TABLE 3.

Predictive performance of different models using the testing (i.e., 20%) set.

Outcome variables	Model types	AUC	Accuracy	PPV	NPV
DN	XF	0.902 ± 0.040	0.862 ± 0.054	0.699 ± 0.149	0.917 ± 0.050
	CHAID	0.699 ± 0.074	0.699 ± 0.067	0.331 ± 0.158	0.780 ± 0.068
	BN	0.744 ± 0.078	0.916 ± 0.057	0.930 ± 0.091	0.914 ± 0.066
	D	0.823 ± 0.055	0.720 ± 0.063	0.459 ± 0.132	0.895 ± 0.064
DPN	XF	0.847 ± 0.081	0.783 ± 0.080	0.642 ± 0.123	0.882 ± 0.073
	CHAID	0.787 ± 0.081	0.757 ± 0.054	0.680 ± 0.143	0.807 ± 0.070
	QUEST	0.720 ± 0.060	0.766 ± 0.056	0.716 ± 0.186	0.805 ± 0.057
	D	0.859 ± 0.050	0.843 ± 0.038	0.775 ± 0.092	0.885 ± 0.055
DA	XF	0.889 ± 0.059	0.851 ± 0.051	0.684 ± 0.129	0.899 ± 0.045
	CHAID	0.764 ± 0.087	0.769 ± 0.049	0.481 ± 0.229	0.842 ± 0.066
	CRT	0.797 ± 0.068	0.802 ± 0.058	0.671 ± 0.207	0.836 ± 0.064
	D	0.825 ± 0.070	0.808 ± 0.065	0.568 ± 0.150	0.907 ± 0.056
DED	ANN	0.725 ± 0.142	0.812 ± 0.091	0.083 ± 0.180	0.864 ± 0.080
	CHAID	0.818 ± 0.161	0.875 ± 0.053	0.523 ± 0.346	0.916 ± 0.050
	BN	0.749 ± 0.179	0.978 ± 0.031	0.867 ± 0.322	0.984 ± 0.028
	D	0.832 ± 0.086	0.799 ± 0.055	0.328 ± 0.156	0.989 ± 0.025
HbA1c	ANN	0.604 ± 0.103	0.760 ± 0.094	0.375 ± 0.460	0.825 ± 0.089
	BN	0.825 ± 0.092	0.728 ± 0.083	0.417 ± 0.180	0.840 ± 0.120

Data are mean ± SD.

XF, ensemble model; ANN, artificial neural network; CRT, classification and regression tree; QUEST, quick unbiased efficient statistical tree; D, discriminate; BN, Bayesian network; DN, diabetic nephropathy; DPN, diabetic peripheral neuropathy; DA, diabetic angiopathy; DED, diabetic eye disease; HbA1c, glycosylated hemoglobin A; AUC, area under the receiver operating characteristic curve; PPV, positive predictive value; NPV, negative predictive value.