A social determinants-based county-level cardiovascular mortality index to identify high-risk counties in the USA.

BACKGROUND: Social determinants of health (SDOH) contribute significantly to geographical variations in cardiovascular disease (CVD) mortality in the USA. We aimed to develop and evaluate a SDOH-related social cardiovascular mortality index (SCMI) as a population-level tool for public health surveillance and planning to identify US counties at risk of high CVD mortality. We also sought to identify the leading SDOH-related variables associated with CVD mortality across US counties. METHODS: We used public registry data from 3141 US counties and applied machine learning techniques to screen 158 SDOH-related variables (centered around 2019), identifying 15 key determinants of total CVD mortality (2018-2020) to develop the SCMI, a composite score ranging from 0.00 to 100.00. We validated SCMI by comparing its predictive performance with social vulnerability index (SVI) for future total CVD mortality (2019-2021) in 1000 bootstrap resamples, based on R2, root mean square error (RMSE), and mean absolute error (MAE). We also used geographical random forest models to identify leading SDOH variables associated with total CVD mortality at the county level. RESULTS: Counties in the highest SCMI quintile (80.01-100.00), reflecting the most disadvantaged community conditions, had on average 202.55 more predicted total CVD deaths per 100,000 population than those in the lowest quintile (0.00-20.00) (p < 0.001). SCMI consistently outperformed SVI in predicting future total CVD mortality, with higher R2 [mean, 0.47 (95% confidence interval, 0.32, 0.57) versus 0.14 (- 0.15, 0.27)] and lower RMSE [72.93 (64.14, 80.69) versus 92.76 (81.14, 105.38)] and MAE [54.63 (48.37, 60.58) versus 72.15 (62.94, 82.94)]. Additionally, the proportion of individuals with frequent mental distress, receiving Supplemental Nutrition Assistance Program, and lower median household income were identified as the leading determinants of total CVD mortality, all positively associated with higher SCMI scores (p < 0.001). CONCLUSIONS: Our novel SDOH-based SCMI consistently outperformed SVI in predicting US counties at risk of high CVD mortality. The identified leading county-level SDOH variables need further evaluation as potentially modifiable targets for community interventions. Our findings can improve county-level cardiovascular risk stratification and support efforts to mitigate geographical variations in cardiovascular mortality in the USA.

Duke Scholars

Author Tazeen Hasan Jafar Duke Global Health Institute

Author Bibhas Chakraborty Biostatistics & Bioinformatics, Division of Biostatistics