topic Re: ROC curves comparison of the AUC's in Statistical Procedures

ROC curves comparison of the AUC's

FGAB — Mon, 03 Feb 2020 21:39:10 GMT

I hope you can help me. I cant figure it out..

I have tested the diagnostic ability of four different biomerkers in obese and in non-obese patients using ROC curves and AUC's.

Now I want to compare the diagnostic value of each biomarker, respectively, in obese vs no-obese patients.

How can I do this?

I have used this code, when comparing the diagnostic value of the biomarkers in obese and non-obese, respectively.

proc logistic data=xxxxxxxx plots(only)=roc rocoptions(id=prob) ;

class nygroup (ref="0") obesity ;

model nygroup (event='1') = obesity biomarker1 biomarker2 biomarker3 biomarker4/nofit outroc=roc;

by obesity;

roc 'Biomarker1' Biomarker1 ;

roc 'Biomarker2' Biomarker2;

roc 'Biomarker3' Biomarker3;

roc 'Biomarker4' Biomarker4;

ROCCONTRAST reference('Biomarker1) / estimate e ;

run;

Re: ROC curves comparison of the AUC's

Norman21 — Sun, 09 Feb 2020 21:09:28 GMT

It looks like you are missing a single quotation mark (') in the ROCCONTRAST statement.

Re: ROC curves comparison of the AUC's

StatDave — Mon, 17 Feb 2020 18:31:30 GMT

If you have two sets of independent subjects, obese and non-obese, then you can fit separate curves for those two groups and compare the independent ROC curves. To fit the separate curves, specify your OBESE predictor in a BY statement rather than in the CLASS and MODEL statements (also, you should avoid specifying the response in the CLASS statement - it is unnecessary). Then use the method discussed in this note to compare the ROC curves of the independent groups.