Abstract 066: Inclusion of Social Determinants of Health in Cardiovascular Disease Risk Prediction Models

Background: Social determinants of health (SDOH) impact CV health, but CVD risk prediction models either exclude SDOHs entirely or include a single factor. Here, we assess the incremental value of adding multiple individual- and neighborhood-level SDOHs to the 2013 Pooled Cohort Equations (PCE). Methods: We pooled data from 38,558 individuals in 6 longitudinal cohorts: Framingham Heart Study (Gen3 and Omni2 cohorts exam 1; 2002-2005), Jackson Heart Study (exam 1; 2000-2004), Coronary Artery Risk Development in Young Adults (year 20 exam; 2005-2006), Multi-Ethnic Study of Atherosclerosis (exam 5; 2010-2011), Reasons for Geographic and Racial Differences in Stroke (exam 1; 2003-2007), and Hispanic Community Health Study/Study of Latinos (exam 1; 2008-2011). Included participants were aged 40-79 with complete data on PCE predictors, were free of baseline atherosclerotic CVD (ASCVD), and were followed for 10 years for incident ASCVD (nonfatal MI, fatal CHD, fatal/nonfatal ischemic stroke). The PCEs were fit separately for each combination of sex and race/ethnicity (Black Women, Black Men, White or Hispanic/Latina Women, White or Hispanic/Latino Men) using Cox proportional hazards models with recalibrated beta coefficients for each CV risk factor (age, total and HDL cholesterol, treated/untreated SBP, current smoking, diabetes). Individual- and neighborhood-level SDOHs were added to the PCEs to assess the change in c-statistic, net reclassification improvement (NRI) and relative integrated discrimination improvement (rIDI). Results: During 340,267 person-years of follow-up, 1966 ASCVD events occurred. In both Black women and White or Hispanic/Latina women, jointly adding individual- and neighborhood-level SDOHs improved model performance the most (Black women Δc=0.034, NRI=0.105, rIDI=0.415; White or Hispanic/Latina women Δc=0.026, NRI=0.136, rIDI=0.329), whereas in White or Hispanic/Latino men, the subset of significant individual-level SDOHs most improved model performance (Δc=0.032, NRI=0.114, rIDI=0.163). Among Black men, no group of SDOHs was both significantly associated with incident ASCVD and improved model performance. Conclusion: Adding individual- and neighborhood-level SDOHs to the PCEs modestly improved model performance and the effect of these groups of SDOHs varied by gender and race. Future studies should examine the impact of inclusion of these and additional individual- and neighborhood-level SDOHs in the PREVENT and other risk prediction models.