Table 5. External validation performances of the models in the hold-out cohort.

	Logistic regression	Neural networks
AUC	0.859 (0.836 to 0.883)	0.856 (0.831 to 0.880)
Sensitivity	82.5% (77 to 87.2)	78.2% (72.3 to 83.3)
Specificity	71.6% (68.3 to 74.7)	76.4% (73.3 to 79.3)
Positive predictive value	45.5% (42.5 to 48.7)	48.8% (45.2 to 52.3)
Negative predictive value	93.4% (91.5 to 95)	92.4 (90.5 to 94)
Positive likelihood ratio	2.91 (2.57 to 3.3)	3.31 (2.87 to 3.82)
Negative likelihood ratio	0.24 (0.18 to 0.32)	0.29 (0.23 to 0.37)
Accuracy	74.1% (71.3 to 76.7)	76.8% (74.1 to 79.3)