Associations of 4 Geographic Social Vulnerability Indices With US COVID-19 Incidence and Mortality

Abstract

Objectives. To examine and compare how 4 indices of population-level social disadvantage—the Social Vulnerability Index (SVI), the Area Deprivation Index (ADI), the COVID-19 Community Vulnerability Index (CCVI), and the Minority Health–Social Vulnerability Index (MH-SVI)—are associated with COVID-19 outcomes.

Methods. Spatial autoregressive models adjusted for population density, urbanicity, and state fixed effects were used to estimate associations of county-level SVI, MH-SVI, CCVI, and ADI values with COVID-19 incidence and mortality.

Results. All 4 disadvantage indices had similar positive associations with COVID-19 incidence. Each index was also significantly associated with COVID-19 mortality, but the ADI had a stronger association than the CCVI, MH-SVI, and SVI.

Conclusions. Despite differences in component measures and weighting, all 4 of the indices we assessed demonstrated associations between greater disadvantage and COVID-19 incidence and mortality.

Public Health Implications. Our findings suggest that each of the 4 disadvantage indices can be used to assist public health leaders in targeting ongoing first-dose and booster or third-dose vaccines as well as new vaccines or other resources to regions most vulnerable to negative COVID-19 outcomes, weighing potential tradeoffs in their political and practical acceptability. (Am J Public Health. 2022;112(11):1584–1588. https://doi.org/10.2105/AJPH.2022.307018)

In a remarkable turn in public health history, the majority of US states followed national guidance on equitable COVID-19 vaccine allocation by adding place-based social disadvantage indices in allocation plans.¹ Social disadvantage indices matter for general public health goals and equity because they prioritize resource allocation to subpopulations with a higher risk of experiencing infections and negative health outcomes from the virus.

Planners used a range of indices for COVID-19 vaccine distribution, increasing vaccine allocations to more socioeconomically disadvantaged areas.¹ However, the indices differ on important dimensions, including the numbers and types of social variables or constructs incorporated and the geographic level.²

The most widely adopted index, used by 28 states, is the Centers for Disease Control and Prevention (CDC) 2011 Social Vulnerability Index (SVI), which comprises 15 variables from the American Community Survey.³ The 2020 COVID-19 Community Vulnerability Index (CCVI), used by 5 states, is based on the SVI and integrates 40 variables, including COVID-19-specific items.⁴ The 2013 Area Deprivation Index (ADI), used by 2 states, is a general policy tool comprising 17 variables.⁵ The 2011 Minority Health–Social Vulnerability Index (MH-SVI), developed by the Office of Minority Health and the CDC, extends the SVI by incorporating 33 American Community Survey variables, including expanded racial/ethnic minority statistics.⁶ The ADI operates at the census block group level (600–3000 people), the SVI’s and CCVI’s lowest resolution is the census tract (1200–8000 people), and the MH-SVI operates at the county level.²

Unlike the SVI, MH-SVI, and CCVI, the ADI does not include race/ethnicity, which can matter for legal and political issues associated with prioritization. For example, targeting underserved populations with use of the SVI has been challenged by policymakers in several states with concerns regarding the legal precedent for using race/ethnicity in health resource allocation and allegations of reverse discrimination.^7,8

In previous work, we found that the SVI was significantly associated with COVID-19 incidence and mortality⁹ and intensity of hospital treatments.¹⁰ However, despite this greater burden, areas with high SVI values are less likely to have robust vaccination rates.^11,12

An important question for public health practitioners and policymakers making vaccine allocation decisions is whether differences in the design of social disadvantage indices affect their association with COVID-19 incidence and mortality. With new COVID-19 variants potentially requiring new vaccines, the initial vaccine supply is not likely to be able to meet demand, and questions regarding optimal prioritization in distribution plans will arise again. In this study, we examined how the SVI compares with the ADI, CCVI, and MH-SVI in predicting COVID-19 outcomes using updated data to inform future policy and resource allocation decisions regarding the use of disadvantage indices.

METHODS

We used data from the CDC (SVI and MH-SVI), Surgo Ventures (CCVI), the University of Wisconsin (ADI), and the New York Times (COVID-19 incidence and mortality data, aggregated from state and local health departments). The sample included 3125 counties or county equivalents in the 50 US states and Washington, DC. Five boroughs of New York City and some smaller counties and boroughs in Alaska were excluded from the analysis because only aggregated COVID-19 data at geographic levels larger than counties and boroughs were reported; 10 other county equivalents with missing data were also excluded. We calculated county-level incidence and mortality rates per 100 000 population by dividing cumulative COVID-19 cases and deaths (as of July 31, 2021) by the total county population and multiplying by 100 000. Rates were log transformed to satisfy normality assumptions for analysis.

Because nationwide COVID-19 data are not available at more granular geographic levels, all disadvantage index data were harmonized at the county level. We transformed ADI data from the census block group level to the county level using a population-weighted average; data for the SVI, MH-SVI, and CCVI were already available at the county level. All indices and subindices are available as percentile rankings, with the SVI, MH-SVI, and CCVI ranging from 0 to 1 and the ADI ranging from 0 to 100. We multiplied SVI, MH-SVI, and CCVI values by a factor of 10 and divided ADI values by a factor of 10 so that the resulting regression coefficients for each index were comparable. Each index or subindex was examined in a separate regression model to avoid multicollinearity.

We used spatial autoregressive models to estimate associations between county-level SVI, MH-SVI, CCVI, and ADI values and COVID-19 incidence and mortality rates with a generalized spatial 2-stage least squares estimator that accounted for spatial autocorrelation and controlled for spillover effects from neighboring counties. An inverse-distance spatial matrix denoting decreasing effects with increasing distance was used for each model, with a spatial lag for the outcome variable and residual errors. Covariates included population density, rural–urban classification, and state fixed effects accounting for differences in pandemic management policies. After analysis, regression coefficients were exponentiated for ease of interpretation to reflect the percentage change in COVID-19 outcomes for a disadvantage index increase of 10 percentile ranks (i.e., 1 decile).

We conducted sensitivity analyses to ensure the robustness of our findings, including New York City as an aggregate county in the analysis, limiting the time period to before vaccine rollout in December 2020, adjusting for community mobility, and assessing temporal trends with serial cross-sectional analyses. All of these analyses produced findings similar to those of the main analyses.

RESULTS

All 4 disadvantage indices had similar positive associations with COVID-19 incidence and mortality. For every index decile increase, the incidence rate increased by 4% for the CCVI, 3% for the ADI, 3% for the SVI, and 3% for the MH-SVI (Table 1). The ADI had a stronger association with COVID-19 mortality than the other indices, increasing by 20% for each decile increase in the index, compared with 9% for the CCVI, 7% for the SVI, and 6% for the MH-SVI. Each SVI, MH-SVI, and CCVI subindex was significantly associated with COVID-19 incidence, and most were significantly associated with mortality (Table 1).

TABLE 1—

Associations Between Disadvantage Indices and US COVID-19 Incidence and Mortality as of July 31, 2021

Disadvantage Index	Incidence, b (95% CI)	Mortality, b (95% CI)
ADIa	1.03 (1.03, 1.04)	1.20 (1.17, 1.22)
SVIb	1.03 (1.03, 1.03)	1.07 (1.06, 1.08)
Socioeconomic status subindex	1.02 (1.02, 1.03)	1.08 (1.06, 1.09)
Household characteristics and disability subindex	1.01 (1.00, 1.01)	1.06 (1.05, 1.07)
Minority status and language subindex	1.03 (1.03, 1.04)	1.02 (1.01, 1.03)
Housing type and transportation subindex	1.02 (1.02, 1.02)	1.03 (1.02, 1.04)
MH-SVIc	1.03 (1.03, 1.03)	1.06 (1.04, 1.07)
Socioeconomic status subindex	1.02 (1.02, 1.03)	1.08 (1.06, 1.09)
Household characteristics and disability subindex	1.01 (1.00, 1.01)	1.06 (1.05, 1.07)
Minority status and language subindex	1.02 (1.01, 1.02)	0.99 (0.98, 1.00)
Housing type and transportation subindex	1.02 (1.02, 1.02)	1.03 (1.02, 1.04)
Health care infrastructure subindex	1.00 (0.99, 1.00)	0.99 (0.98, 1.00)
Medical vulnerability subindex	1.02 (1.02, 1.03)	1.08 (1.07, 1.09)
CCVId,e	1.04 (1.03, 1.04)	1.09 (1.08, 1.11)
Socioeconomic status subindex	1.02 (1.01, 1.02)	1.06 (1.05, 1.07)
Minority status and language subindex	1.03 (1.03, 1.04)	1.02 (1.01, 1.03)
Housing type, transportation, household composition, and disability subindex	1.03 (1.02, 1.03)	1.05 (1.04, 1.06)
Epidemiological factors subindex	0.98 (0.97, 0.98)	1.05 (1.04, 1.07)
Healthcare system factors subindex	1.03 (1.02, 1.03)	0.99 (0.98, 1.01)
High risk environments subindex	1.02 (1.01, 1.02)	1.06 (1.05, 1.07)
Population density subindexe	1.02 (1.01, 1.03)	0.99 (0.98, 1.01)

Note. ADI = Area Deprivation Index; CCVI = COVID-19 Community Vulnerability Index; CI = confidence interval; MH-SVI = Minority Health–Social Vulnerability Index; SVI = Social Vulnerability Index. The regression coefficient was exponentiated from log-transformed data representing the percentage change in COVID-19 outcomes for a disadvantage index increase of 10 percentile ranks.

^a ADI includes 17 census/American Community Survey (ACS) measures. National rankings for US census block groups are provided as a percentile ranging from 1 to 100. In this analysis, ADI rankings were divided by 10 to aid in comparisons with the SVI, MH-SVI, and CCVI.

^b SVI includes 4 subindices composed of 15 ACS measures. The overall SVI and each subindex are percentile ranks ranging from 0 to 1, with higher values indicating greater social vulnerability/disadvantage. Each index was multiplied by 10 to aid in comparisons with the ADI, MH-SVI, and CCVI.

^c MH-SVI is an extension of the SVI and incorporates 4 indices included in the SVI and 2 additional indices composed of 33 ACS measures. Similar to the SVI, the overall MH-SVI and each subindex are percentile ranks ranging from 0 to 1, with higher values indicating greater social vulnerability/disadvantage. Each index was multiplied by 10 to aid in comparisons with the ADI, SVI, and CCVI.

^d CCVI includes 7 subindices composed of 40 measures derived from the ACS; the Behavioral Risk Factor Surveillance System; the National Cancer Institute; the National Center for HIV, STD and TB Prevention; the Centers for Medicare & Medicaid Services; the Bureau of Labor Statistics; and other government and nonprofit organizations. The overall CCVI and each subindex are percentile ranks ranging from 0 to 1, with higher values indicating greater social vulnerability/disadvantage. Each index was multiplied by 10 to aid in comparisons with the ADI, MH-SVI, and SVI.

^e Models incorporating the overall CCVI and the CCVI population density subindex did not include additional covariates to avoid multicollinearity.

DISCUSSION

Despite differences in component measures, all 4 indices we assessed demonstrated an association between greater disadvantage and COVID-19 incidence of a similar magnitude. Although all of the indices were associated with COVID-19 mortality, there was more variation in the magnitudes of the relationships, suggesting that the different social variables used to construct each index may mediate or moderate other pathways relevant to illness severity and death.

Although the CCVI was developed to tailor the SVI to COVID-19 by incorporating additional variables, neither the CCVI nor the MH-SVI produced stronger mortality or incidence associations. The mortality association was weakest for the MH-SVI and strongest for the ADI, the index that offers the most fine-grained geographic resolution and involves the lowest risk of legal challenges given its exclusion of race variables. Yet, as a CDC-issued index, the SVI commands significant authority among public health practitioners nationally.

Policymakers should consider data availability and practical application when selecting an index for use in ensuring equitable COVID-19 testing, treatment, and vaccine resources. For federal and state policymakers who frequently have data available at the census tract or county level, the SVI, MH-SVI, and CCVI may all be similarly effective. However, there may be political tradeoffs in areas where debates about the use of race in resource allocation limit use of these indices. For county health department or health system planners who may have neighborhood-level data, the ADI can be used to target concentrated areas of social disadvantage for resource allocation. Targeting these smaller geographic units may be advantageous for ensuring equitable testing or other local resources within counties, although there may be tradeoffs in accuracy because margins of errors are greater in smaller geographic areas. Thus, each index might be applied at different geographic levels and in different policy contexts to promote equitable COVID-19 testing, treatment, and vaccine resources.

A potential limitation of this study is that our county-level analyses did not incorporate individual-level patient risk factors such as medical comorbidities, nor did they focus on more granular neighborhood-level effects. These county-level analyses were conducted to harmonize the geographic level of data across indices for comparison purposes, but they may not have accounted for heterogenous areas of disadvantage within counties. In addition, our use of rates as outcome variables disregarded differences in county population size, which may have biased or reduced the efficiency of our estimates. This is a limitation of spatial autoregression methods because programs in commonly used software packages do not currently allow analytic weights to be included in model estimations.

PUBLIC HEALTH IMPLICATIONS

Overall, our findings suggest that despite differences in design, each of the social disadvantage indices assessed in this study can be used in different ways to assist public health leaders’ efforts to promote efficient and equitable vaccine allocation, weighing potential tradeoffs in their political and practical acceptability.