Table 2.
Predictor space stratified by category
| Category | Initial predictor space | Selected by ML4HEN-COX | |
|---|---|---|---|
| Demographics | 12 (0.09%) | 3 (5.9%) | age |
| sex | |||
| Townsend deprivation index at recruitment | |||
| Lifestyle | 11 (0.08%) | 6 (11.8%) | overall health rating—fair |
| smoking status—current | |||
| smoking status—never | |||
| overall health rating—excellent | |||
| weight change compared with 1 year ago—none | |||
| alcohol intake | |||
| Medical history | 7,917 (57.4%) | 5 (9.8%) | hypertension (self-reported) |
| lipid-lowering medication | |||
| diabetes | |||
| hypertension (EHR) | |||
| BP-lowering medication | |||
| Surgical history | 5,740 (41.6%) | 0 | |
| Family history | 32 (0.23%) | 2 (3.9%) | illnesses of father—heart disease |
| illnesses of siblings—heart disease | |||
| Physical exam | 7 (0.05%) | 3 (5.9%) | systolic blood pressure |
| hip circumference | |||
| waist circumference | |||
| Genetics | 5 (0.04%) | 4 (7.8%) | genome-wide polygenic score for CAD (GPSCAD) |
| principal component 3 of genetic ancestry (PC3) | |||
| PC2 | |||
| PC4 | |||
| Laboratory values | 58 (0.42%) | 28 (48.3%) | HDL cholesterol |
| glycated hemoglobin | |||
| LDL cholesterol | |||
| testosterone | |||
| apolipoprotein B | |||
| cystatin C | |||
| lipoprotein(a) | |||
| neutrophil count | |||
| apolipoprotein A | |||
| alkaline phosphatase | |||
| C-reactive protein | |||
| monocyte count | |||
| triglycerides | |||
| red blood cell distribution width | |||
| reticulocyte percentage | |||
| alanine aminotransferase | |||
| basophil count | |||
| total protein | |||
| calcium | |||
| total bilirubin | |||
| mean sphered cell volume | |||
| white blood cell count | |||
| mean corpuscular volume | |||
| monocyte percentage | |||
| hemoglobin concentration | |||
| albumin | |||
| urate | |||
| platelet crit | |||
| 13,782 | 51 | ||
Predictor variables selected by ML4HEN-COX are ranked by leave-one-out C-statistic change within each category (Table S4).