Bayesian Network Modeling of Environmental, Social, and Behavioral Determinants of Cardiovascular Disease Risk

Background: Cardiovascular disease (CVD) is the leading global cause of death and is shaped by interacting biological, environmental, lifestyle, and social factors. Traditional models often treat risk factors in isolation and may miss dependencies among exposures and biomarkers. Objective: To map interdependencies among environmental, social, behavioral, and biological predictors of CVD risk using Bayesian network models. Methods: A cross-sectional analysis was conducted using NHANES 2017–2018 data. After complete-case procedures, the analytic sample included 601 adults and 22 variables: outcomes (systolic/diastolic blood pressure, total/LDL/HDL cholesterol, triglycerides) and predictors (BMI, C-reactive protein (CRP), allostatic load, Dietary Inflammatory Index, income, education, age, gender, race, smoking, alcohol, and serum lead, cadmium, mercury, and PFOA). Spearman’s correlations summarized pairwise associations. Bayesian networks were learned with two approaches: Grow–Shrink (constraint-based) and Hill-Climbing (score-based, Bayesian Gaussian equivalent score). Network size metrics included number of nodes, directed edges, average neighborhood size, and Markov blanket size. Results: Correlation screening reproduced expected patterns, including very high systolic–diastolic concordance ( p ≈ 1.00), strong LDL–total cholesterol correlation ( p = 0.90), inverse HDL–triglycerides association, and positive BMI–CRP association. The final Hill-Climbing network contained 22 nodes and 44 directed edges, with an average neighborhood size of ~4 and an average Markov blanket size of ~6.1, indicating multiple indirect dependencies. Across both learning algorithms, BMI, CRP, and allostatic load emerged as central nodes. Environmental toxicants (lead, cadmium, mercury, PFOS, PFOA) showed connections to sociodemographic variables (income, education, race) and to inflammatory and lipid markers, suggesting patterned exposure linked to socioeconomic position. Diet and stress measures were positioned upstream of blood pressure and triglycerides in the score-based model, consistent with stress-inflammation–metabolic pathways. Agreement across algorithms on key hubs (BMI, CRP, allostatic load) supported network robustness for central structures. Conclusions: Bayesian network modeling identified interconnected pathways linking obesity, systemic inflammation, chronic stress, and environmental toxicant burden with cardiovascular risk indicators. Findings are consistent with the view that biological dysregulation is linked with CVD and environmental or social stresses.