how to use the macro after the imputation?

[Lucai_sister]

Hi, guys. I would like to study the mean difference in the number of relapse among the three treatment groups. Because the data set was partially missing, after using multiple imputation, 100  imputation sets were obtained.

After using pro genmod analysis, I want to use the nlmeans macro to compare the comparison results of each imputation set.  But the following result is prompted. Do you know why?

here is my data

imputation      id      relapse_count    lntime   group

1                          1             0                        -1             1

1                          2             2                       -3             2

.....

100                    33            3                       -4             3

here is my SAS code:

stype1:
proc genmod data=relapse;
class group(ref="1") id/param=glm;
model relapse_count=group/link=log dist=negbin offset=lntime;
repeated subject=id;
lsmeans group/ilink cl e;
estimate 'group2 vs 1 of RR' group -1 1 0/e;
estimate 'group1 vs 3 of RR' group 1 0 -1/e;
estimate 'group2 vs 3 of RR' group 0 1 -1/e;
ods output coef=coeffs;
store out=insmodel;
by _Imputation_;
run;

stype2:
%include 'path\nlest.sas';
%include 'path\nlmeans.sas';
%NLMeans(instore=insmodel, coef=coeffs, link=log, title=Difference of relapse Rates)

Here is the error:

here is my coeffs:

Am I missing a procedure? I really want researchers to help me.

[sbxkoenk]

Hello,

Here's how you should use multiple imputation.

Multiple imputation inference involves three distinct phases:

1. The missing data are filled in m times to generate m complete data sets or by-groups. (PROC MI)
2. Perform regression or any other analysis on each of the m complete data sets or by-groups. (PROC GENMOD)
3. Average the values of the parameter estimates across the M samples to produce a single point estimate. (PROC MIANALYZE)

After 3. you can run the macro's you need!

See also here:

Multiple Imputation for Missing Data: Concepts and New Development (Version 9.0)
Yang C. Yuan, SAS Institute Inc., Rockville, MD
https://support.sas.com/rnd/app/stat/papers/multipleimputation.pdf

https://support.sas.com/rnd/app/stat/topics/multiple-imputation.html

[EDIT] @Lucai_sister 
My answer is not entirely correct, I believe.
I am sure my colleagues @SAS_Rob and @StatDave know better, so please follow their advice.
See below for their replies!

Koen

[Lucai_sister]

I really appreciate your advice!

[SAS_Rob]

The %NLMEANS macro does not have a BY statement which is what you would need to make this work.  My suggestion would be to run the macro for each _imputation_ independently and then combine the estimates and standard errors into a single data set.  Once you do that you should be able to use Proc MIANALYZE to get the final combined estimates.

[Lucai_sister]

Thank you so much!

[StatDave · Accepted solution]

As Rob suggests, to get a final combined rate estimate you need to run NLMeans on each imputation data set and then combine the results. BY processing can be done by using the NLMeans macro in the RunBY macro. This is discussed in the Details section of the NLMeans macro documentation and is illustrated in the last example in the Results tab. In your case, you should drop the ESTIMATE statements since only the LSMEANS statement is needed. Then, after downloading the RunBY macro and defining it in your session using a %INC statement, you should just need to add the following after your GENMOD step. I also suggest you make sure you are using the latest versions of the NLMeans and NLEST macro (which NLMeans calls). You can download the latest versions from support.sas.com.

%macro code;
  %nlmeans(instore=insmodel, coef=coeffs, link=log, where=&_BY1=&_LVL1, 
           options=nonames noprint, title=rate diffs)
  proc append base=alldiff data=est; run;
%mend;
%runby(data=relapse, by=_imputation_)

data alldiff; 
  set alldiff;  
  _imputation_=_n_;
  run;
proc mianalyze data=alldiff;
   modeleffects estimate;
   stderr standarderror;
   run;

[Lucai_sister]

The code works fine. Thank you so much!