Optimizing cardiovascular disease mortality prediction: a super learner approach in the tehran lipid and glucose study

Table 4 Four clinical models: the strategy of selecting the best scale of variables

Method	Clinical Model	Integrated Brier Score	Prediction Error	C-index	AUC (CI)
Cox-PH	1	0.011	0.011	90.03	93.83 (93.12–94.54)
	2	0.011	0.012	90.21	93.42 (92.71–94.13)
	3	0.011	0.011	91.47	93.48 (92.77–94.19)
	4	0.011	0.012	88.06	91.97 (91.26–92.68)
GBM	1	0.013	0.013	89.79	90.52 (89.81–91.23)
	2	0.013	0.013	90.14	91.14 (90.43–91.85)
	3	0.013	0.013	90.18	91.67 (90.96–92.38)
	4	0.013	0.013	86.52	87.25 (86.54–87.96)
SVM	1	0.014	0.013	75.99	85.25 (84.54–85.96)
	2	0.014	0.013	89.10	90.01 (89.30–90.72)
	3	0.014	0.013	89.63	90.13 (89.42–90.84)
	4	0.014	0.013	88.95	89.95 (89.24–90.66)
SL	1	0.011	0.011	91.60	94.34 (93.63–95.05)
	2	0.011	0.011	90.55	93.59 (92.88–94.30)
	3	0.011	0.011	92.81	93.73 (93.02–94.44)
	4	0.012	0.011	87.97	91.86 (91.15–92.57)

Model I included Age, Sex, Smoking status, Education, Marital Status, Family History of Stroke, SBP, DBP, BMI, Waist, Hip, FBS, TG, HDL, Physical Activity, Lipid Drug, Anti-Hypertension Drug, Aspirin, Corticosteroid; Model III included Age, Sex, Smoking, Education, Marital Status, Family History of Stroke, Anti-Hypertension Drug, BMI categorization, Waist-to-Hip Ratio, T2DM, high TG, low HDL, and Physical Activity. GBM = Gradient boosting model; SVM = support vector model; SL = super learner; CI confidence interval

ISSN: 1472-6947