Table 2.

Discrimination and calibration performance for prediction of 9-year risk of stroke. Comparisons included the 2017 Framingham Stroke Risk Profile (FSRP), a recalibrated and refitted FSRP, Cox, random survival forest (RSF), logistic regression (LR), support vector machine (SVM), gradient boosted tree (GBT), and multilayer perceptron (MLP) models

	Men		Women
Model Type	Discrimination	Calibration	Discrimination	Calibration
	AUROCs	χ2	AUROCs	χ2
	[95%CI]	[95%CI]	[95%CI]	[95%CI]
FSRP	0.781	5541	0.772	19402
	[0.772-0.790]	[4996-6107]	[0.764-0.780]	[17784-21019]
Recalibrated and refitted	0.824	138	0.825	140
FSRP	[0.816-0.831]	[96-185]	[0.819-0.833]	[97-186]
Cox	0.829	122	0.831	129
	[0.822-0.837]	[83-166]	[0.824-0.838]	[89-172]
RSF	0.826	61	0.832	62
	[0.818-0.834]	[36-90]	[0.824-0.839]	[36-95]
LR	0.831	56	0.832	57
	[0.823-0.838]	[31-86]	[0.825-0.838]	[34-85]
SVM	0.830	712	0.831	24
	[0.823-0.838]	[582-852]	[0.824-0.838]	[11-41]
GBT	0.833	44	0.836	47
	[0.825-0.840]	[24-67]	[0.829-0.843]	[30-69]
MLP	0.831	515	0.833	19
	[0.824-0.839]	[410-627]	[0.826-0.841]	[8-35]