression with a random effect and much research.

It can be done!! 😃