Table 2.

Performance of type 2 diabetes risk models via various classifiers.

Classifier	Prediction modela	Discrimination		BS	NRI (95% CI)c %
		AUC (95% CI)	ΔAUC (95% CI)b
CPH	Conventional model	0.815 (0.795 to 0.833)	-	0.053	-
	Conventional + GRS model	0.815 (0.796 to 0.833)	0.001 (−0.009 to 0.010)	0.133	41.2 (27.8 to 54.1)
ANN	Conventional model	0.816 (0.797 to 0.834)	-	0.044
	Conventional + GRS model	0.818 (0.799 to 0.836)	0.002 (−0.015 to 0.019)	0.045	41.0 (25.1 to 52.7)
RF	Conventional model	0.843 (0.825 to 0.860)	-	0.041	-
	Conventional + GRS model	0.861 (0.844 to 0.877)	0.018 (0.002 to 0.034)	0.040	46.4 (35.2 to 57.6)
GBM	Conventional model	0.851 (0.834 to 0.868)	-	0.033	-
	Conventional + GRS model	0.885 (0.869 to 0.899)	0.033 (0.001 to 0.065)	0.033	45.1 (18.0 to 57.7)

Bold values represent P < 0.05. CPH, Cox proportional hazards regression model; ANN, artificial neural network; RF, random forest; GBM, gradient boosting machine; GRS, genetic risk score; AUC, the area under receiver operating characteristic curve; BS, brier score; NRI, net reclassification improvement; CI, confidence interval.

^aConventional model included FPG, WC, TG, parental history of diabetes, and hypertension.

^bΔAUCs were the differences of AUCs among conventional-genetic-combined models and conventional model.

^cNRI showed the reclassification of the prediction models with genetic risk scores compared to conventional model.