Table 3.

Discrimination obtained after sixfold cross-validation on the training set (n = 296)

Algorithm	No. of variables included	AUC	Sensitivity	Specificity	PPV	Accuracy	ϕ
GL	12	0.81 (±0.09)	0.72 (±0.2)	0.82 (±0.1)	0.50 (±0.3)	0.82 (±0.1)	0.52 (±0.1)
DRF	21	0.85 (± 0.06)	0.78 (± 0.2)	0.84 (± 0.1)	0.50 (± 0.2)	0.84 (± 0.1)	0.53 (± 0.2)
GBM	28	0.74 (± 0.1)	0.68 (± 0.2)	0.86 (± 0.1)	0.58 (± 0.4)	0.83 (± 0.1)	0.51 (± 0.2)
DL	32	0.84 (± 0.07)	0.70 (± 0.2)	0.87 (± 0.1)	0.60 (± 0.3)	0.85 (± 0.1)	0.54 (± 0.1)

GL, generalized linear modeling; DRF, distributed random forest; GBM, gradient boosting machine; DL, deep learning; AUC, area under the receiver operating characteristic curve; PPV, positive predictive value