Table 1.
Development (N = 138,619) | Holdout (N = 34,655) | |
---|---|---|
Age (years) | 56.2 (8.1) | 56.1 (8.1) |
Males | 70,896 (51.1%) | 17,606 (50.9%) |
Ethnicity | ||
White | 132,610 (95.7%) | 33,092 (95.5%) |
Black | 1,945 (1.4%) | 499 (1.4%) |
East Asian | 1,095 (0.8%) | 290 (0.8%) |
South Asian | 1,614 (1.2%) | 402 (1.2%) |
Other | 1,355 (1.0%) | 372 (1.1%) |
Current smoker | 14,501 (10.5%) | 3,604 (10.4%) |
Diabetes | 6,568 (4.7%) | 1,635 (4.7%) |
Cholesterol (mg/dL) | 217.5 (37.8) | 217.4 (37.6) |
HDL-C (mg/dL) | 55.4 (13.9) | 55.3 (13.9) |
LDL-C (mg/dL) | 136.3 (29.2) | 136.2 (29.0) |
SBP (mm Hg) | 137.5 (18.4) | 137.3 (18.3) |
Antihypertensive | 26,100 (18.8%) | 6,501 (18.8%) |
Genome-wide polygenic score for CAD(GPSCAD) | −0.03 (0.99) | −0.03 (0.99) |
Incident CAD events over median 11-year follow-up | 4,103 (3.0%) | 1,037 (3.0%) |
Predicted 10-year risk (%) | ||
FRS | 6.9 (6.4) | 6.9 (6.4) |
PCE | 8.3 (7.7) | 8.2 (7.7) |
QRISK3-2017 (QRISK3) | 10.0 (8.4) | 9.9 (8.4) |
The development cohort was used for a 5-fold cross-validation procedure to build ML4HEN-COX, while the holdout cohort was used to test performance in unseen data (Figure 1). GPSCAD was adjusted for the first four PCs of genetic ancestry and scaled to mean 0 and standard deviation 1. None of the above variables were significantly different between groups at the p < 0.05 level. HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; SBP, systolic blood pressure.