Comparing Predictive Power of Area-Level Socioeconomic Status Indices Across Health Outcomes and Geographic Levels

Many researchers want to control for both individual-level demographic/health variables and area-level socioeconomic status (area-SES) when studying health outcomes. However, comparative assessments of area-SES indices across geographic levels and a range of health outcomes are scarce.

Compare predictive power for 3 commonly used area-SES indices: the Graham Social Deprivation Index (SDI), the Area Deprivation Index (ADI), and the CDC Social Vulnerability Index (SVI), for a variety of health outcomes, at different geographic levels (county, 5-digit zip-code, census tract, and census block group). Also compare these indices to the simpler Townsend Deprivation Index (TDI) and population percent in poverty (area-Poverty).

Principal research methods are logistic and ordinary least squares regression.

Medicare fee-for-service beneficiaries, COVID-19 decedents, and drug overdose decedents.

SDI, SVI, ADI, TDI, area-Poverty.

All-cause mortality, diabetes incidence and prevalence, hypertension, renal disease, and 30-day hospital readmission for Medicare beneficiaries; COVID-19 mortality; overdose mortality; Medicare fee-for-service spending.

All measures predict the health outcomes, controlling for age, gender, race/ethnicity, and comorbidities, at zip code, tract, and block-group levels. Predictive power is comparable for SDI, SVI, and a standardized version of ADI, and generally superior to TDI, area-Poverty, and non-standardized ADI. Predictive power is highest at tract level, similar at block-group; reasonably strong at zip code, but weaker at county level.

Across a range of health outcomes, we find similar predictive power for SDI, SVI, and standardized ADI, ideally measured at census tract level. SDI has the value of being more parsimonious, with similar performance. Non-standardized ADI cannot be recommended.