Table 3.

Performance of the three predictive models.

Metrics	Model 1 (Rad−VASARI combined)	Model 2 (Rad_score only)	Model 3 (VASARI only)
AUC	0.966	0.937	0.831
Accuracy	0.931	0.912	0.761
Sensitivity	0.954	0.908	0.776
Specificity	0.892	0.919	0.723
PPV	0.939	0.952	0.681
NPV	0.917	0.850	0.596

AUC: area under the receiver operating characteristic curve; PPV: positive predictive value; NPV: negative predictive value.

The AUC cut-off was determined based on Youden index maximization criterion. Specifically, Youden index = true positive rate (sensitivity) – false positive rate (1-specificity). In the ROC curve, a series of Youden indices was calculated, then the maximum Youden index of this series was picked out and the corresponding value of the test variable which matched this maximum Youden index was the cut-off value.