Exact GEE logistic regression

[iuri_leite]

Dear Friends,

I am writing to ask if there is anyway for me to run a exact GEE logistic regression in SAS?

I am running it using PROC GENMOD.

PROC GENMOD DESCENDING;
CLASS COPID WEEK (REF='0') COUNTRY (REF='BR');
MODEL HEPAT_B=WEEK COUNTRY/ DIST=BIN LINK=LOGIT;
REPEATED SUBJECT=COPID/TYPE=UN;
ESTIMATE 'TIME' WEEK 1 -1/EXP;
RUN;

WEEK is a dichotomous variable and when WEEK IS 1 HEPAT_B is always zero. When a use a logistic exact model I am able to estimate the effect of week. But when I run the GEE logistic, the model do not estimate the effect of week.

I could not see any option for exact logistic.

Thank you in advance.

Regards,

Iuri

[Ksharp]

I never heard that model.

Maybe @SteveDenham  @StatDave  knows ?

[SteveDenham]

If WEEK is dichotomous, you might try removing it from the CLASS statement to see if that addresses your issue.  I am not familiar with any algorithms that apply exact tests in generalized estimating equations.

SteveDenham

[StatDave]

Sparseness in the data like that typically cause problems in any modeling approach that uses an iterative fitting algorithm like maximum likelihood or GEE. In such cases, you might be able to get what you want using a stratified, conditional model or a nonmodeling approach. Using stratification, you could try using a conditional model and exact test in PROC LOGISTIC or using a nonmodeling approach with an exact analysis in PROC MULTTEST. These are both illustrated in this note.

[iuri_leite]

Dear Dave,

I had tried this before. It was your suggestion.

Many thanks.

Regards,

Iuri