. 2024 Oct 14;11:23. doi: 10.1186/s44156-024-00059-8

Table 2.

Comparison of the predictive performances between two models in the test set

Prediction model	AUC (95% CI)	P value*	Sensitivity† (95% CI)	Specificity† (95% CI)	PPV† (95% CI)	NPV† (95% CI)
Reference model	0.72 (0.57–0.86)	Reference	0.64 (0.43–0.82)	0.80 (0.59–0.94)	0.76 (0.53–0.92)	0.69 (0.49–0.85)
DCNN model + Reference model	0.86 (0.76–0.96)	0.04	0.84 (0.64–0.95)	0.76 (0.55–0.91)	0.78 (0.58–0.91)	0.83 (0.61–0.95)

Prediction model

AUC
(95% CI)

P value*

Sensitivity†
(95% CI)

Specificity†
(95% CI)

PPV†
(95% CI)

NPV†
(95% CI)

Reference model

0.72

(0.57–0.86)

Reference

0.64

(0.43–0.82)

0.80

(0.59–0.94)

0.76

(0.53–0.92)

0.69

(0.49–0.85)

DCNN model + Reference model

0.86

(0.76–0.96)

0.04

0.84

(0.64–0.95)

0.76

(0.55–0.91)

0.78

(0.58–0.91)

0.83

(0.61–0.95)

^*P value was calculated to compare the AUC of the reference model with that of the combined model, using Delong’s test

†The threshold probabilities of the reference model and the combined model to calculate each sensitivity, specificity, PPV, and NPV were 0.51 and 0.45, respectively, according to the cut-off points with the best Youden index

AUC, area under the receiver-operating-characteristic curve; CI, confidence interval; DCNN, deep convolutional neural network; NPV, negative predictive value; PPV, positive predictive value