topic Re: Odds ratios for categorical variables using total as reference in Statistical Procedures

Odds ratios for categorical variables using total as reference

crockerm12 — Tue, 22 Aug 2017 18:35:19 GMT

Hello! I'm trying to produce odds ratios for hospital readmission based upon 22 different categories of disease. I would also like to see the OR for the TOTAL of all disease categories (all 22 added together), and use this as a reference group. I'm not sure how to do this using total as the ref. Any suggestions? Thanks in advance for any assistance.

proc logistic data=index_events descending;
class discat;
model readmit=discat;
weight discwt;
format discat discat.;
run;

Re: Odds ratios for categorical variables using total as reference

Reeza — Tue, 22 Aug 2017 18:38:51 GMT

If you have weights you should use PROC SURVEYLOGISTIC not LOGISTIC.

To compare to all you can replicate the data and have overall as a category in discat but this may affect other calculations. At that point I suspect you need an ESTIMATE statement .

Re: Odds ratios for categorical variables using total as reference

crockerm12 — Tue, 22 Aug 2017 19:37:55 GMT

I have tried creating an overall category in discat but then all observations fall under overall instead of the 22 other categories. Do you have any additional specific advice on how to do that?

Re: Odds ratios for categorical variables using total as reference

Reeza — Tue, 22 Aug 2017 21:22:51 GMT

You have to replicate the data, not just recode it.

data all_comparison;
set sashelp.class;

output;
sex='T'; 
output;
run;

proc freq data=all_comparison;
table sex;
run;

Re: Odds ratios for categorical variables using total as reference

StatDave — Wed, 23 Aug 2017 15:05:56 GMT

What you want to do can be done using the DIFF=ANOM option in the LSMEANS statement in PROC LOGISTIC. See this note.