Protective predictors of cardiovascular disease: an explainable AI approach.

OBJECTIVES

To develop interpretable machine learning (ML) models using nationally representative survey data to identify protective factors against cardiovascular disease (CVD), addressing gaps in traditional clinical risk scores across diverse populations.

STUDY

DESIGN

Cross-sectional analysis of the 2021 BRFSS.

METHODS

We analyzed 116,608 adult records after data cleaning. Three ML models (XGBoost, convolutional neural network, random forest) were trained on 11 demographic and behavioral features (age, sex, race/ethnicity, income, smoking, alcohol use, depression, diabetes, insurance status, and fruit and vegetable intake). Performance was assessed using precision, recall, F1-score, AUROC, and AUPRC. SHapley Additive exPlanations (SHAP) were used for interpretability.

RESULTS

XGBoost outperformed other models, achieving precision 0.90, recall 0.82, F1-score 0.86, AUROC 0.76, and AUPRC 0.95. SHAP indicated younger age, higher income, insurance coverage, and absence of diabetes or depression as strong protective predictors.

CONCLUSIONS

An explainable XGBoost model predicts cardiovascular resilience by emphasizing absence of diabetes, mental health stability, socioeconomic advantage, and younger age, supporting proactive and equitable prevention and more efficient resource allocation in CVD care.