Issue with obtaining relative estimates after multiple imputation - multinomial regression

[SG1993]

Hi All,

I've used the below code for multiple imputation. After doing imputation, I'm running multinomial logistic regression then running proc analyse. And getting only pooled/combined estimates. But I need relative estimates.

Thank you.
.
.

proc mi data=data.newdata seed=876 nimpute=5 out=outfcs;

class group sex hispan;

fcs nbiter=40 logistic (group/details);

var group sysbp01 diabp01 a1c_01 chol_01 bmi_01 sex hispan;

run;

proc logistic data = outfcs;

class group (ref = "3") sex (ref = "0") hispan (ref= "1") / param = ref;

model group = sysbp01 diabp01 a1c_01 chol_01 bmi_01 age01 sex hispan / link = glogit covb;

by _Imputation_;

ods output ParameterEstimates=lgsparms CovB=lgscovb;

run;

proc mianalyze parms (classvar=classval)=lgsparms;

class sex hispan;

modeleffects intercept sysbp01 diabp01 a1c_01 chol_01 bmi_01 age01 sex hispan ;

run;

[StatDave · Accepted solution]

When you say "relative estimates," I assume that you mean relative risk estimates. If so, then you can do this with the NLMeans and RunBY macros. See this note that shows how you can add LSMEANS and STORE statements in PROC LOGISTIC followed by the NLMeans macro to obtain relative risk estimates from a multinomial model. To do this for each of your multiple imputations, you will use the RunBY macro to run the NLMeans macro repeatedly over the imputations. You can then combine the relative risk estimates using PROC MIANALYZE. This is discussed and illustrated in the NLMeans macro documentation. See "Analysis of imputed data" in Example 2 in the Results tab of the macro documentation.