Table 2.

Predictive performance of each model.

Model and data set		AUROCa, mean (95% CI)	Sensitivity, mean (95% CI)	Specificity, mean (95% CI)	Positive predictive value, mean (95% CI)	Negative predictive value, mean (95% CI)
Random forest
	Internal data set	0.916 (0.916-0.916)	0.904 (0.904-0.905)	0.746 (0.744-0.747)	0.579 (0.578-0.580)	0.953 (0.952-0.953)
	External data set	0.721 (0.720-0.721)	0.910 (0.909-0.912)	0.270 (0.266-0.273)	0.159 (0.159-0.160)	0.952 (0.951-0.953)
XGBoostb
	Internal data set	0.919 (0.919-0.919)	0.904 (0.904-0.905)	0.731 (0.729-0.732)	0.565 (0.563-0.566)	0.952 (0.952-0.952)
	External data set	0.697 (0.695-0.699)	0.908 (0.906-0.909)	0.250 (0.245-0.255)	0.156 (0.155-0.156)	0.946 (0.945-0.947)
Deep neural network
	Internal data set	0.881 (0.878-0.884)	0.906 (0.905-0.907)	0.622 (0.608-0.635)	0.485 (0.477-0.492)	0.944 (0.943-0.945)
	External data set	0.655 (0.654-0.657)	0.907 (0.905-0.908)	0.197 (0.192-0.201)	0.147 (0.146-0.147)	0.932 (0.931-0.933)
Logistic regression
	Internal data set	0.875 (0.875-0.875)	0.901 (0.901-0.901)	0.605 (0.605-0.605)	0.469 (0.469-0.469)	0.940 (0.940-0.940)
	External data set	0.631 (0.631-0.631)	0.904 (0.904-0.904)	0.155 (0.155-0.155)	0.140 (0.140-0.140)	0.914 (0.914-0.914)

^aAUROC: area under the receiver operating characteristic curve.

^bXGBoost: extreme gradient boosting.