topic Re: How to code for many regression models with repeating variables in SAS Programming

How to code for many regression models with similar variables

TJ87 — Wed, 31 Jul 2013 23:25:36 GMT

Dear SAS Community,

I'm trying to calculate logistic regression odds ratios (associations) between 20 different chemicals and 9 different cancer sites in my dataset.

I would like to use macro language or proc sql to efficiently code for the 180 logistic regression models.

My variables are person id (ID), sites A-I (site), and chemical exposures 1-20 (exp). Here is an idea of what my data looks like. The variables for cancer sites are coded as dummy variables (1= case, 0=non-case).

The ellipsis (...) indicates that I'm only showing 1 cancer site variable instead of all 9, only 1 chemical out of 20, and only 5 subjects from the whole study.

ID SiteA ... Exp1 ...

1 1 10.5

2 0 0.5

3 0 6

4 0 1.3

5 1 12.2

...

My basic model looks like this:

proc logistic data = analysis;

model &site (event="1")= &exp; *Odds ratios for chemicals as continuous variables;

run;

Could someone please suggest how I can write concise code for all these tests using the same underlying model? Thank you.

Re: How to code for many regression models with repeating variables

Scott_Mitchell — Thu, 01 Aug 2013 01:03:05 GMT

Hi TJ,

I haven't tested the attached as you haven't provided any data, but hopefully it makes sense. If you have each of your scenarios in a dataset, which I assume you can achieve on your own, using a Call Execute will allow for you to use values in the relevant variables as part of your statements, functions and the like.

DATA _NULL_;

SET SAMPLES (END=EOF);

STR =COMPBL( "PROC LOGISTIC DATA = HAVE; MODEL "|| SITE || " (EVENT='1')=" || EXP || ";";

CALL EXECUTE (STR);

IF EOF THEN DO;

STR = 'RUN;';

CALL EXECUTE (STR);

END;

RUN;

The following gives you a simple example of code that works. We are accessing the Dictionary View VCOLUMN to identify all the columns in SASHELP.CLASS and executing a sort on each of the variables.

DATA _NULL_;

SET SASHELP.VCOLUMN (WHERE =(LIBNAME = "SASHELP" AND MEMNAME = "CLASS")) END=EOF;

STR =COMPBL( "PROC SORT DATA = SASHELP.CLASS OUT = SORTCLASS ; BY " || NAME || ";");

CALL EXECUTE (STR);

IF EOF THEN DO;

STR = 'RUN;';

CALL EXECUTE (STR);

END;

RUN;

Re: How to code for many regression models with repeating variables

TJ87 — Mon, 20 Jan 2014 18:58:30 GMT

Thank you very much for your suggested code. I found this link to be especially helpful for testing multiple independent variables. The authors also run the 'call execute' subroutine that you recommend.

Re: How to code for many regression models with similar variables

Rick_SAS — Tue, 21 Jan 2014 14:58:36 GMT

Perhaps I'm missing something, but if you just want 9 different independent analyses, transform the data into a long data set with SITE=A-I and use a BY statement with SITE as the BY variable.

If you want to specify a single model that incorporates the sites, you can specify SITE as a CLASS variable and include interactions between the variables and the sites.

Lastly, if SITE is a random variable, rather than a fixed effect, you might want to look at using PROC GLIMMIX instead of PROC LOGISTIC. has posted many examples of using PROC GLIMMIX.