• Definition
• Data Sources(s)
• Groups with Substantial Contributions
• Release Cycle
• Release Date
• Geography
• Background
  • Social Determinants of Health & Deprivation Indexes
    • The Importance of Social Determinants of Health
    • The Importance of Deprivation Indexes
    • Selecting an Deprivation Index
  • The Area Deprivation Index
    • Estimating ADI Scores
• Methods
  • Data Processing & Analysis
    • ADI by Census Block Group
      • CBG ADI score
      • ADI percentile rank and ADI quintile categories
    • ADI for Our Community
      • Total ADI score
      • ADI percentile rank distribution in Our Community (overall and by race/ethnicity)
    • 17 Indicators at the ZCTA- and County-level
    • 17 Indicators for Our Community
• Limitations, Important Notes & Discussion
• References

Area Deprivation Index

Definition

A composite score of area-level socioeconomic deprivation. Smaller values indicate an area is “less deprived” with base mean of 100 and a standard deviation of 20.

Data Sources(s)

U.S. Census Bureau, American Community Survey, 5-Year Data

Groups with Substantial Contributions

Singh, G. K. (2003)¹

Release Cycle

Annually

Release Date

January

Geography

Census Block Group

Background

Social Determinants of Health & Deprivation Indexes

The Importance of Social Determinants of Health

In 2015, U.S. healthcare spending reached \$3.2 trillion amounting to \$9,990 per person and 17.8% of its gross domestic product (GDP). While this was almost 50% more than any country, the U.S. only ranked 31st in life expectancy.^2–4 It is relatively well-known that the U.S. is an outlier when compared to other countries because it spends more on healthcare and gets less in return in the form of better health outcomes, such as life expectancy and the prevalence of chronic disease.^5,6 The U.S. also performs poorly on measures of population health when compared to 13 other high-income countries (called OECD countries), with the U.S. having the lowest life expectancy, the highest infant mortality rate, and higher prevalence of chronic disease. Further, the U.S. has fewer practicing physicians, fewer physician visits, fewer hospital beds and fewer discharges per capita than the median of 13 high-income countries.⁴

Despite high spending on healthcare in the U.S., healthcare accounts for only 10% of the variability in health outcomes,³ while social determinants of health (SDOH) – the sphere (e.g., social, economic and physical) of a person’s economic stability, education, social and community context, health and healthcare, and neighborhood and environmental factors – accounts for 20% of the health outcomes, with individual behavior accounting for 40%, and genes 30%.^7,8 Despite SDOH accounting for 20% of the variability in health outcomes, the U.S. spends only 9% of its economy on social services, which is significantly less than other OECD nations. In fact, the U.S. was the only country that spent more on healthcare than social services.⁴ Thus, the idea that individual-level attributes play the only role in health determinants (i.e., determining individual health) and ignoring or downplaying neighborhood and area effects seems misguided.⁹

The impact of material deprivation (e.g., grossly inadequate food, clothing, shelter, water, sanitation, etc.) and simply just living in a socioeconomically disadvantaged neighborhood (one factor in SDOH) has been known for centuries,^10–14 and has been linked to many adverse health outcomes, including: unideal birthweight,¹⁵ childhood obesity,¹⁶ sexually transmitted infections¹⁷, life expectancy (disability-adjusted life years),¹⁸ hospitalizations¹⁹, readmissions/rehospitalization^20–24 and mortality.²⁵ It is important to note that in talking about SDOH, they include the most downstream nonmedical factors in the causal chain/pathway influencing health, such as health-related knowledge, attitudes, beliefs, or behaviors (e.g., smoking). That being said, it is the upstream SDOH that have a more direct causal impact on improving health and reducing health disparities. These upstream factors can be conceptualized as “living and working conditions in homes and communities” (e.g., the quality of the house one lives in) and by even more upstream indicators of “economic and social opportunities and resources” (e.g., civic participation).²⁶

This means that our ZIP code alone – where we live and where our children are born – can have a significant impact on our life chances beyond the amount of dollars spent on healthcare. It is these social factors – parks, playgrounds, neighborhoods, education, income, social integration, health coverage and safety that are as much about who we are and how healthy we are, as our genes – that have a sizable impact in determining how healthy we are and how we feel. SDOH encompass social (e.g., job opportunities), structural/physical/environmental (e.g., waterways) determinants and the conditions of these determinants (e.g., how many job opportunities, quality of waterways) in which we are born, grow, live, work and age.^8,27 General examples of constructs that make up SDOH include socioeconomic status (SES), education, the physical environment, employment, and social support networks, as well as access to health care.^7,8 Thus, social deprivation is made of and/or correlates with other commonly known constructs, such as SES and social exclusion,²⁸ but should not be confused with them. When considering the deprivation of an area, it is important to choose socioeconomic indicators that approximate material and social conditions, along with the relative socioeconomic disadvantage in a given community.^1,7,29–31

The Importance of Deprivation Indexes

Since the release of the 2008 World Health Organization report, Closing the Gap in a Generation: Health Equity through Action on the Social Determinants of Health, there has been increased attention in the U.S. on SDOH and a recognition that health and healthcare access varies along a social stepwise incremental gradient (i.e., health improves with increasing social advantage and affects everyone along this gradient)^26,27,32. There is wide international consensus and substantial research that targeted resource allocation can reduce a range of disparities, and that SDOH can help guide this.³³ And while there has been intense interest in the U.S. to leverage SDOH to direct clinical and community health interventions, and to adjust quality measures and payments, the U.S. has not yet leveraged social deprivation data in the same way other countries have.³² For example, the U.S. has not taken advantage of their rigorously researched and validated censuses or administrative data sets to aggregated SDOH into indexes in the ways the United Kingdom and New Zealand have by creating deprivation indexes representing aspects of material and social deprivation. The U.S. “desperately needs one" to "measure socioeconomic variation across communities, assess community needs, inform research, adjust clinical funding, allocate community resources, and determine policy impact.³² Indices provide these countries with comparable data and serve as a universal language and tool set to define organizing principles for population health” and help meet the Triple Aim of lower of lower costs, improved care, and population health.^32,34

To immediately integrate social determinant based indexes into the causal pathways of improving health outcomes, SDOH need to be leveraged at the ecological level (e.g., ZIP code level as ecological correlation) so data can be acquired efficiently and quickly through open and publicly available sources to reduce marginal costs for healthcare (e.g., the average cost of a single hospital visit).³² Though data is acquired at the individual level by such organizations as the U.S. Census Bureau, it is ecological in nature because it is aggregated beyond the individual level, thus ecological fallacy is always a concern. That being said, the ecological deprivation indexes are still valuable in predicting outcomes and for use in estimating disease prevalence using small area estimation to provide important statistics at different ecological levels.

There has also been an effort to bring SDOH indicators and their related indexes into patient electronic health records (EHRs) to “better inform clinical recommendations for individual patients, facilitate referrals to community services, and expand understanding of factors impacting treatment adherence and health outcomes. This information could also help care teams target disease prevention initiatives and other health improvement efforts for clinic panels and populations.”³⁵

Selecting an Deprivation Index

A multitude of different indexes for measuring socioeconomic disadvantage have been developed.³² The Community Needs Index (CNI) developed in 2004 by Catholic Healthcare West (i.e. Dignity Health) in partnership with Solucient (now Truven Health Analytics) is commonly used by hospitals.³⁶ The CNI methods have been published in the public domain and include measures of poverty and race/ethnicity that have individually been associated with health. Unfortunately, the CNI has several weaknesses, including: 1) Relying on paid and proprietary data sources; 2) Lacking rigorous statistical based validation methods; and 3) Not having been extensively studied or validated in the peer-reviewed academic literature in how well it predicts health outcomes.^36,37 Another index is the SocioNeeds Index®³⁸ by Conduent (formerly The Healthy Communities Institute), which suffers from the same shortcomings as the CNI. The Social Deprivation Index (SDI) is a promising and rigorously developed index that focuses on health care access and health outcomes in primary care service areas (PCAs).³³ Other indexes that link to a given patient’s home address include the Neighborhood SocioEconomic Status Index and the Neighborhood Deprivation Index,^35,39 and other international indexes in countries such as New Zealand and the United Kingdom.³²

It is unknown whether a “best” method exists for calculating neighborhood socioeconomic disadvantage. An ideal index would have published, peer-reviewed methods; would be inexpensive, convenient (i.e. no front-line data collection), valid, reliable; and would correlate with multiple health outcomes across many ages. The index would also have the potential to be expanded and used at a national level to provide comparable data and a universal language “tool” for organizing principals based on SDOH for population health.³²

The Area Deprivation Index

Despite these other promising indexes, we chose to use the Area Deprivation Index (ADI) as our area-based measure of socioeconomic disadvantage because it met the aforementioned criteria for use of the current version, can easily be used immediately, and has potential for expansion.¹ In general, the ADI is widely studied, well-established and has been validated for small geographic regions (e.g., County, Census Tract, Census Block Group).^1,23 Further, ADI quintiles and deciles have been correlated with all-cause mortality and childhood mortality,⁴⁰ mortality from cardiovascular disease and cancer,^41–44 prevalence of cervical cancer,^42,45 and 30-day rehospitalization rates.²³ The ADI was created based on similar indexes in other countries that were being used for resource planning and health policy development.²³ Calculation of the ADI is inexpensive and only requires free and downloadable data from the U.S. Census Bureau.

Thus, the strength of the ADI is that it can be used to examine relationships between socioeconomic factors and health, and to inform policy development, provision of care, resource allocation, and to adjust risk.⁴⁶ Further, it does not require time-consuming and potentially intrusive questions for people and families, and it can easily be made available to different stakeholders.²³ Other single indicator and construct-based approaches have been attempted to predict health outcomes (e.g., rehospitalization), but they have shown mixed results.^23(p.773) In general, the ADI offers a holistic approach to predicting health outcomes by integrating multiple indicators of socioeconomic status based on theoretical grounding and sound measurement principles.

Estimating ADI Scores

The ADI was originally created by Singh¹ using principal components analysis (PCA) to estimate a component score coefficient (or weight) matrix to weight the 17 indicators, which include categories such as income/poverty, housing, employment, and education (see Table 1), and produce component scores for each observation, in this case a Census Block Group (CBG). These weights are called factor score coefficients or “Singh coefficients” by other authors, but it is important to note the distinction between component coefficient scores that result from PCA and factor coefficient scores that result from factor analysis (FA), as the different statistical approaches of PCA and FA produces different coefficient weights and ultimately different component and/or factor scores.

It is not explicitly stated what factor/component score method Singh and other authors have used; though, it is likely the regression score method as it is widely available and popular in common statistical software and languages, such as SPSS (where it is the default method), SAS, R and Mplus.⁴⁷ If using PCA, only exact component scores can be produced, which is equivalent to the regression score method (Note, this is true in SPSS as well, even though other methods are available, SPSS will default to the regression method for PCA).⁴⁸ Since factor score coefficients from FA are approximations, as opposed to component score coefficients from PCA, alternative methods to compute them exist and compete.^47,48

Prior to creating the final ADI score, each of the 17 indicators must first be standardized in some way to ensure they are on the same scale. When items are on the same scale unit they can remain “raw” and unstandardized, but if on different scales the items must be standardized for analysis of correlations (e.g., PCA) or centered for analysis of covariances (e.g., FA).^47,48 This step is very important because if the 17 indicators are not on the same scale, their contribution to the overall ADI will depend on the indicator’s numerical units and not its component/factor score. For example, median home value (e.g. \$100,000) will overshadow unemployment rate (e.g. 0.06) regardless of the factor score weight. Following this standardization, the weighted indicators were summed and converted into a standardized score with a base mean of 100 and a standard deviation of 20. Smaller ADI values indicate an area that is “less deprived.”^1,23 For example, an area with an ADI of 83 indicates that area has less deprivation than an area with an ADI score of 109.

The University of Wisconsin (UW) Health Innovation Program published ADI scores for every census block group and ZIP code in the United States^23,46,49 and released the data under a Creative Commons license (HIPxChange website). The ADI released by UW was calculated using 2000 Decennial Census data (U.S. Census Bureau) and the original Singh coefficients.^1,23 Since that time, the Singh coefficients have been updated, recalibrated, and adapted for different regions of the country.^49,50

It is important to be mindful when using the ADI for geographical levels other than those levels defined by the U.S. Census Bureau as the ADI relies on data published by the U.S. Census Bureau. The ZIP code approach used by UW was feasible using the 9-digit ZIP code crosswalk, which was designed to link ZIP codes directly to Census Block Groups and accompanies the Census Block Group level ADI. ZIP Code Tabulation Areas (ZCTAs) are boundaries published by the Census to be geographically similar to postal ZIP codes, but are substandard because large geographic areas with linkage can result in imprecise estimates. This is especially true in areas where concentrations of poverty border affluent regions.⁴⁶ ZIP code to ZCTA crosswalk tables are available from UDS Mapper.⁵¹

Methods

Data Processing & Analysis

ADI by Census Block Group

CBG ADI score

BroadStreet extracted the American Community Survey (ACS) 5-Year Data at the CBG level from the Census Data API⁵² using the Python programming language and the census package from DataMade. The ADI was created using 17 indicators (see Table 1) following the standard methods described previously for each CBG. Methods included re-calibration of the component score coefficients/weights using PCA in order to create updated component scores for each CBG:^1,23,49

1. The weight coefficients were estimated from PCA on the the combined sample of all CBG in the US using the R missMDA package.⁵³ The missMDA package performs principal component methods on incomplete data sets, aiming to obtain scores, loadings and graphical representations despite missing values. Since the ACS data on these 17 variables using outdated missing data methods (e.g., listwise deletion, mean imputation, etc.) resulted in 20-30% of CBG being removed, Bayesian multiple imputation approach was implemented using the missMDA.
   • Note: in performing PCA on all CBG in the United States combined, the assumption is that the factor structure will not vary significantly by geographic location or is invariant by geographic location. This assumption may be unrealistic and introduce measurement error to the estimated ADI values.⁵⁴
2. The 17 indicators calculated from the ACS (see Table 1) were then z-score standardized for each CBG using the population mean (μ) and population standard deviation (σ) for all block groups in the United States combined.
   • As noted previously, this step is important in order to properly weight each indicator.⁴⁷
3. The z-score standardized indicators were then weighted by multiplying each indicator by its weight coefficient to calculate the final indicator score for each indicator.
4. The indicator scores (i.e., weighted indicators) were then summed to calculate an overall ADI score for each CBG or the final overall component score for each CBG (because of aggregation each CBG level acts as an “individual subject.”
5. The standard base value were standardized to an ADI μ = 100 and a standard deviation of σ = 20. This creates the final, arbitrarily scaled ADI score.

ADI percentile rank and ADI quintile categories

CBG ADI scores were ranked for every CBG in United States to create the ADI percentile rank. Then the ADI percentile rank was categorized based on quintile with the highest quintile denoting the 20% most deprived CBGs. Categorizing ADI percentile ranks into 5 quintile categories was a method originally used by Singh¹. Other investigators utilize a 2-category method comparing the highest 15% of ADI scores to the lower 85% of ADI scores.⁵⁰ There is early evidence of a continuous relationship between ADI score and health outcomes.²³ Until a threshold effect is found, categorization remains discretionary.

ADI for Our Community

Total ADI score

The ADI was calculated for Our Community and for the individual counties or ZCTAs comprising Our Community.

The following variables were used:

1. The CBG total population (source: U.S. Census Bureau, ACS 5-year estimate data)
2. The CBG ADI score (calculated in the steps above)
3. The CGB ADI percentile rank of the CBG ADI score (calculated in the steps above)

Steps for calculating ADI for Our Community were as follows:

1. Our Community geography was assembled as either a list of counties or a list of ZCTAs (this selection was based on the preference of the user).
2. The total population of each CBG was assigned to a county or to a ZCTA within Our Community. Because CBGs sometime span boundaries of ZCTAs, the proportion of the CBG population attributed to each ZCTA was assigned based on a crosswalk file obtained from the Missouri Census Data Center.
3. The population of the CBG attributed to Our Community was calculated as a percent of the population of Our Community.
4. A population-weighted mean of all CBG ADI scores was calculated to create a total ADI score for Our Community. The same process was used to create total ADI scores for each county or ZCTA within Our Community.

ADI percentile rank distribution in Our Community (overall and by race/ethnicity)

The percent of the total population of Our Community living within each ADI percentile rank and ADI quintile category was calculated. The same was done for the population of White (non-Hispanic) and Black (non-Hispanic) residents attributed to Our Community. In the case where a CBG was only partially in a ZCTA of Our Community, the racial/ethnic distribution within the CBG was assumed to be homogeneous. Example: If a CBG has a population where 60 out of 100 people identify as White (non-Hispanic) AND only half of the CBG population is attributable to Our Community THEN we assume Our Community has 30 out of 50 people who identify as White.

17 Indicators at the ZCTA- and County-level

The 17 indicators (see Table 1) were reported individually for the counties or ZCTAs comprising Our Community. In order to calculate and display the values for each of the 17 indicators, we calculated them as described in Table 1 using data extracted from the ACS at the ZCTA- or county-level. The percentile rank of each indicator was then calculated for the United States. ZCTA indicators were ranked compared to all other ZCTAs nationally. Ranks were assigned with #1 being the indicator associated with the least amount of deprivation (based on the Factor Score Coefficient). Counties were ranked compared to counties.

17 Indicators for Our Community

The 17 indicators (see Table 1) were reported individually for Our Community. In order to calculate and display the values for each of the 17 indicators, we calculated them as described in Table 1 using data extracted from the ACS at the CBG-level. CBG-level data was then population-weighted for each indicator to create estimates for Our Community. CBG to ZCTA crosswalk methods (if Our Community was defined using a list of ZCTAs) have been described previously by the Missouri Census Data Center. If Our Community was comprised of ZCTAs or counties, then, for each indicator, Our Community was ranked compared to all Census Block Groups in the United States. Counties and ZCTA are ranked compared to all U.S. ZCTAs and counties, respectively as above.

Table 1. Indicators Comprising The Area Deprivation Index from American Community 5-Year Estimates
Indicator Name	Factor Score Coefficient (Coefficients from Knighton for comparison)49	ID# - Table, File Name, Concept	Variable Logic - Variable names* & calculations
Percentage of population age ≥ 25 years with less than 9 years of education	0.096018 (0.0969)	B15003 - Educational attainment for the population 25 years and over	Population 25 years and over with no schooling completed through 8th grade completed divided by total population = (B15003_002E + B15003_003E + B15003_004E + B15003_005E + B15003_006E + B15003_007E + B15003_008E + B15003_009E + B15003_010E + B15003_011E + B15003_012E) / B15003_001E x 1000
Percentage of population ages ≥ 25 years with at least a high school diploma	-0.121689 (-0.1090)	Educational attainment for the population 25 years and over	Population with at least regular high school diploma or GED or alternative credential divided by total population = (B15003_017E + B15003_018E + B15003_019E + B15003_020E + B15003_021E + B15003_022E + B15003_023E + B15003_024E + B15003_025E) / B15003_001E x 100
Percentage of employed persons ages ≥ 16 years in white-collar occupations	-0.118919 (-0.0942)	C24010 - Sex by occupation for the civilian employed population 16 years and over	Males and females in the management, business, science, and arts occupations (or) sales and office occupations divided by total population = (C24010_003E + C24010_039E + C24010_027E + C24010_063E) / C24010_001E x 100
Median family income ($)1	-0.131858 (-0.1082)	B19113 - Median family income in the past 12 months (in inflation-adjusted dollars)1	Median family income in the past 12 months (in inflation-adjusted dollars) = B19113_001E
Income disparity (see 2 variables below)	0.111968 (0.0823)	B19001 - Household income in the past 12 months (in inflation-adjusted dollars)	Income disparity is defined as the log of 100X the ratio of the number of households with less than \$10,000 annual income to the number of households with greater than or equal to \$50,000 annual income1= Singh, 2003
Households with < \$10,000 annual income	Used to calculate Income Disparity	See income disparity	Households with income < \$10,000 = B19001_002E
Households with ≥ \$50,000 annual income	Used to calculate Income Disparity	See income disparity	Households with income ≥ \$50,000 = B19001_011E + B19001_012E + B19001_013E + B19001_014E + B19001_015E + B19001_016E + B19001_017E
Median home value ($)	-0.090188 (-0.0764)	B25077 - Median value ($) for owner-occupied housing units	Median home value ($) = B25077_001E
Median gross rent ($)	-0.094116 (-0.0675)	B25064 - Median gross rent ($)	Median gross rent ($) = B25064_001E
Median monthly mortgage ($)	-0.092298 (-0.0823)	B25088 - Median selected monthly owner costs ($) by mortgage status	Median selected monthly owner costs ($) for housing units = B25088_001E
Percentage of owner-occupied housing units (home ownership rate)	-0.076278 (-0. 0708)	B25003 - Tenure	Housing units that are owner occupied divided by total housing units = B25003_002E / B25003_001E x 100
Percentage of civilian labor force population ages ≥ 16 years unemployed (unemployment rate)	0.082570 (0.0826)	B23025 - Employment status for the population 16 years and over	The population in the civilian labor force who are unemployed divided by total civilian labor force = B23025_005E / B23025_003E x 100
Percentage of families below the poverty level	0.126313 (0.1074)	B17010 - Poverty status in the past 12 months of families by family type by presence of related children under 18 years by age of related children	Families with income in the past 12 months below poverty level divided by total families = B17010_002E / B17010_001E x 100
Percentage of population below 150% of the poverty threshold	0.139956 (0.1157)	C17002 - Ratio of income to poverty level in the past 12 months	Population whose ratio of income to poverty level in the last 12 months is less than 1.50 divided by the total population with known poverty status = (C17002_002E + C17002_003E + C17002_004E + C17002_005E) / C17002_001E x 100
Percentage of single-parent households with children ages under 18 years	0.066693 (0.0810)	B11005 - Households by presence of people under 18 years by household type (alt. SF1P20)49	Single-parent households with one or more people under 18 years divided by all single-parent households = (B11005_005E + B11005_008E) / (B11005_005E + B11005_008E +B11005_014E + B11005_017E) x 100
Percentage of occupied housing units without a motor vehicle	0.066868 (0.0806)	B25044 - Tenure by vehicles available	Owner and renter-occupied housing units with no vehicle available divided by total occupied housing units = (B25044_003E + B25044_010E) / B25044_001E x 100
Percentage of occupied housing units without a telephone	0.045355 (0.0809)	B25043 - Tenure by telephone service available by age of householder	Owner and renter-occupied housing units with no telephone service available divided by total occupied housing units = (B25043_007E + B25043_016E) / B25043_001E x 100
Percentage of occupied housing units without complete plumbing (log)	0.038256 (0.0422)	B25049 - Tenure by plumbing facilities (alt. B25016 - Tenure by plumbing facilities by occupants per room)49	Owner and renter-occupied housing units lacking plumbing facilities divided by total occupied housing units on a logarithmic scale = LOG of (B25049_004E + B25049_007E) / B25049_001E
Percentage of occupied housing units with more than 1 person per room (i.e. crowding)	0.069035 (0.0731)	B25014 - Tenure by occupants per room	Owner and renter-occupied housing units with 1.01 or more occupants per room divided by total occupied housing units = (B25014_005E + B25014_006E + B25014_007E + B25014_011E + B25014_012E + B25014_013E) / B25014_001E x 100
NOT IN Area Deprivation Index - only add-ons
Race	N/A	B02001 - Race	TBD
Insurance Status	N/A	B27010 - Types of health insurance coverage by age	TBD
*Database labels are aligned with Census Data API: Variables in /data/2015/acs5/variables. More information on the Census Data API for developers is available at the Census Data API User Guide. All totals are the total for the select geography. The Index Variable for all is geo.id or geo.id2. 1Previous-year estimates where used if most recent American Community Survey Data was unavailable.

Limitations, Important Notes & Discussion

There are several potential limitations for the ADI that are worth mentioning: 1) More current and different factor analysis methods exist that may produce different results;⁵⁵ 2) Using geographic-based measures, such as the ADI, can lead to ecological fallacy in which geographic aggregate traits are incorrectly attributed to individuals, especially using surveys that aggregate results such as the U.S. Census American Community Survey or the Decennial Census. Thus, the ADI should only be used in conjunction with more thorough assessments;^23(p.772); and 3) Though the ADI has been built on sound measurement principals and statistical factor analysis, more work needs to be done on the predictive validity of the ADI or where the “rubber meets the road” to determine how the ADI can help build healthy communities.

References