topic How I get pooled Dunnett adjusted confidence intervals calculated from multiple imputed datasets? in Statistical Procedures

How I get pooled Dunnett adjusted confidence intervals calculated from multiple imputed datasets?

Wenqing_Jiang — Wed, 09 Dec 2020 09:44:06 GMT

I hope to get pooled Dunnett adjusted confidence intervals from multiple imputed datasets. I first create 100 imputed datasets using PROC MI. Then I get the Dunnett adjusted CIs for treatment effect difference of A vs. C and B vs. C using Estimate statement with adjust=dunnett option in PROC MIXED. But it seems it’s not feasible to use PROC MIANALYZE to poll the Dunnett adjusted upper/lower limits or request an Dunnett adjustment to the CIs from PROC MIANALYZE？Can anyone give me some advice? Many thanks！

Below is my SAS code:

PROC MI data=toimp out=anami seed=124695 nimpute=100;
by TRT ;
var BASE w3 w5 w7 w9 w13 w17;
EM maxiter=1000;
mcmc chain=multiple impute=full;
run;

PROC MIXED data=anami;
class TRT;
model week17 = TRT BASE / solution;
lsmeans TRT/ADJUST=DUNNETT cl pdiff=control("C");
ods output Diffs=lsdiffs LSMeans=lsm solutionF=Parms;
by _Imputation_;
run;

PROC MIANALYZE parms(classvar=full)=lsdiffs;
class TRT;
modeleffects TRT;
ods output ParameterEstimates=MIAN_lsdiffs;
run;

Re: How I get pooled Dunnett adjusted confidence intervals calculated from multiple imputed datasets

SteveDenham — Wed, 09 Dec 2020 14:33:03 GMT

Hard way, but more likely to be correct, is to get the pooled differences and standard errors of the differences from MIANALYZE, and then do the calculation of the pooled confidence interval in a DATA step. You know the number of comparisons, so you know the critical Dunnett value to use in constructing the CI.

I don't trust any of the "easier" ways I was thinking of...

SteveDenham

Re: How I get pooled Dunnett adjusted confidence intervals calculated from multiple imputed datasets

Wenqing_Jiang — Mon, 21 Dec 2020 03:09:35 GMT

Thank you Steve. I tried to do some literature search but it was scoreless. It is a hard way and calls for more theoretical work in the future.