PROC MIANALYZE procedure after PROC GLIMMIX binary outcome (NO odds ratio and CL)

[mtechnologist]

Hi SAS users,

I used Multiple imputation with GLIMMIX for binary outcome variable with one binary independent variable (as example).  When I ran PROC MIANALYZE, the results only showed the estimate and SE but did not show 95% CL nor the significance (attach pic of the table). Also did not provides the odds ratio table.    

I used the following code:

PROC GLIMMIX DATA=shis.shis_mi3 METHOD=QUAD empirical=classical;
CLASS rgn_code SEX1 (ref=FIRST);
MODEL PA (EVENT=LAST)= SEX1 /CL DIST=BINARY LINK=LOGIT SOLUTION
ODDSRATIO (DIFF=LAST LABEL); 
RANDOM INTERCEPT / SUBJECT=rgn_code S CL TYPE=VC;
by imputation_;
nloptions gconv=0;
ods output ParameterEstimates=PAparm1;
COVTEST /WALD;
RUN;

/* Prepare pooled estimates for MI (excluding original unimputed data) */
data imputed;
set PAparm1;
_imputation_=imputation_;
If _imputation_>0;

/* Calculate and print pooled estimates for MI */
proc mianalyze parms = imputed;
CLASS SEX1;
modeleffects Intercept SEX1;
run;

[SAS_Rob · Accepted solution]

You likely have a WARNING message in your LOG regarding the between imputation variance being zero.  This essentially means that the estimates for each of the imputations is identical.  When the between imputation variance is zero then the number of Degrees of Freedom is undefined so you cannot get a confidence interval or p-values.
Unfortunately, there is not a good approach to take in this case. This is one of the limitations of multiple imputation in general. You might try doing a literature search to see if you can find any references that deal with this issue.

Because GLIMMIX does not report a standard error for the odds ratio directly, you would not be able to combine those either.  I suspect you would run into the same issue with the between imputation variance anyway. 

Normally you would combine the log (OR), that is the parameter estimates, and then exponentiate those values in a subsequent data step.

[mtechnologist]

Well explained Rob. Thanks.