Abstract
<title>Abstract</title> <p> <bold>Background.</bold> Young Black and Hispanic adults in the United States bear a disproportionate burden of cardiovascular stroke-risk factors. Using the National Health and Nutrition Examination Survey (NHANES), we quantified racial and ethnic disparities in multifactorial risk-factor control among adults aged 18–55 years and evaluated whether machine learning models built from non-clinical characteristics could identify individuals with poor control. <bold>Methods.</bold> Cross-sectional analysis of 10 NHANES cycles (1999–2000 through 2017–2018), adults aged 18–55 years (N = 32,567 main; N = 31,056 laboratory). Two composite outcomes: poor control (main) = elevated blood pressure (systolic ≥ 130 mm Hg or diastolic ≥ 80 mm Hg) or current smoking (≥ 1 of 2); poor control (laboratory) = ≥ 2 of 4 components adding HbA1c ≥ 6.5% and LDL-C ≥ 130 mg/dL. Model A used demographic, socioeconomic, and healthcare-access features. Model B added HDL-C, total cholesterol, and triglycerides (non-outcome lipid biomarkers). Both models excluded outcome-defining variables. Four algorithms were trained with survey weights and evaluated on a stratified held-out test set. <bold>Results.</bold> Survey-weighted prevalence of poor control (main) was 45.8% overall and 52.0% among Non-Hispanic Black adults (prevalence ratio 1.10 versus Non-Hispanic White; absolute difference +4.7 percentage points). XGBoost achieved survey-weighted test AUROC = 0.716 (main) and 0.822 (laboratory). Disparities persisted within insurance strata. <bold>Conclusions.</bold> Demographic, socioeconomic, and healthcare-access characteristics were associated with poor cardiovascular risk-factor control and provided moderate discrimination in the main cohort. These associations do not establish causal effects, but they may help identify groups that could benefit from targeted preventive-care outreach. </p>