Incorporating Social Drivers of Health Information Into Health Economics and Outcomes Research: Neighborhood-Level Proxies Versus Individual Level Data
Author(s)
Karl M. Kilgore, PhD1, Christie Teigland, PhD2, Matt McClellan, BS3, Zulkarnain Pulungan, PhD3, Barton Jones, MS3, Barth Kelly, MS3;
1Inovalon, Director, Research Science and Advanced Analytics, Bowie, MD, USA, 2Inovalon, Vice President, Research Science & Advanced Analytics, Bowie, MD, USA, 3Inovalon, Bowie, MD, USA
1Inovalon, Director, Research Science and Advanced Analytics, Bowie, MD, USA, 2Inovalon, Vice President, Research Science & Advanced Analytics, Bowie, MD, USA, 3Inovalon, Bowie, MD, USA
Presentation Documents
OBJECTIVES: Evidence shows that addressing Social Drivers of Health (SDOH) can improve health outcomes and reduce costs but access to data on patients’ SDOH remains a barrier. Healthcare plans/providers commonly use aggregate proxies to assign SDOH. This study examines systematic differences between SDOH at the patient-level vs. neighborhood-level. Recent advances in data-sharing technologies, such as tokenization of patient identified information (PII), has made it feasible to match patient’s actual SDOH to information on healthcare diagnoses, utilization, treatments, and costs.
METHODS: A random sample of 100,000 patients in 2019-2021 was drawn from a nationally representative all-payer claims database and matched to a database containing person-level SDOH for the US population. Patients were matched to SDOH measured at 3 levels: 1) 5-digit-ZIP-level (mean≈1000 households per-neighborhood); 2) 9-digit-ZIP-level (mean≈20 households per-neighborhood); and 3) individual SDOH data using tokenized PII. SDOH characteristics included household size, marital status, home ownership, home type, education, household income, and net worth.
RESULTS: Approximately 81% of patients were matched to their individual SDOH characteristics. R² statistics from regressions of person-level values on neighborhood-level values were used as measures of the accuracy of neighborhood-level values as proxies. 9-digit-ZIP R² ranged from 0.09 (household size) to 0.52 (net worth), indicating these near-neighborhood proxies predicted only 9%-52% of person-level SDOH. 5-digit-ZIP R² were approximately one-half of the smaller neighborhood coefficients.
CONCLUSIONS: The accuracy of aggregate neighborhood characteristics as proxies for individual characteristics varied significantly by SDOH characteristic and size of neighborhood used. At best, <50% of variation across individuals was left unexplained by neighborhood-level measures, and in some cases ≥90% was left unaccounted. Design of effective interventions to address social inequities requires accurate data. Recent enhancements in data availability and matching at the individual patient-level offer healthcare stakeholders improved information to identify and address health disparities.
METHODS: A random sample of 100,000 patients in 2019-2021 was drawn from a nationally representative all-payer claims database and matched to a database containing person-level SDOH for the US population. Patients were matched to SDOH measured at 3 levels: 1) 5-digit-ZIP-level (mean≈1000 households per-neighborhood); 2) 9-digit-ZIP-level (mean≈20 households per-neighborhood); and 3) individual SDOH data using tokenized PII. SDOH characteristics included household size, marital status, home ownership, home type, education, household income, and net worth.
RESULTS: Approximately 81% of patients were matched to their individual SDOH characteristics. R² statistics from regressions of person-level values on neighborhood-level values were used as measures of the accuracy of neighborhood-level values as proxies. 9-digit-ZIP R² ranged from 0.09 (household size) to 0.52 (net worth), indicating these near-neighborhood proxies predicted only 9%-52% of person-level SDOH. 5-digit-ZIP R² were approximately one-half of the smaller neighborhood coefficients.
CONCLUSIONS: The accuracy of aggregate neighborhood characteristics as proxies for individual characteristics varied significantly by SDOH characteristic and size of neighborhood used. At best, <50% of variation across individuals was left unexplained by neighborhood-level measures, and in some cases ≥90% was left unaccounted. Design of effective interventions to address social inequities requires accurate data. Recent enhancements in data availability and matching at the individual patient-level offer healthcare stakeholders improved information to identify and address health disparities.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
SA7
Topic
Study Approaches
Disease
No Additional Disease & Conditions/Specialized Treatment Areas