Skip to content

Abstract Search

Primary Submission Category: Chronic disease

Developing a Predictive Model for Cardiovascular Disease with Only Social and Behavioral Risk Factors

Authors:  Stephanie Kjelstrom, Richard Hass, Brandon George, Sharon Larson,

Presenting Author: Stephanie Kjelstrom*

Background: Cardiovascular disease (CVD) is the leading cause of mortality in the US. Current CVD risk scores that physicians use contain behavioral and clinical risk factors (smoking status, blood pressure, lipids, diabetes etc.). However, many studies have identified CVD social, environmental, and behavioral risk factors (SEBRFs) which are precursors to clinical factors and often missing from patient charts. Creating a predictive model with only SEBRFs may offer earlier risk stratification than traditional CVD risk scores, and could be administered in both community and healthcare settings.

Objective: To develop a predictive model for CVD using solely SEBRFs.

Methods: We analyzed the 2021 Medical Expenditure Panel Survey, utilizing the social determinants of health questionnaire data (18,435 participants). Multivariable logistic regression regularization methods LASSO, elastic net, and forward selection with BIC were employed for model development. Fifty-two variables were included in the models. Performance was assessed with an area under the curve (AUC), observed vs expected (O:E) ratio, and k-fold cross-validation (CV) with 10 folds.

Results: LASSO with cross-validation achieved the highest AUC via k-fold CV (82.7% [81.3, 83.2]) and excellent O:E (1.003), closely followed by elastic net (82.3% [81.6, 83.4]) and forward selection with BIC (82.5% [81.3, 83.1]). Significant predictors included age, sex, Medicaid status, mental health diagnoses, current stress, adverse childhood events, smoking, exercise, transportation issues, mold exposure, income, household size, debt collection contact, education, discrimination, and marital status.

Conclusion: SEBRFs alone can yield highly predictive CVD models. A risk score based on these upstream factors could enable earlier patient risk stratification before clinical manifestations, facilitating preventative interventions.