Identifying Communities at Risk for Poor Health using Multidimensional vs. Unidimensional Neighborhood Disadvantage Indices

Technical Summary: Identifying Communities at Risk for Poor Health using Multidimensional vs. Unidimensional Neighborhood Disadvantage Indices

Problem Statement
Neighborhood disadvantage is a well-established determinant of poor health outcomes, typically measured using composite indices that aggregate multiple economic, social, and housing variables. Widely used indices, such as the Area Deprivation Index (ADI), Social Vulnerability Index (SVI), and Social Deprivation Index (SDI), often incorporate between 7 and 29 variables to create a multidimensional measure of disadvantage. However, the necessity of including such a multitude of variables has not been rigorously questioned. Furthermore, existing indices often conflate the concepts of "disadvantage" and "affluence," potentially obscuring the distinct effects of resource scarcity versus resource abundance on health. This study investigates whether a more parsimonious, unidimensional approach (focusing strictly on poverty/income for disadvantage or high-income/education for affluence) offers greater precision in identifying communities at risk for poor health compared to traditional multidimensional indices.

Methodology
This national, cross-sectional study analyzed 83,522 US census tracts using 2022 data. The researchers compared three specific neighborhood indices derived from the National Neighborhood Data Archive (NaNDA), based on American Community Survey (2018-2022) 5-year estimates:

1. Multidimensional Disadvantage Index: A mean of seven variables covering economic (poverty, public assistance, low income), social (unemployment, education, single-parent households), and housing (non-owner occupancy) domains.
2. Unidimensional Disadvantage Index: A mean of three strictly economic variables (poverty rate, public assistance/food stamps, and low income <$40,000).
3. Unidimensional Affluence Index: A mean of three variables capturing high socioeconomic resources (Bachelor's degree or higher, professional/managerial employment, and high income >$125,000).

Neighborhood health status was obtained from the CDC PLACES database for 2022, providing model-based prevalence estimates for obesity, diabetes, and coronary heart disease. The analysis involved:

• Categorizing all indices into quintiles.
• Constructing contingency tables to examine the correspondence in classification between the indices.
• Calculating disease prevalence within cross-classified cells.
• Fitting generalized linear mixed models (GLMM) with a logit link and beta distribution to regress disease prevalence on index quintiles, adjusting for population density and including a random effect for US state.

Key Results
The study found that the choice of index significantly alters the identification of high-risk and low-risk communities:

• Gradient Differences: While all indices showed a gradient where disease prevalence increased with disadvantage, the magnitude of the difference between the most and least disadvantaged quintiles varied. The unidimensional disadvantage and reverse-scored affluence indices showed larger gaps in disease prevalence (e.g., a 2.2–2.9% difference in coronary heart disease between quintiles) compared to the multidimensional index (1.0% difference).
• Classification Mismatch: The multidimensional index frequently misclassified tracts relative to health outcomes.
  • It classified approximately 18% of tracts as "most disadvantaged" (Q5) that were actually less disadvantaged (Q1–Q4) according to the unidimensional index; these tracts had significantly lower coronary heart disease prevalence (5.7% vs. 7.9%).
  • Conversely, it classified tracts as "least disadvantaged" (Q1) that were actually more disadvantaged (Q2–Q5) by the unidimensional index; these tracts had higher disease prevalence (ranging from 7.4% to 9.1%).
• Regression Findings: Generalized linear mixed models revealed that the multidimensional index tended to underestimate disease burden in the most disadvantaged tracts and overestimate disease burden in the least disadvantaged tracts. In contrast, the unidimensional disadvantage and affluence indices provided a sharper distinction, with the least disadvantaged tracts showing the lowest disease prevalence when measured by the affluence index.

Key Contributions

• Empirical Comparison: The study provides a direct, nationwide comparison of multidimensional versus parsimonious unidimensional indices in predicting neighborhood-level chronic disease prevalence.
• Differentiation of Concepts: It highlights the importance of separating "disadvantage" from "affluence" rather than treating them as opposite ends of a single continuum, demonstrating that a unidimensional affluence index may be more precise for identifying low-risk areas.
• Evidence of "Noise": The findings suggest that adding social and housing variables to economic indicators in composite indices may introduce noise that dilutes the index's ability to discriminate between healthy and unhealthy neighborhoods.

Significance and Claims
The authors conclude that the inclusion of multiple social, educational, housing, and transportation items in area-based disadvantage indices may compromise their performance regarding population health. They claim that a unidimensional affluence or disadvantage index with a more parsimonious set of items offers greater precision in identifying communities at risk for poor health.

The study posits that current public health policies relying on multidimensional indices (such as the Community Deprivation Index used in CMS programs) risk misdirecting resources. Specifically, using a multidimensional index could redirect resources away from over 4,000 census tracts (approx. 16 million people) that are actually high-risk but classified as low-risk by the multidimensional metric, or conversely, direct resources to over 6,000 tracts (approx. 24 million people) that are classified as high-risk but have lower actual disease prevalence. The authors argue for a shift toward simpler, more focused indices to improve the targeting of health care resources.