Area-level social determinants of health and individual-level social risks: Assessing predictive ability and biases in social risk screening

Abstract

Introduction: Area-level social determinants of health (SDoH) and individual-level social risks are different, yet area-level measures are frequently used as proxies for individual-level social risks. This study assessed whether demographic factors were associated with patients being screened for individual-level social risks, the percentage who screened positive for social risks, and the association between SDoH and patient-reported social risks in a nationwide network of community-based health centers. Methods: Electronic health record data from 1,330,201 patients with health center visits in 2021 were analyzed using multilevel logistic regression. Associations between patient characteristics, screening receipt, and screening positive for social risks (e.g., food insecurity, housing instability, transportation insecurity) were assessed. The predictive ability of three commonly used SDoH measures (Area Deprivation Index, Social Deprivation Index, Material Community Deprivation Index) in identifying individual-level social risks was also evaluated. Results: Of 244,155 (18%) patients screened for social risks, 61,414 (25.2%) screened positive. Sex, race/ethnicity, language preference, and payer were associated with both social risk screening and positivity. Significant health system-level variation in both screening and positivity was observed, with an intraclass correlation coefficient of 0.55 for social risk screening and 0.38 for positivity. The three area-level SDoH measures had low accuracy, sensitivity, and area under the curve when used to predict individual social needs. Conclusion: Area-level SDoH measures may provide valuable information about the communities where patients live. However, policymakers, healthcare administrators, and researchers should exercise caution when using area-level adverse SDoH measures to identify individual-level social risks.

Understanding individual health-related social needs in the context of area-level social determinants of health: The case for granularity

“…This analysis adds to the growing body of literature comparing area-level SDoH metrics with individually measured HRSNs [28][29][30][31][32][33]46], cautioning against making assumptions about individuals using aggregate area-level data (also known as the ecological fallacy). Given these findings, there would be real risk in intervening within high-risk census tracts alone, as this would miss many of the individuals living in census tracts considered to have low SDoH risk scores but have self-reported HRSNs.…”

“…However, the SDI metric identified 57% of individuals with no HRSNs living in disadvantaged communities in the Cottrell study, which varies considerably from the 17% in our study sample (corresponding to a specificity of 83%). A recently published follow-up study from the same network of community health centers expanded this analysis to include two additional area-level SDoH indices (the ADI and Material Community Deprivation Index) and quantified the relationships between these metrics and individual-level social risks [ 29 ]. They found that these area-level measures had low sensitivity and would likely miss most individuals with social risks, which is similar to our analysis, with the ADI metric missing ∼ 81% of individuals with HRSNs.…”

“…Particularly given the new regulatory requirements, pragmatic approaches are better aligned with many health-systems efforts and quality measures to implement universal HRSN screening [ 19 , 47 , 48 ]. Additionally, the focus on one urban county in this analysis adds an important juxtaposition to analyses of health centers dispersed across national networks of health centers [ 29 , 31 ]. Finally, utilizing maps as a means of highlighting the heterogeneity of these findings we believe adds an important visual context for potential city and state policy implications.…”

“…Comparisons between individual-level social risks and arealevel SDoH metrics in a variety of settings have shown the limitations of this approach [28][29][30][31][32][33], finding that area-level indices are variable predictors of individual-level social risk. However, past studies have been conducted in geographically dispersed communities and among heterogeneous patient populations, drawing conclusions across a wide range of settings.…”

Variation in multimorbidity by sociodemographics and social drivers of health among patients seen at community-based health centers

“…Additionally, recent work has demonstrated poor alignment of areal-level measures of social drivers of health and individual-level social risks demonstrating the risk of the ecological fallacy and may help explain some of this paradox. 46 …”