Deep Neural Network algorithm successfully predicts incident cardiovascular disease in non-Hispanic Black adults, indicating significance of social determinants of health
Non-Hispanic Black men experience the highest rate of cardiovascular disease (CVD) across all ages when compared to other ethnic and racial groups. Utilizing data collected via the Jackson Heart Study (JHS), researchers supported by NHLBI and NIMHD evaluated standard/biobehavioral risk factors and social determinants of health (SDOH) as features in a complex deep neural network model for CVD risk prediction. They achieved this by utilizing 3 modeling algorithms for 10-year cardiovascular disease predictions among JHS patients and interpreting insights from these models using Shapely Additive Explanation (SHAP) values to compare against input features, to understand the extent of feature importance; features included all standard/biobehavioral, psychosocial/socioeconomic, and environmental factors.
In their analysis, researchers found that while 7 out of 15 of the top features were standard/biobehavioral CVD risk factors, including sex, HDL, history of smoking, and blood pressure medication status, 8 out of the top 15 factors were SDOH, including insurance status and type, discrimination burden, physical activity resources, accessible grocery stores offering fresh and nutritious food. Researchers found it relevant to highlight that while the deep neural network model held promise for predicting initial CVD incidents, SDOH factors did not significantly improve predictive accuracy for initial CVD events past biobehavioral risk factors; however. SDOH factors did rank higher in importance that other publicly acknowledged standard risk factors such as HbA1C, systolic block pressure, physical activity levels, and LDL readings. Limitations of this study include participant age potentially favoring biobehavioral over other risk factors, and the potential relative importance of features with higher levels of missing data. Researchers discuss how their models highlighted the “trickle-down” impact of SDOH factors on lived experience and biobehavioral responses, ultimately impacting CVD development and progression for groups experiencing disadvantage.
Citation:
Morris MC, Moradi H, Aslani M, Sims M, Schlundt D, Kouros CD, Goodin B, Lim C, Kinney K. Predicting incident cardiovascular disease among African-American adults: A deep learning approach to evaluate social determinants of health in the Jackson heart study. PLoS One. 2023 Nov 10;18(11):e0294050. doi: 10.1371/journal.pone.0294050. PMID: 37948388; PMCID: PMC10637695
Black-White residential segregation and cardiovascular mortality rates
Based on prior research that indicates robust associations between racial segregation and increased risk factors for cardiovascular disease (CVD) among Black Americans, a team of researchers, one of whom was supported by NHLBI, utilized county-level data on CVD deaths among non-Hispanic (NH) Black and NH White adults aged 25 years and above to investigate associations between racial CVD mortality rate disparities and racial residential segregation.
The team used 2014-2017 data from the National Center for Health Statistics (NCHS), as well as county-level covariates from the Robert Wood Johnson Foundation county health rankings for 2017. County-level values of the Black-White Interaction Index were calculated using U.S. Census Bureau population estimates, in order to measure racial residential segregation. Outcome measures included county-level age-adjusted CVD mortality rates for NH Black and NH white populations, and relative risk ratio (RR) for Black-White mortality due to CVD.
The median poverty rate, violent crime rate, and food insecurity rate in highly segregated counties were significantly higher than the same rates in the least-segregated counties. Additionally, researchers found that the most highly segregated counties had higher rates of NH Black mortality due to CVD when compared to the least segregated counties. Black-White disparities in CVD mortality between 2014 and 2017 were more significant in highly segregated counties. While limitations of this study include the lack of causal inference, data limitations, and the lack of accounting for other forms of racial segregation, this study advances sound population-level approaches to quantifying associations between health disparities and county-level residential data via cross-sectional analyses.
Citation:
Reddy KP, Eberly LA, Julien HM, Giri J, Fanaroff AC, Groeneveld PW, Khatana SAM, Nathan AS. Association between racial residential segregation and Black-White disparities in cardiovascular disease mortality. Am Heart J. 2023 Oct;264:143-152. doi: 10.1016/j.ahj.2023.06.010. Epub 2023 Jun 24. PMID: 37364747.
