Table 2.

Diagnostic performance of new biomarker panel.

Marker	Training and Test Set				Validation Set
	AUC	Accuracy (%)	Specificity (%)	Sensitivity (%)	AUC	Accuracy (%)	Specificity (%)	Sensitivity (%)
RF
Panel	0.992	95	95	96	0.993	95	96	93
CA19-9 + CEA	0.924	90	95	76	0.960	92	94	83
CA19-9	0.921	90	95	74	0.960	90	94	78
CEA	0.666	77	95	20	0.797	78	95	25
GLM
Panel	0.983	94	95	92	0.983	94	95	92
CA19-9 + CEA	0.852	87	95	62	0.928	91	94	80
CA19-9	0.848	88	95	66	0.923	92	94	83
CEA	0.732	78	95	24	0.814	80	96	28
GLM + RF
Panel	0.984	94	95	92	0.986	95	96	92
CA19-9 + CEA	0.934	91	95	78	0.962	91	93	87
CA19-9	0.933	90	95	75	0.964	90	94	80
CEA	0.732	78	95	24	0.814	80	96	28
RIDGE
Panel	0.987	95	95	93	0.985	95	96	92
CA19-9 + CEA	0.852	87	95	62	0.928	91	94	80
CA19-9	0.848	88	95	67	0.924	92	94	83
CEA	0.732	78	95	24	0.816	80	96	28
SVM
Panel	0.990	95	95	95	0.991	97	98	92
CA19-9 + CEA	0.900	89	95	71	0.964	92	97	77
CA19-9	0.912	88	95	68	0.967	92	96	77
CEA	0.627	78	95	25	0.692	78	95	27

AUC, area under the curve; RF, random forest; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; GLM, generalized linear model; SVM, support vector machine.