topic Re: ROC curve in Statistical Procedures

ROC curve

Teresa12 — Tue, 10 Aug 2021 13:58:16 GMT

I need to create a ROC curve.

My data looks like :

data Data1;
input id score diagnosis;
datalines;
1 0.1 0
1 0.2 0
1 1.0 1
1 0.8 0
1 0.9 0
2 1.8 1
2 1.2 1
3 1.7 1
3 1.1 0
4 0.1 0
4 1.9 0
4 1.8 1
4 1.2 1
4 2.0 0
5 0.7 0
5 1.2 1

;

I have several tests per ID.
My sas code following

ods graphics on;
proc logistic data=data1 plots(only)=(roc(id=obs) effect) descending;
model diagnosis(event='1')=score / ctable outroc=RocStats ;
ROC; ROCCONTRAST;
run;
ods graphics off;

My question: I don't know how to integrate the repeated ID into my model.
I use only diagnosis and score. O am confused about this issue

Thanks:

Re: ROC curve

gcjfernandez — Tue, 10 Aug 2021 18:09:50 GMT

Unfortunately, Proc Logistic is not appropriate for generating ROC chart with repeated measures. You need to use GLIMMIX to account for repeated measures. But there is no option available in GLIMMIX to generate ROC chart. Please check this paper where the author has included a SAS macro for generating ROC using SAS IML.

http://ferran.torres.name/download/shared/roc/roc.pdf.

Re: ROC curve

StatDave — Tue, 10 Aug 2021 19:45:09 GMT

You can use the predicted probabilities from any model on a binary response in PROC LOGISTIC to generate an ROC curve. See this note. If you want to use a GEE model to handle the repeated measures allowing an exchangeable correlation matrix, then you could use these statements.

proc gee data=data1;
class id;
model diagnosis(event='1')=score / dist=bin ;
repeated subject=id / type=exch;
output out=out p=p;
run;
proc logistic data=out;
model diagnosis(event="1")= / nofit;
roc pred=p;
run;

Re: ROC curve

Teresa12 — Tue, 10 Aug 2021 20:17:58 GMT

Thank you very much for your answer!!!