topic Re: Logistic Regression Modeling Outcomes "1 vs. 0" instead of "0 vs. 1"; DESC n in Statistical Procedures

Logistic Regression Modeling Outcomes "1 vs. 0" instead of "0 vs. 1"; DESC not functioning

darkwob — Mon, 03 Aug 2020 12:01:03 GMT

Hi All,

I am running a logistic regression model with mortality as the dependent variable (event of interest is "1", which equals mortality) and 3 treatments as the independent variable ("0" signifies the patient did not receive therapy, whereas "1" signifies the patient did receive therapy).

When running the model, SAS defaults to reporting the event of "0 vs 1" in the odds ratio plots (see below).

Code:

proc logistic data=WORK.TOCI plots=(oddsratio(cldisplay=serifarrow) );
class Received_Hydroxychloroquine Received_Azithromycin Treatment / param=glm;
model Mortality(event='1')=Received_Hydroxychloroquine Received_Azithromycin
Treatment / link=logit technique=fisher;
run;

I would like to model this analysis based on "1 vs. 0", in other words the predictability of mortality if the patient received therapy. I adjusted the code by adding DESC, and the order seems to have reversed, but the odds ratio plot is still reporting "0 vs. 1" and the point estimates have not changed.

Code:

proc logistic descending data=WORK.TOCI plots=(oddsratio(cldisplay=serifarrow) );
class Received_Hydroxychloroquine Received_Azithromycin Treatment / param=glm;
model Mortality(event='1')=Received_Hydroxychloroquine Received_Azithromycin
Treatment / link=logit technique=fisher;
run;

Is there any additional issue in the code of the program that is causing this to happen, or am I misinterpreting the results?

Thanks in advance,

Matt

Re: Logistic Regression Modeling Outcomes "1 vs. 0" instead of "0 vs. 1"; DESC n

FreelanceReinh — Mon, 03 Aug 2020 12:41:50 GMT

Hi @darkwob and welcome to the SAS Support Communities!

The output says "Probability modeled is Mortality='1'" -- which is what you want and what you requested with the option event='1' in the MODEL statement. So, the descending option in the PROC LOGISTIC statement is basically redundant. Instead, add the descending option of the CLASS statement to turn the odds ratios of the dichotomous CLASS variables around.

class Received_Hydroxychloroquine Received_Azithromycin Treatment / param=glm descending;

Re: Logistic Regression Modeling Outcomes "1 vs. 0" instead of "0 vs. 1"; DESC n

SteveDenham — Mon, 03 Aug 2020 12:47:52 GMT

Response removed by author

(Because it was wrong, basically)

SteveDenham

Re: Logistic Regression Modeling Outcomes "1 vs. 0" instead of "0 vs. 1"; DESC n

darkwob — Mon, 03 Aug 2020 12:46:21 GMT

Great, this solved the issue, much appreciated!