Table 2.

Performance of machine-learning based risk prediction models in the test set of the pooled cohort equation cohort.

Predicted 5-year ASCVD risk	Overall		Discrimination		Calibration		Clinical usefulness
	Brier	Brierscaled	C-statistic (95% CI)	P value	Hosmer–Lemeshow χ2	P value	Net benefit at threshold of 3.75%	Net benefit at threshold of 5%
Pooled cohort equation (African)	0.032	35.0%	0.726 (0.716‒0.737)	0.004	506.0	< 0.001	0.0086	0.0049
Pooled cohort equation (white)	0.031	7.3%	0.738 (0.727‒0.749)	–	171.1	< 0.001	0.0106	0.0072
Logistic regression	0.030	4.6%	0.749 (0.738‒0.759)	< 0.001	15.3	0.053	0.0109	0.0079
Random forest	0.031	2.7%	0.720 (0.709‒0.731)	< 0.001	805.8	< 0.001	0.0094	0.0064
TreeBag	0.032	5.9%	0.674 (0.662‒0.685)	< 0.001	403.0	< 0.001	0.0067	0.0038
AdaBoost	0.031	3.9%	0.740 (0.729‒0.751)	0.434	19.9	0.011	0.0107	0.0074
Neural network (16 variables)	0.031	4.4%	0.751 (0.740‒0.761)	< 0.001	86.1	< 0.001	0.0108	0.0078
Neural network (8 variables)	0.031	4.2%	0.748 (0.738‒0.759)	< 0.001	91.2	< 0.001	0.0105	0.0077

ASCVD atherosclerotic cardiovascular disease, CI confidence interval.