Table 3

Sensitivity and specificity of best model and AUC on each ROC curve

	AUC	Best model on ROC curve			AUC	Best model on ROC curve
	AUC	Sensitivity	Specificity		AUC	Sensitivity	Specificity
RF model1	0.719	0.714	0.617	MLR model1	0.699*	0.648	0.648
RF model2	0.716	0.608	0.720	MLR model2	0.711	0.648	0.668
RF model3	0.743	0.607	0.778	MLR model3	0.734	0.629	0.729
RF model4	0.864	0.804	0.823	MLR model4	0.817*	0.748	0.773
RF model5	0.940	0.870	0.898	MLR model5	0.840*	0.685	0.889
RF model6	0.967	0.929	0.877	MLR model6	0.854*	0.750	0.834
vrRF model1	0.606*	0.467	0.685	vrMLR model1	0.635*	0.610	0.605
vrRF model2	0.602*	0.516	0.632	vrMLR model2	0.622*	0.654	0.542
vrRF model3	0.638*	0.541	0.671	vrMLR model3	0.634*	0.607	0.594
vrRF model4	0.796*	0.594	0.874	vrMLR model4	0.680*	0.517	0.835
vrRF model5	0.895†	0.741	0.919	vrMLR model5	0.801*	0.630	0.950
vrRF model6	0.918	0.821	0.944	vrMLR model6	0.798*	0.643	0.959

*Significantly lower than that in the corresponding RF model: p<0.01.
†Significantly lower than that in the corresponding RF model: p<0.05.
AUC, area under the curve; MLR, Multiple Logistic Regression; RF, Random Forest; ROC, receiver operating characteristic; vrMLR, variable restricted multiple logistic regression (use only nine variables according to a previous study); vrRF, variable restricted random forest (only use single year for prediction).