Composite socio-environmental risk score for cardiovascular assessment: An explainable machine learning approach

Abstract

Background

Cardiovascular disease (CVD) is the leading global cause of death, with socio-environmental factors significantly influencing morbidity and mortality. Understanding these factors is essential for improving risk assessments and interventions.

Objective

To develop and evaluate the predictive power of a composite socio-environmental (SE) cardiovascular risk score in forecasting major adverse cardiovascular events (MACE) among patients, considering both traditional and novel socio-environmental risk factors.

Methods

A Survival Random Forest (RSF) model was used to create a composite socio-environmental (SE) cardiovascular risk score using 22 census-tract level variables from 62,438 patients in the CLARIFY registry undergoing coronary artery calcium (CAC) scoring. A Cox Proportional Hazard (CPH) model was then applied to assess the association between the SE-MACE risk score and MACE in a hold-out test set. SHapley Additive exPlanations (SHAP) values were used to identify variable importance.

Results

The study included 62,438 individuals (mean age 59.6 years, 53.2 % female, 87.7 % White). Hypertension (55.4 %), diabetes (15.7 %), and dyslipidemia (72.3 %) were common, with a median CAC score of 168. The RSF model showed a concordance index of 0.58, with significant factors including smoking prevalence, insurance status, and median household income impacting cardiovascular risk. The SE-MACE risk score was robustly associated with MACE (HR, 1.21 [95 % CI, 1.11-1.32]), independent of clinical variables and the CAC score. Kaplan-Meier analysis highlighted clear risk stratification across SE-MACE score quartiles.

Conclusion

The SE-MACE risk score effectively incorporates socio-environmental factors into cardiovascular risk assessment, identifying individuals at higher risk for MACE and supporting the need for holistic assessment frameworks. Further validation in diverse settings is recommended to confirm these findings.