Help using Base SAS procedures

Need to label ROC curve with cutoff value

Reply
Occasional Contributor
Posts: 9

Need to label ROC curve with cutoff value

I need an ROC curve that labels the observations with the actual value of a variable, not the observation number. I am plotting the curve for a model of mortality as a function of an injury score, so I would like to know what injury score has the highest sensitivity/specificity for my model. I am interested in seeing what cutoff I can use to predict 90% or 95% mortality.  I have tried the following:

 

 

ods graphics on;
proc logistic data=mt2.elev plots=roc;
model mortality(event='1')= issloc/outroc=myroc;run;
ods graphics off; 
proc gplot data=myroc; symbol1 i=join v=circle;
plot _sensit_*_1mspec_; run;quit;

I also used some code from a class I took to make this table, but it doesn't have the value for injury score which is what Im looking for.

SAS Output

Probability Level No. of CorrectlyPredicted Events No. of CorrectlyPredicted Nonevents No. of NoneventsPredicted asEvents No. of EventsPredicted asNonevents Sensitivity 1 - Specificity
0.018955144490989530.9795920.154204
0.021067144495085430.9795920.14714
0.023408143498781740.9727890.140765
0.026002142500979550.9659860.136975
0.028875142509271250.9659860.122674
0.0355731375211593100.9319730.102171
0.039461345266538130.9115650.092695
0.0437541265314490210.8571430.084425
0.0484911155415389320.7823130.067023
0.059461115457347360.7551020.059786
0.0657811035530274440.700680.047209
Ask a Question
Discussion stats
  • 0 replies
  • 254 views
  • 0 likes
  • 1 in conversation