SAS Macro Proc Logistic将P值放入数据集中

Question

I've googled lots papers on the subject but don't seem to find what I want. I'm a beginner at SAS Macro, hoping to get some help here. Here is what I want:

I have a dataset with 1200 variables. I want a macro to run those 1199 variables as OUTCOME, and store the P-values of logistic regression in a dataset. Also the dependent variable "gender" is character, and so are the outcome variables. But I don't know how to put class statement in the macro. Here is an example of how I run it as a single procedure.

 proc logistic data=Baseline_gender ;
 class gender(ref="Male") / param=ref;
 model N284(event='1')=gender ; 
 ods output ParameterEstimates=ok;
 run;

My idea was to create ODS output and delete the unnecessary variables other than the P-value and merge them into one dataset according to the OUTCOME variable names in the model: eg

 Variable P-value
 A1       0.005
 A2       0.018
 ..       ....

I tried to play with some proc macro but I just cant get it work!!! I really need help on this, Thank you very much.

Answer 1

SRSwift might be onto something (don't know enough about his method to tell), but here's a way to do it using a macro.

First, count the number of variables in your dataset. Do this by selecting your table from the dictionary.columns table. This puts the number of variables into &sqlobs . Now read the variable names from the dictionary table into macro variables var1-var&sqlobs .

%macro logitall;
proc sql;
create table count as
select name from dictionary.columns
where upcase(libname) = 'WORK'
  and upcase(memname) = 'BASELINE_GENDER'
  and upcase(name) ne 'GENDER'
;

select name into :var1 - :var&sqlobs
from dictionary.columns
where upcase(libname) = 'WORK'
  and upcase(memname) = 'BASELINE_GENDER'
  and upcase(name) ne 'GENDER'
;
quit;

Then run proc logistic for each dependent variable, each time outputting a dataset named after dependent variable.;

%do I = 1 %to &sqlobs;
  proc logistic data=Baseline_gender ;
    class gender(ref="Male") / param=ref;
    model &&var&I.(event='1')=gender ; 
    ods output ParameterEstimates=&&var&I.;
  run;
%end;

Now put all the output datasets together, creating a new variable with the dataset name using indsname= in the set statement.

data allvars;
  format indsname dsname varname $25.; 
  set
  %do I = 1 %to &sqlobs;
    &&var&I.
  %end;
  indsname=dsname;
  varname=dsname;
  keep varname ProbChiSq;
  where variable ne 'Intercept';
run;
%mend logitall;

%logitall;

Answer 2

Here is a macro free approach. It restructures the data in advance and uses SAS's by grouping. The data is stored in a deep format where the all the outcome variable values are stored in one new variable.

Create some sample data:

data have;
   input 
        outcome1 
        outcome2 
        outcome3 
        gender $;
   datalines;
1 1 1 Male
0 1 1 Male
1 0 1 Female
0 1 0 Male
1 1 0 Female
0 0 0 Female
;
run;

Next transpose the data into a deep format using an array:

data trans;
    set have;
    /* Create an array of all the outcome variables */
    array o{*} outcome:;
    /* Loop over the outcome variables */
    do i = 1 to dim(o);
        /* Store the variable name for grouping */
        _NAME_ = vname(o[i]);
        /* Store the outcome value in the  */
        outcome = o[i];
        output;
    end;
    keep _NAME_ outcome gender;
run;
proc sort data = trans;
    by _NAME_;
run;

Reusing your logistic procedure but with an additional by statement:

proc logistic data = trans;
    /* Use the grouping variable to select multiple analyses  */
    by _NAME_;
    class gender(ref = "Male");
    /* Use the new variable for the dependant variable */
    model outcome = gender / noint; 
    ods output ParameterEstimates = ok;
run;

Answer 3

Here is another way to do it using macro. First define all the variables to be used as outcome in a global variable and then write the macro script.

%let var = var1 var2 var3 ..... var1199;

%macro log_regression;
  %do i=1 %to %eval(%sysfunc(countc(&var., " "))+1);
    %let outcome_var = %scan(&var, &i);
    %put &outcome_var.;

    proc logistic data = baseline_gender desc;
    class gender (ref = "Male") / param = ref;
    model &outcome_var. = gender;
    ods output ParameterEstimates = ParEst_&outcome_var.;
    run;

    %if %sysfunc(exist(univar_result)) %then %do;
      data univar_result;
      set univar_result ParEst_&outcome_var.;
      run;
    %end;
    %else %do;
      data univar_result;
      set ParEst_&outcome_var.;
      run;
    %end;

  %end;
%mend;