Machine Learning and Social Determinants of Health—An Opportunity to Move Beyond Race for Inpatient Risk Prediction in Patients With Heart Failure

Patients admitted with acute heart failure have higher short- and long-term risk of mortality and morbidity with increasing risk for each subsequent hospitalization.¹ In-hospital risk of mortality varies considerably with consistently lower risk among self-described Black patients, partially resulting in multiple risk scores with race as a covariate that assigns lower risk for Black individuals.² Given that race is a social construct, use of race outside the context of the individual, institutional, and societal levels poses risks.³ Models that use race as a covariate and are not carefully developed may have unintended results, including incorrect conclusions that may persist and impact medical decision making, resource allocation, and policy changes. Moreover, models may oversimplify some racial differences that are seen while missing opportunities to address modifiable risk factors that may contribute to these differences and disparities. Race-agnostic models have been recently proposed to assess biological parameters, including recent efforts in the estimated glomerular filtration calculations from serum creatinine with many health care systems revising these clinical calculators.⁴