Artificially Intelligent Social Risk Adjustment: Development and Pilot Testing in Ohio

Abstract

Prominent voices have called for a better way to measure, predict, and adjust for social factors in healthcare and population health. Local area characteristics are sometimes framed as a proxy for patient characteristics, but they are often independently associated with health outcomes. We have developed an “artificially intelligent” approach to risk adjustment for local social determinants of health (SDoH) using random forest models to understand life expectancy at the Census tract level. Our Local Social Inequity score draws on more than 150 neighborhood-level variables across 10 SDoH domains. As piloted in Ohio, the score explains 73 percent of the variation in life expectancy by Census tract, with a mean squared error of 4.47 years. Accurate multidimensional, cross-sector, small-area social risk scores could be useful in understanding the impact of healthcare innovations, payment models, and SDoH interventions in communities at higher risk for serious illnesses and diseases; identifying neighborhoods and areas at highest risk of poor outcomes for better targeting of interventions and resources; and accounting for factors outside of providers’ control for more fair and equitable performance/quality measurement and reimbursement.

Imputing Race and Ethnicity

“…41 Area-based measures are often independently associated with health outcomes, and including them improves health equity by better accounting for social context. 42…”

“…Area-level racial/ethnic-related characteristics can involve measures of racial residential segregation and isolation, dissimilarity indices, and historical redlining, reflecting the place-based approach to health 41. Area-based measures are often independently associated with health outcomes, and including them improves health equity by better accounting for social context 42…”