Rural–urban comparisons of area-based socioeconomic disadvantage and vulnerability indices

Abstract

Background

Policymakers and health systems increasingly rely on area-based socioeconomic disadvantage indices to inform health policy and resource allocation. However, the extent to which indices score the same areas differently across the rural–urban continuum remains poorly understood.

Methods

We compared 5 nationwide US indices (2022 versions) at the census tract level across 4 rural–urban designations using correlations, reliability testing, and decile score distributions. For the most discordant pair, the Neighborhood Atlas Area Deprivation Index (NA-ADI) and the Social Vulnerability Index (SVI), we used ridge regression and dominance analysis to identify specific items driving index disagreement.

Results

Disadvantage indices are not interchangeable. The NA-ADI's unstandardized items demonstrated rural–urban and demographic biases. In metropolitan areas, NA-ADI's overweighted housing cost and income items masked vulnerability among housing cost-burdened, predominantly urban minority populations, classifying these tracts as low deprivation. Conversely, in nonurban areas, NA-ADI inflated deprivation scores in predominantly White tracts, with relatively lower proportions of disadvantage indicators, due to inherently lower housing costs.

Conclusion

The NA-ADI's lack of item standardization risks exacerbating disparities that disadvantage indices aim to alleviate. Transitioning to more methodologically rigorous, transparent indices and developing context-specific rural metrics are urgent imperatives for effective, efficient resource allocation and to address health disparities.

Key Points.

• Socioeconomic disadvantage indices are not interchangeable; index selection affects which communities are prioritized for resources.

• Index construction matters; the Neighborhood Atlas Area Deprivation Index's unstandardized item inputs systematically divert resources away from high-need urban minority communities toward predominantly White rural communities.

• Existing “one-size-fits-all” nationwide indices do not capture unique drivers of rural disadvantage, necessitating context-specific metrics.

Introduction

Increased attention to health disparities has motivated efforts to more effectively allocate limited resources. To identify areas at high risk of health disparities, US policymakers, health systems, and researchers increasingly rely on one of several existing composite indices of small area-level socioeconomic disadvantage. These indices provide a single metric capturing interrelated indicators of income, education, employment, housing, and other domains. Federal, state, and local government agencies and health systems increasingly leverage these indices to inform value-based health care reimbursement models (eg, Centers for Medicare and Medicaid Services [CMS] payment models),^1-8 coordinate public health interventions (eg, vaccine allocation),^9-11 target infrastructure upgrades (eg, lead pipe remediation),^12,13 and facilitate health services research (eg, embedding indices in electronic health record [EHR] software such as Epic).^14,15

A detailed understanding of index use in rural areas is of particular importance to public health, as rural populations have shorter life expectancy, higher median age, and increased rates of chronic health conditions and poverty compared with urban areas.^16-19 Rural communities face disparate health services challenges related to cost, quality, and access,^18,20 and experience varying social and physical environmental exposures.²¹ Despite these differences, frequently used “one-size-fits-all” nationwide indices implicitly assume that disadvantage indicators hold the same meaning across the rural–urban continuum. Recent analyses found substantial index disagreement by rural–urban status but did not compare indices at consistent geographic scales or identify drivers of index disagreement.^22,23

Although disadvantage indices are helpful tools for identifying spatial patterns of social risk, well-informed index selection is essential. Nuanced differences in index purposes, data sources, indicators, and mathematical construction affect which communities are prioritized for policy and resource allocation, meaning that arbitrary index selection can unintentionally exacerbate disparities.²⁴ Prior work comparing indices demonstrated disagreement, particularly within urban areas.^25-27 The extent to which index agreement varies across the rural–urban continuum and which items drive any differences, however, remain unclear. To address these gaps, this study quantified index agreement among 5 nationwide, publicly available indices by rural–urban status and identified items driving disagreement to inform future policy, resource allocation, and research.

Data and methods

Disadvantage index selection

This study compared 5 publicly available nationwide indices, selected based on frequency of use and approval by the Office of the Assistant Secretary for Planning and Evaluation (ASPE) for policy application.⁸ Selected indices included the Neighborhood Atlas Area Deprivation Index (NA-ADI) and 2 standardized alternative versions (Standardized Area Deprivation Index [sADI]; Reproducible Area Deprivation Index [ReADI]), the Social Vulnerability Index (SVI), and the frequently used but not ASPE-reviewed Neighborhood Disadvantage Index (ND). The ND was included instead of the ASPE-approved Social Deprivation Index (SDI) as 2019 is the most recently available version.²⁸ All indices were analyzed at the census-tract level (1200–8000 people), as tracts are more appropriate for measuring small area-level disadvantage than block groups (600–3000 people).²⁹ Tract populations are of sufficient size to capture neighborhood-level social structure and power processes important for health-related resource distribution.^29-31 All indices used 2018–2022 American Community Survey (ACS) 5-year estimates for tracts in all 50 US states; Washington, DC; and Puerto Rico, except where noted. Index items are summarized in Table 1 (see Tables S1 and S2 for detailed index descriptions).

Table 1.

Comparison of 2022 socioeconomic disadvantage index items.

	2022 Index (No. of items)
Domain and itema	NA-ADI (17)	sADI (17)	ReADI (17)	SVI (16)	ND (3)
Income
Below poverty level (%)	•	•	•	−	•
Below 150% of poverty level (%)	•	•	•	•	−
Income disparity (ratio)	•	•	•	−	−
Median household/family income ($)	•b	•b	•b	−	−
Receiving public assistance income (%)	−	−	−	−	•
Annual income <$40 000 (%)	−	−	−	−	•
Housing cost burden (%)	−	−	−	•	−
No health insurance (%)	−	−	−	•	−
Employment (%)
Unemployment	•	•	•	•	−
White collar occupation	•b	•b	•b	−	−
Education (%)
College education	−	−	•b	−	−
High school diploma or higher	•b	•b	−	−	−
Less than or no high school diploma	−	−	•	•	−
Less than 9 years of education	•	•	−	−	−
Housing
Owner-occupied housing (%)	•b, c	•b, c	•b	−	−
Median monthly mortgage ($)	•b, c	•b, c	•b	−	−
Median gross rent ($)	•b	•b	•b	−	—
Median home value ($)	•b, c	•b, c	•b	—	—
Household characteristics (%)
Single-parent households	•	•	•	•	−
Age 65+ y (older adults)	−	−	−	•	−
Age ≤17 y (children)	−	−	−	•	−
Limited English language proficiency	−	−	−	•	−
Persons with a disability	−	−	−	•	−
Households without internet	•	•	•	−	−
Households without a motor vehicle	•c	•c	•	•	−
Housing without complete plumbing	•c	•c	•	−	−
Housing type (%)
Multi-unit (10+) structures	−	−	−	•	−
Crowding (>1 person/room)	•c	•c	•	•	−
Mobile homes	−	−	−	•	−
Persons in group quarters	−	−	−	•	−
Minority status (%)
Minority populations	−	−	−	•	−

Abbreviations: ACS, American Community Survey; NA-ADI, Neighborhood Atlas Area Deprivation Index; ND, Neighborhood Disadvantage Index; ReADI, Reproducible Area Deprivation Index; sADI, Standardized Area Deprivation Index; SVI, Social Vulnerability Index.

Source: 2022 index versions using 2018–2022 ACS 5-year estimates. • = variable in index; − = variable not in index.

^aFor complete index item information, see Table S2.

^bIndicates negative factor loadings.

^cTract factor loadings below |0.65| at the block group level; loadings not available for ReADI.

The NA-ADI provides relative neighborhood disadvantage rankings for research, planning, and policy.³² Originally created by Singh in 2003³³ and later adapted by Kind and colleagues at the University of Wisconsin,³⁴ the NA-ADI captures 4 domains via 17 items (Table 1): income, employment, education, and housing.³⁵ Census-block group percentile scores range from 1 to 100 (highest deprivation). The 2022 NA-ADI was downloaded and aggregated to the tract level using population-weighted means.³⁵ Individual NA-ADI item source data were obtained directly from 2018–2022 ACS data.³⁶ Two alternative NA-ADI versions (sADI, ReADI) addressed a well-known NA-ADI construction issue: unstandardized, overweighted housing cost and income indicators^26,27; the ReADI also addressed NA-ADI's outdated 1990 tract-level factor weights, imputation of missing block group data using tract- and county-level data, and lack of population-size weighting.^26,37

The sADI was created by the University of Michigan's National Neighborhood Data Archive (NaNDA) to address NA-ADI's item standardization issue.³⁸ Index construction and scoring replicate NA-ADI but standardize its 17 items (Table 1) prior to aggregation (see Rollings et al³⁸ for details and replication reliability). The 2018–2022 sADI was downloaded from the Inter-university Consortium for Political and Social Research (ICPSR) open data repository and aggregated to tract level using population-weighted means.³⁸

The ReADI, created by Rehkopf and colleagues at Stanford University, standardizes NA-ADI items (Table 1) and applies updated factor weights, modernized item thresholds (eg, income disparity), advanced imputation methods, and population size and distribution adjustments.^37,39 The 2022 tract-level ReADI scoring follows NA-ADI.³⁹

The SVI was developed by the Centers for Disease Control and Prevention and the Agency for Toxic Substances and Disease Registry to identify communities vulnerable to natural disasters and hazardous events.⁴⁰ Although not originally developed for health policy or research, the SVI is commonly used for both purposes.^11,41,42 The 2022 SVI includes 16 equally weighted items (Table 1) across 4 domains: socioeconomic status, household characteristics, minority population status (unique to SVI), and housing type and transportation. The SVI scores are percentile-ranked from 0 to 1 (highest vulnerability). Domain and overall scores are provided for all 50 US states and Washington, DC. The SVI data were downloaded⁴³ and scaled to 0–100 as needed for study comparisons.

The ND is a theoretically derived measure of concentrated disadvantage developed by NaNDA.⁴⁴ Factor analyses are repeated annually to capture variance across dynamic structural socioeconomic characteristics. The 2022 ND factor analysis resulted in 3 income-related items (Table 1). The ND provides unranked, raw mean proportion scores ranging from 0 to 1 (highest disadvantage). The ND was included considering its updated factor analysis, parsimony, and potential to serve as an alternative to the frequently used NA-ADI and SVI. The ND scores for 2022 were downloaded from ICPSR⁴⁴ and scaled to 0–100 as needed for study comparisons.

Rural–urban status

Tract-level rural–urban status was assigned using the 2020 US Department of Agriculture's Economic Research Service (USDA-ERS) Rural–Urban Commuting Area (RUCA) codes. Tracts are assigned 1 primary (1 to 10) and 1 of 21 secondary (9 metropolitan, 5 micropolitan, 3 small town, and 4 rural) codes representing urbanization, population density, and daily commuting patterns.^45,46 Following the USDA-ERS–recommended 4-level, 30% commuting scheme based on secondary RUCA codes, tracts were categorized as metropolitan, micropolitan, small town, or rural.⁴⁷ This approach provided more spatially granular, tract-level designations than dichotomous and county-level measures while maintaining sufficient tract numbers for meaningful analyses.^48,49

Analysis, part I: index agreement

Good index agreement was defined by tracts scoring in the highest 2 disadvantage deciles according to each pairwise comparison. Poor index agreement was defined by tracts scoring in the highest 2 and lowest 6 disadvantage deciles according to an index pair (eg, high index 1/low index 2, high index 2/low index 1). This definition was based on frequent use of indices to identify high disadvantage areas,²⁵ captured substantial index disagreement, and maintained sufficient tract counts for analyses across rural–urban designations (Methods S1a; Figure S1).

Following these definitions, index agreement was quantified overall and by rural–urban status using tract counts and percentages, pairwise Spearman correlations, Cohen's weighted kappa (an interrater reliability measure complementing association strength captured by correlation coefficients⁵⁰), and decile score distributions. The part I index pair with the highest poor agreement was further analyzed in part II.

Sensitivity analyses compared correlations using decile, ventile, and continuous scores. Correlations and kappa statistics were also compared for 2022 index versions with and without tracts in Puerto Rico, as the SVI lacks those scores. Additionally, to address concerns about data collection and ACS 5-year estimates that spanned the 2020 pandemic,⁵¹ sensitivity analyses were completed using the 2019 NA-ADI, sADI, and ND and the 2018 SVI (only the 2022 ReADI is available). Pairwise Spearman correlations and Cohen's weighted kappa statistics were compared with 2022 results.

Analysis, part II: identifying items driving poor index agreement

To identify index items driving poor agreement between the most discordant pair identified in part I, individual tract-level index items were standardized (normal distribution with a mean of 0 and SD of 1). Duplicate items were removed, and a high-income indicator (≥$125 000/year) was added related to NA-ADI's standardization issue (Methods S1b).

A 2-step ridge regression and dominance analysis process was then applied. This approach allowed us to overcome index item multicollinearity and to calculate both magnitude and direction of item effects on, and their relative importance to, poor index agreement (see Methods S1c). Using ridge regression, odds ratios (ORs) were calculated to determine the magnitude and direction of the top 10 index items driving poor agreement overall and by rural–urban status. Unlike standard logistic models, which become unstable and drop highly correlated variables (eg, median rent, home value), ridge regression applies a penalty that reduces correlated variable coefficients, allowing for objective selection of stable index item predictors.^52,53

While ORs describe an item's effect, holding other variables constant, they do not quantify an item's contribution to index disagreement. Therefore, dominance analysis was used to determine the relative importance (percentage contribution) of each of the 10 items to poor index agreement.^54,55 Dominance analysis calculated the marginal contribution (change in R²) across all 1023 possible combinations of the top 10 items.^56-59 By iteratively testing every combination, this approach partitions the explained variance among correlated items rather than prioritizing one and zeroing out others. Resulting dominance statistics provided a percentage for each index item, totaling 100%, that indicated the relative item contribution to poor index agreement. For outcomes perfectly correlated with at least 1 item, an ordinary least squares (OLS) linear probability model was used. For remaining outcomes, logistic regression models were used.^57,60,61

Analyses were conducted using Stata (version MP18; StataCorp). This study of publicly available, tract-level data did not require University of Michigan Institutional Review Board review.

Results

Analyses identified 84 337, of 85 395 total, tracts classified by at least 2 indices and RUCA codes (82 100 [96.1%] unique tracts included across all comparisons). Table S3 presents detailed tract counts overall and by rural–urban status.

Index agreement

Part I results demonstrated that, except for the standardized alternative NA-ADI versions ReADI and sADI, indices are not interchangeable overall or across the rural–urban continuum (Figure 1; Table S3). The ReADI/sADI (17.4% overall) and ND/sADI (16.2% overall) had the highest index agreement, correlations, and reliability. Although the ND/ReADI had higher overall agreement (16.4%), correlations and reliability were slightly lower than the ND/sADI. The NA-ADI/SVI had the lowest agreement (8.2% overall; Figure 1; Table S3), disagreeing for 13.1% of all tracts where 40.4 million people reside (12.1% metro [32.2 million people] to 19.2% rural [4.0 million people]; Table S4, population details). Across all index comparisons, reliability was highest among metropolitan tracts and lowest among rural tracts.

Figure 1.

(a–c) Pairwise comparisons of tract-level 2022 disadvantage index decile scores by rural–urban status (RUCA codes). In panel c, counts and percentages reflect the sum total of all poor index agreement (see Table S3a for good index agreement). Percentage denominators for each cell are displayed in Table S3b. Poor index agreement = tracts classified by the highest 2 (9–10) and lowest 6 (1–6) deciles by 2 indices, eg: high SVI, low NA-ADI = highest 2 (20%) SVI decile scores, lowest 6 deciles (60%) NA-ADI decile scores. High NA-ADI, low SVI = highest 2 (20%) NA-ADI decile scores, lowest 6 deciles (60%) SVI decile scores. Panel a: higher numbers and darker shading represent higher correlation and greater index agreement. Panel b: higher numbers and darker shading represent higher reliability and greater index agreement. Panel c: poor index agreement: higher numbers and darker shading represent worse (greater poor) index agreement. *Comparison pair contains 2 versions of an Area Deprivation Index (NA-ADI, sADI, ReADI). All indices used 2018–2022 ACS 5-year estimates. Abbreviations: ACS, American Community Survey; Metro, metropolitan; Micro, micropolitan; NA-ADI, Neighborhood Atlas Area Deprivation Index; ND, Neighborhood Disadvantage Index; ReADI, Reproducible Area Deprivation Index; RUCA, Rural–Urban Commuting Area; sADI, Standardized Area Deprivation Index; ST, small town; SVI, Social Vulnerability Index.

Decile score distributions by rural–urban status were noticeably skewed for NA-ADI, sADI, ReADI, and ND, highlighting greater disadvantage among nonurban (micropolitan, small town, and rural) versus metropolitan tracts (Figure 2; details: Table S5). This shift was most pronounced for the NA-ADI: 37.2% of tracts with the highest disadvantage decile score were nonurban versus just 2.9% of tracts with the lowest decile score. The SVI decile score distribution was less skewed, with nonurban tracts comprising 13.2% and 7.2% of the highest and lowest disadvantage decile scores, respectively (Table S5).

Figure 2.

(a–e) Distribution of tract-level 2022 disadvantage index decile scores by rural–urban status (RUCA codes). All index versions used 2018–2022 ACS 5-year estimates. Analyses were completed at the census-tract level. Abbreviations: ACS, American Community Survey; NA-ADI, Neighborhood Atlas Area Deprivation Index; ND, Neighborhood Disadvantage Index; ReADI, Reproducible Area Deprivation Index; RUCA, Rural–Urban Commuting Area; sADI, Standardized Area Deprivation Index; SVI, Social Vulnerability Index.

Sensitivity analyses confirmed that these results were consistent regardless of continuous, decile, or ventile scoring (Figure S2), Puerto Rico tract inclusion or exclusion (Figure S3), and pre- or postpandemic index version use (Figure S4).

Index items driving poor index agreement

Analyses of the index items driving NA-ADI/SVI poor agreement, and visible skew in Figure 2, revealed 2 distinct tract profiles (Figure 3; Tables S4, S6, and S7 index item details and correlations; Tables S8 and S9 and Methods S1 ridge regression; Table S10 dominance analysis).

Figure 3.

(a-d) Relative importance of 2022 index items to poor index agreement by rural–urban status (RUCA codes). The dotted/dashed outlines indicate the combined importance of housing cost (dashes) and income (dots) items. See Table S2 for detailed index item descriptions and Tables S7–S10 for detailed item analyses. “+” indicates that index item means were higher in poor vs good agreement tracts (odds ratio >1), indicating more or less disadvantage depending on the item; “−” indicates that index item means were lower in poor vs good agreement tracts (odds ratio <1), indicating more or less disadvantage depending on the item. Good index agreement = tracts classified by the highest 2 disadvantage deciles (9 and 10) by both indices. Poor index agreement = tracts classified by the highest 2 (9–10) and lowest 6 (1–6) deciles by 2 indices: high SVI, low NA-ADI = highest 2 (20%) SVI decile scores, lowest 6 deciles (60%) NA-ADI decile scores; high NA-ADI, low SVI = highest 2 (20%) NA-ADI decile scores, lowest 6 deciles (60%) SVI decile scores. Abbreviations: HH, household; Med., median; MM, median monthly; NA-ADI, Neighborhood Atlas Area Deprivation Index; occ., occupied; RUCA, Rural–Urban Commuting Area; SVI, Social Vulnerability Index; WC, white collar.

The urban penalty

Tracts classified by high disadvantage according to the SVI but not NA-ADI were largely metropolitan (9.2% metro tracts [Figure 3A]; 26.7 million people [Table S4d]). Dominance analysis revealed that this type of poor index agreement was driven by the NA-ADI's reliance on unstandardized housing costs (monthly mortgage and rent in all tracts, as well as median home value in metro tracts) and median household income items measured in raw dollars (dominance: 69.8% overall, 67.9% metro, 80.8% micro, 72.7% small town, 79.7% rural; Figure 3, left panels; Table S10).

While the NA-ADI interpreted higher dollar values as indicators of low deprivation, a closer review revealed additional tract characteristics consistent with high deprivation. Compared with tracts with good agreement, these high-SVI, low–NA-ADI metropolitan tracts had larger minority populations (78.3% poor vs 75.8% good agreement), higher housing cost burden (42.5% vs 40.1%), more multi-unit structures (30.4% vs 13.6%), double the proportion of household crowding (12.5% vs 5.6%), and limited English proficiency (14.8% vs 7.4%) (Table S4d). These findings suggest that the NA-ADI's unstandardized construction obscured these potential vulnerabilities, penalizing urban minorities for residing in high-cost real estate markets.

Rural disadvantage inflation

Tracts classified by high disadvantage according to the NA-ADI but not the SVI were concentrated in nonurban areas (18.1% of rural [3.7 million people], 15.9% of small town [0.9 million people], and 11.7% of micro [2.4 million people] tracts) (Figure 3B–D; Table S4). Dominance analysis showed that this type of poor agreement was primarily driven by lower proportions of minority populations (dominance: 28.2% overall, 27.7% metro, 20.6% micro, 25.0% small town, 20.8% rural) and other disadvantage indicators (Figure 3, right panels; Table S10). Compared with good agreement tracts, these tracts exhibited less household crowding, greater vehicle access, fewer multi-unit structures, and lower proportions of limited English proficiency.

Other drivers of high NA-ADI, low SVI poor agreement included higher median income (micro, small town), fewer single-parent households (micro, rural), more high school graduates (micro, small town), fewer people without health insurance (micro, rural), lower housing cost burden (small town, rural), more owner-occupied housing (small town, rural), fewer children (micro, small town), and lower median home values (rural). Because rural real estate values are inherently lower, the NA-ADI's unstandardized construction artificially inflated deprivation scores in predominantly White, rural tracts despite relatively lower proportions of other disadvantage indicators.

Discussion

This nationwide analysis showed that index selection affects which communities are classified as highly deprived and prioritized for resource allocation. Indices constructed using standardized items demonstrated strong agreement (sADI/ReADI/ND/SVI), while the unstandardized NA-ADI systematically skewed risk profiles across the rural–urban continuum. Although the NA-ADI's lack of item standardization is well documented,²⁶ its frequent use by government agencies, health systems and Epic's EHR software, and researchers necessitates continued caution.^14,15 When used to inform resource allocation, selecting the NA-ADI over standardized alternatives systematically shifts resources away from cost-burdened, urban minority communities toward predominantly White, rural communities with characteristics consistent with less disadvantage (eg, lower housing cost burden), risking exacerbation of the very health disparities these indices aim to alleviate.

These findings confirmed prior work regarding the NA-ADI's urban misclassification^25-27,62-64 and extended them across the rural–urban continuum. By failing to standardize raw dollar amount items (housing costs, income) with other percentage indicators, the NA-ADI functions more as a proxy for home value than a multidimensional disadvantage measure. This standardization issue yields 2 types of misclassification. First, disadvantage indicators, such as high poverty and crowding, are masked in urban tracts with expensive real estate markets and high median incomes. Second, the NA-ADI artificially inflates disadvantage scores in rural and non-metro tracts characterized by inherently lower housing costs and incomes typical of these areas. These systematic misclassifications contribute to sociodemographic and rural–urban bias: the NA-ADI artificially inflates disadvantage scores among predominantly White, rural communities and masks disadvantage in urban minority communities.^27,62 Recognizing this bias, CMS shifted away from using the NA-ADI to inform payment adjustments in 2025 in favor of the updated and standardized Community Deprivation Index (CDI)^65-67; however, the NA-ADI is still widely used and promoted by other government agencies, health systems, and health services researchers.¹⁵

Findings also addressed a common misconception that SVI's inclusion of minority populations drives NA-ADI/SVI differences.⁶⁸ We performed additional sensitivity analyses that repeated index comparisons and item analyses for the NA-ADI, sADI, SVI, and SVI(-RE), a recreated version of the SVI we calculated without minority populations (Methods S2). These analyses examined the extent to which item standardization and inclusion of minority populations accounted for poor index agreement. Poor NA-ADI/SVI agreement was driven by the NA-ADI's lack of item standardization: 13.1% of tracts had poor NA-ADI/SVI index agreement vs 3.0% for sADI/SVI (Figure 1; Figure S5, Table S11 item analyses).

Standardized sADI/SVI(-RE) poor index agreement was 2.1%, revealing that the SVI's inclusion of minority populations (race and ethnicity) only accounted for poor index agreement among 0.9%, or 786 of 82 100 US tracts (Figure S6). The 2.1% sADI/SVI(-RE) poor index agreement was attributed to index items capturing characteristics inherent to urban–rural areas (eg, income and rent in metro tracts; age, housing cost burden, and vehicle access in non-metro tracts), rather than concentrated disadvantage within each diverse landscape (Figures S6–S10, Table S12; note analyses were underpowered for 6 of 10 models). To avoid misdirecting health resources based on rural–urban status rather than need, these tracts require careful index selection that appropriately defines disadvantage specific to these areas.

Furthermore, these sensitivity analyses and main study findings underscored the limitations of applying nationwide indices across the rural–urban continuum, especially considering lower index reliability results in non-metro tracts.^67,69 This “one-size-fits-all” approach ignores differences and disparities between and within rural and urban areas.^21,23,70-74 Because indices apply static weights to disadvantage indicators, they do not account for varied rural–urban contextual meanings. For example, lack of vehicle access may signal isolation in a rural tract, but simply reflect reliance on robust public transit in a metropolitan tract.⁷⁵ Similarly, unstandardized home values and housing costs have substantially different cost-burden implications depending on local rural–urban real estate markets.⁷⁶

Existing indices also do not account for unique demographic, geographic, and environmental drivers of rural disparities. For example, current indices do not capture occupational health risks inherent to rural economies (eg, concentrations of essential workers, prisons, and meat-packing and manufacturing facilities), which disproportionately affect rural areas and minority populations where these facilities are often concentrated.^77-80 Furthermore, indices fail to capture how broadband and transportation access compound the effects of disadvantage on telehealth and ambulatory care access.^81,82 Measuring rural disadvantage requires attention to unique regional indicators of disadvantage and may benefit from metrics created specifically for rural environments.^22,23

Limitations

Findings should be interpreted within the context of several limitations. First, we excluded several newer and frequently used indices (SDI,²⁸ Child Opportunity Index,⁸³ Balanced Area Deprivation Index (bADI),⁶⁴ and the CDI used by CMS⁶⁷) that lacked 2018–2022 versions or used proprietary source data. Second, tract-level analyses aggregated the block group–level NA-ADI; however, index disagreement is likely greater at this smaller geographic scale according to prior work.^23,26,84 Block group ACS data are also often suppressed due to small populations, leading to 56% of block groups missing at least 1 2022 NA-ADI item and imputed using tract- or county-level data.³⁸

Third, 3 non-metropolitan high-SVI/low–NA-ADI poor index agreement models were underpowered according to an evaluation of tracts per index item (see Methods S1e).⁸⁵ While relative importance percentages for these 3 comparisons should be cautiously interpreted, these models did not inform main study findings concerning the NA-ADI's lack of item standardization; the remaining 7 of 10 item analyses models supporting main findings were amply powered.

Fourth, current study indices relied on a “deficit” approach, whereas identifying and leveraging rural strengths and assets is likely necessary to mitigate rural disadvantage.⁷⁴ Moreover, the analyzed indices capture a “snapshot” of disadvantage and do not reflect multigenerational socioeconomic factors, evolving sociocultural systems, or decades of policy choices and disinvestment in rural communities, health care, and public health institutions that contributed to contemporary rural health disparities and health deserts.^74,86-88 Finally, rural definitions vary by metric and alternative definitions may shift results.^49,89

Implications and recommendations

While disadvantage indices are essential for identifying high-risk areas, relying on flawed metrics threatens effective implementation and efficient resource allocation. The 2 most widely used indices present policymakers, health systems, and researchers with a problematic choice: the NA-ADI lacks item standardization that penalizes urban minorities, while the SVI's inclusion of minority populations may introduce legal and research challenges.^68,90 Careful index selection requires rigorous alignment between index construction and targeted policy or research goals.^25,63 Study findings suggest the following recommendations for index selection.

Government agencies

Following CMS models, other state and local government agencies should transition away from the unstandardized NA-ADI.^65,66 Payment models and policies should continue to use standardized alternatives, such as the CDI,⁶⁷ to prevent penalization of urban minority residents and safety-net providers. Additional standardized alternatives include the sADI or ReADI analyzed in this study, the bADI, or Duke University's ADI.⁹¹ Agencies should also consider using more parsimonious measures that capture the strongest drivers of contemporary disadvantage, such as the ND. Further testing of these alternative standardized metrics and implications must be prioritized.

Health systems

Health systems that have used the NA-ADI to inform local resource allocation, public health interventions, infrastructure upgrades, and EHR linkages, including via Epic, should compare communities or eligibility identified by the NA-ADI and a standardized alternative. This sensitivity analysis process enables practitioners to identify any masked urban vulnerability and adjust interventions and resource allocation accordingly. Indices require face validity or “ground-truth” testing and evaluation of anticipated unintended consequences, such as impacts on eligibility.⁶³

Policy and health services research

When targeting predominantly rural populations, exceptions to nationwide indices may be necessary considering lower index reliability in nonurban areas, in addition to using standardized metrics.²² Researchers and grant funders should prioritize and incentivize development of rural-specific disadvantage (deficit) and opportunity (asset) indices. New metrics are necessary to account for rural deficits and assets, infrastructure, health care access, and within-rural area heterogeneity.^21,23,74

Conclusion

This study demonstrated substantial differences in how tract-level disadvantage classifications vary by index across the rural–urban continuum. High disagreement was primarily driven by the widely used NA-ADI's construction issue. The NA-ADI's lack of standardized indicators resulted in biased misclassifications, masking disadvantage among urban minority populations experiencing high housing cost burden while artificially inflating disadvantage scores among predominantly White, rural communities. As government agencies, health systems, and researchers increasingly rely on disadvantage indices to inform resource allocation, adjust value-based care models, and address health disparities, careful consideration of index selection is critical. To avoid exacerbating the disparities that these indices aim to alleviate, users must critically evaluate index construction, transition away from the NA-ADI, and prioritize using standardized, contextually appropriate disadvantage measures.

Funding

This work received support from the National Institutes of Health/National Institute of Nursing Research (NIH/NINR) U01NR020556 (H.L./K.A.R.), National Institutes of Health/National Institute on Aging (NIH/NIA) K01AG095314 (K.A.R.), and the University of Michigan's Institute for Healthcare Policy and Innovation (H.L.)

Supplementary material

Supplementary material is available at Health Affairs Scholar online.