topic Re: Find optimal threshold using ROC Curve in Statistical Procedures

Find optimal threshold using ROC Curve

mariko5797 — Fri, 20 Jan 2023 22:18:46 GMT

data HAVE;
 input ID Test1 Test1Res Test2 Test2Res @@;
 cards;
A 1 0.5 1 1.23
A 1 0.8 1 1.97
A 0 . 1 1.54
B 1 0.6 1 1.20
B 1 0.7 0 .
B 1 0.9 1 1.21
C 0 . 0 .
C 0 . 0 .
C 1 0.5 0 .
D 0 . 0 .
D 1 0.8 0 .
D 0 . 1 1.29
;
run;
/*ID = Subject Identifier*/
/*Test1 = Primary test result positive (1) or negative (0)*/
/*Test1Res = Primary test result*/
/*Test2 = Secondary test result positive (1) or negative (0)*/
/*Test2Res = Secondary test result*/

proc logistic data= HAVE descending plots(only)= roc;
 model Test1=Test2Res / outroc= ROC;
run;
/*ROC dataset contains Probability Level, No. of TP, FP, TN, and FN, and the Sensitivity and 1-Specificity*/

To find the optimal point, I would want to find when sensitivity is the greatest and 1-specificity is the lowest, i.e. the highest Probability Level or sum(Sens, Spec), correct? From here, how would I find which threshold level the sensitivity and specificity are associated with?

Re: Find optimal threshold using ROC Curve

StatDave — Fri, 20 Jan 2023 22:21:06 GMT

What you want is called the Youden index. When you are looking for a particular statistic in SAS, check the list of Frequently Asked-for Statistics (FASTats) in the Important Links section of the Statistical Procedures Community page. For you case, see the ROC curve and Youden index items in FASTats.

Re: Find optimal threshold using ROC Curve

Ksharp — Sat, 21 Jan 2023 11:39:52 GMT

https://communities.sas.com/t5/Statistical-Procedures/Calculating-Optimal-Cutoff-Point-for-Repeated-Measurement-data/m-p/843880#M41817