Abstract P201: Social Determinants Of Health Modulate Risk Of Heart Failure Development In A Federally Qualified Health Center Population

Abstract

Introduction: Social determinants of health (SDoH) are known predictors of heart failure, though their incorporation into existing risk models is limited, and these models are not widely implemented. In particular, there are limited studies of heart failure risk within federally qualified health centers (FQHCs) tailored to quantify the effects of several SDoH including sexual and gender minority status, socioeconomic factors, and neighborhood characteristics. Hypothesis: We assessed the hypothesis that risk factors among several SDoH domains would modulate heart failure risk when controlled for traditional biologic risk factors and significantly improve the performance of a predictive model among patients at Howard Brown Health, a FQHC. Methods: We retrospectively identified all patients at Howard Brown Health from January 1, 2012 through December 31, 2019 using electronic health records. The association between risk factors and heart failure diagnosis, defined as having one heart failure-associated ICD code, was assessed by multivariate logistic regression. Missing data were imputed with the R mice package. Neighborhood-level data were obtained from the 2019 American Community Survey. Machine learning was used to predict heart failure with several algorithms including gradient boosting machine via the R caret package. Results: Among 99,047 patients, 130 (0.13%) had heart failure. On multivariate logistic regression, these patients were significantly more likely than peers to be older (OR 0.0002, p<0.001), smoke (OR = 0.0035, p<0.001), have diabetes (OR 0.3, p<0.001), have dyslipidemia (OR 0.3, p<0.001), have hypertension (OR 0.001, p<0.001) or have coronary heart disease (OR 0.8, p<0.001. Non-Hispanic whites were at lower risk than non-Hispanic blacks (OR 0.0008, p<0.01) and at higher risk than Hispanic patients (OR -0.0006, p<0.05). Patients with lower household incomes, government insurance or no insurance, or those who lived in neighborhoods with lower median household incomes were also all more likely to have heart failure (all p<0.05). The machine learning predictive models incorporating SDoH and without SDoH respectively had sensitivities of 87% and 82%, specificities of 85% and 80%, positive predictive values of 99.98% and 99.98%, and negative predictive values of 0.9% and 0.8%. Conclusions: Biologic comorbidities portend the highest heart failure risk. Race and economic factors, including neighborhood level variables, predict the highest heart failure risk among measured SDoH in this FQHC population. A machine learning model is able to predict heart failure with biologic comorbidities, and SDoH improve model performance. In conclusion, future research incorporating SDoH—in particular race and economic characteristics—into risk assessments or predictive modeling may aid early detection and prevention of heart failure and reduce health disparities.