Hello Odunayo David Adeniyi I do not think so.

The ROC curve workspace is Column and the regression (linear or nonlinear, it does not matter) is in an XY workspace.

In other words, the ROC curve is used for comparison and regression to measure the relationship and prediction.

It's quite possible, especially in R. For details see the attached screenshots and the corresponding Google searches.for full details.. Best wishes, David Booth

David Eugene Booth : thank you for the great links.

Odunayo David Adeniyi : apart from what was said above, let me add the following.

Usually, ROC is not used for regression. Instead, you can use the prediction-realization diagram. There's also a variant we called the pooled prediction-realization diagram (PPRD) where results from different methods can be plotted, see page 64 of the following publication:

Article Assessing Point Forecast Bias Across Multiple Time Series: M...

Hello,

of course it can be done.

But not in one step.

First you need to create a new predictor using prog reg or proc logistic. Then you have not only inputs of the new predictor, but also point estimates of all its coefficients.

In the data step you can then calculate the predictor (derived before via proc reg or proc logistic - both these procedure allow to store the derived predictor also in the dataset).

Then proc logistic follows with ROC statement si a final step.

E.g. new risk score is

New_risk_score= 4.267 * PAR5 + 1.150 * PAR4 + 2.002 * PAR3 + 1.589 * PAR2 + 2.561 * PAR1 ;

Then

proc logistic data=XXX;

model Event (REF='0') = New_risk_score / expb CLODDS=WALD ;

roc;

RUN;

will create requested ROC.

If e.g. parameter PAR5 is the strongest single predictor of Event then following code compare statistically the new risk score with PAR5

proc logistic data=XXX;

model Event (REF='0') = New_risk_score PAR5 / nofit expb CLODDS=WALD ;

roc 'New risk score' New_risk_score;

roc 'Parameter 5' PAR5;

roccontrast reference( 'Parameter 5') /estimate ;

RUN;

Best regards,

LP

As David Eugene Booth explained very thoroughly typically the multiple predictors are combined using logistic regression and the predictions are input to the ROC curve (though in theory there are alternatives Article Combining Predictors for Classification Using the Area under...

A prediction model should be assessed by calibration as well to assess accuracy of individual predictions though as well as ROC/discrimination - more info here https://www.youtube.com/watch?v=v_e85TDYj9Y and Article Calibration: The Achilles heel of predictive analytics