OBJECTIVES: To compare proxies for individual level Social Determinants of Health (SDoH) drawn from two different neighborhood sizes—Census Block Group (CBG) vs. 9-digit ZIP Code (ZIP9)—in modeling healthcare outcomes. The use of aggregate proxy data where SDoH characteristics of residential areas are imputed to the individual and used to risk adjust health outcomes is common in research on social risk factors. This study compares two models relating individual patient outcomes to a set of similarly defined SDoH proxy variables calculated from two different levels of aggregation: ZIP9 vs. CBG.

METHODS: The study sample included 1,813,937 Medicare Advantage (MA) beneficiaries, continuously enrolled in 2015, extracted from a national claims database. Based on address, beneficiaries were matched to household SDoH variables from: 1) a commercial market research database of the US population, aggregated at the ZIP9 level (>30M areas), and 2) the American Community Survey, aggregated at the CBG level (approximately 220K areas). Common SDoH variables in the two databases were race, education, marital status, home ownership, and income. These variables, recoded to have consistent response groups, were used as predictor variables in two generalized linear regression models. Response variables (outcomes) at the individual level were based on 3 HEDIS/PQA quality measures: Breast Cancer Screening (BCS), Plan All-Cause Readmissions (PCR) and Medication Adherence for Diabetes (MA-D).

RESULTS: Income and home ownership were significantly related to BCS in both models, but the strength of the relationship (measured by standardized parameter estimates) was approximately 1.4 times greater for ZIP9 than for CBG level data. Income was related to PCR, but only using ZIP9 data; income was not significant using CBG level proxies. No consistent trends emerged for MA-D.

CONCLUSIONS: Using data drawn from smaller neighborhood areas may serve as better proxies to capture the effect of social risk factors on patient outcomes.