Historic Redlining and Impact of Structural Racism on Diabetes Prevalence in a Nationally Representative Sample of U.S. Adults

Diabetes is the seventh leading cause of death in the U.S. and is associated with decreased quality of life and increased health care use and cost (1). Poorly controlled diabetes over time leads to significant complications, with diabetes being the leading cause of kidney disease, amputation, and blindness in the U.S. (1). Racial and ethnic minorities are more likely to have diabetes, experience complications of the disease, and have worse glycemic control (1–3). For example, Black people in the U.S. are 73% more likely to have diabetes and more than twice as likely to die of diabetes-related causes compared with non-Hispanic White people (2,4,5).

Differences in health outcomes for racial and ethnic minorities is a major concern for efforts aimed to improve population health (6). Recently, more attention is being given to structural factors existing upstream of social determinants of health that perpetuate these racial and ethnic disparities (7–10). For example, a recent systematic review on the role of structural racism as an upstream factor affecting diabetes outcomes found significant relationships between structural racism and worse clinical outcomes (namely, HbA_1c and blood pressure), worse self-care behaviors (diet and physical activity), lower standards of care, higher mortality rate, and more years of life lost for adults with diabetes (11). Structural racism, defined as the totality of ways in which societies foster racial discrimination via mutually reinforcing inequitable systems, is hypothesized to influence health through societal sectors such as housing, education, employment, criminal justice, and health care, and through differential distribution of resources (7,12,13).

Much of the current evidence on structural racism uses historic redlining as a proxy measure. Historic redlining is the previously legal practice initiated in 1934 by the Federal Housing Administration of systematically denying residents in economically disadvantaged areas access to credit and insurance (14,15). The Home Owners’ Loan Corporation (HOLC) created residential security maps grading neighborhoods on a color-coded scale (green, indicating the best, through red, indicating hazardous); hence the term redlining (14,15). Redlined areas were more often inhabited by racial and ethnic minority groups, which led to structural exclusion from homeownership and wealth due to laws allowing banks to deny access to credit and insurance for homes graded as hazardous (14,15). Analyses to date show present-day impacts of historic redlining, including residential segregation, poor mental health, lower life expectancy at birth, higher preterm birth, worse COVID-19 outcomes, and worse cardiovascular disease risk (16–23). Specific to diabetes, a study showed that redlining was associated with 53.7% higher diabetes mortality rates compared with residents of neighborhoods where redlining did not occur (24).

However, limited information exists on pathways and mechanisms underlying the relationship between structural racism, as measured by historic redlining, and diabetes outcomes. Mechanisms identified in the literature theoretically linking structural racism to health include economic injustice and social deprivation, environmental and occupational health inequities, psychosocial trauma, targeted marketing of health-harming substances, maladaptive coping behaviors, and inadequate health care (7). Therefore, the objective of this study was to investigate the direct and indirect relationships between structural racism and the prevalence of diabetes in a national sample, using historic redlining as a surrogate for structural racism and hypothesized pathways based on a conceptual model developed from qualitative research on the impact of structural racism on health in inner-city environments.

The conceptual model was developed on the basis of review of the extant literature on structural racism, known pathways, and data from more than 30 focus groups and 350 stakeholder interviews (25). Thus, the model is grounded in both the literature and lived experience of participants and identifies social factors, including discrimination, incarceration, poverty, residential segregation, substance abuse, housing instability, food insecurity, education, and unemployment, which results in chronic stress and, ultimately, in poor health (25). Figure 1 presents the conceptual model developed from this work to test the direct and indirect influences of structural racism on diabetes prevalence.

To test this conceptual model, multiple data sources were combined at the census-tract level. First, data on diabetes prevalence were sourced from the Centers for Disease Control and Prevention’s PLACES 2019 database (26). Second, data on historic HOLC gradings were sourced from the Mapping Inequality project (27). Finally, additional census tract data on incarceration, poverty, discrimination, substance abuse related to smoking and binge drinking, housing instability, education, unemployment, food access, and demographics were sourced from the Opportunity Insights database (https://opportunityinsights.org/data/?geographic_level=99&topic=0&paper_id=0#resource-listing). The resulting analytic sample consisted of 11,375 U.S. census tract observations.

The outcome measure was prevalence of diabetes among adults measured at the census tract level. This measure is based on model-based census tract estimates for the crude diagnosed diabetes prevalence for the adult (aged >18 years) U.S. population (26,28).

Each census tract was assigned an HOLC grade on the basis of the census tract’s overlap with historically HOLC-graded areas from historic HOLC maps (27). Census tracts that mapped one to one with historic gradings were assigned that grade on the basis of the following scoring scale: neighborhoods with a grade of “A,” representing areas graded as “best,” were given a score of 1; those with a grade of “B,” representing areas graded as “still desirable,” were given a score of 2; those with a grade of “C,” representing areas graded as “definitely declining,” were given a score of 3; and areas with a grade of “D,” representing areas graded as “hazardous,” were given a score of 4. In cases where census tracts overlapped several different HOLC-grading areas, a grade was assigned on the basis of the weighted average grade of its overlapping areas. For example, a census tract that had a 50% area overlap with a B-graded area and a 50% overlap with a C area would be assigned a score of 2.5. This historic redlining score has been used in prior work studying the effect of historic redlining exposure (24,29).

Each of the hypothesized pathways in Fig. 1 was defined using data from the Opportunity Insights database (https://opportunityinsights.org/data/?geographic_level=99&topic=0&paper_id=0#resource-listing) or the structural racism effect index data (30) at the census tract level on incarceration, poverty, discrimination, substance abuse related to smoking and binge drinking, housing instability, education, and unemployment. Incarceration was defined as the fraction of the population within a census tract that was incarcerated. Poverty was defined as the income and poverty domain score of the structural racism effect index (30). The income and poverty domain is an index of variables representing individuals living below 100% of the federal poverty level, individuals living below 200% of federal poverty level, individuals receiving public assistance, family income, per capita income, and the supplemental poverty measure from the Census Bureau (30). Discrimination was defined using the social cohesion domain score of the structural racism effect index (30). The social cohesion domain score is an index of variables representing changed address in the past year, single-parent households, income gap, and residential segregation in the census tract (30). Substance abuse was based on the prevalence of binge drinking in the past 30 days, and the prevalence of smoking reported in a census tract. Housing instability was defined as the housing domain score of the structural racism effect index (30). The housing domain is an index of variables representing housing units without telephone and without plumbing, and with crowding, group quarters, foreclosure risk, and an eviction rate (30). Education was defined as the education domain score of the structural racism effect index (30). The education domain is an index of variables representing receiving a bachelor’s degree or higher, high school diploma, and per-pupil spending in school districts (30). Employment was based on the percentage of the census tract population employed. Food access was captured using the built-environment domain of the structural racism effect index (30). The built-environment domain is an index of variables representing building vacancy rate, mobile homes, and no internet access, as well as the U.S. Department of Agriculture Food Access Research Atlas measure of low food access for Supplemental Nutrition Assistance Program recipients (30).

Structural equation modeling was used to investigate direct and indirect pathways of the relationship between historic redlining and diabetes prevalence at a census tract level. Structural equation modeling is a methodology that combines regression and factor analysis techniques to allow incorporation of multiple independent and dependent variables simultaneously (31–33). The model specified in Fig. 1 was used to specify hypothesized paths, with statistically significant relationships indicating data supported the hypothesized pathway. Before testing the model, assumptions for structural equation modeling were tested, including multivariate normality for all variables. Next, the model was specified using the maximum likelihood estimation procedure and standardized coefficients. Standardized coefficients allow for measures on different scales to be compared by standardizing against the mean and SD of each variable (31). Stata software, version 17, was used.

Following best practices for reporting structural equation modeling results, multiple fit statistics were investigated to understand discrepancy between the sample and model covariance matrices. Because the χ² statistic is sensitive to sample size, fit was primarily investigated using the root mean square error of approximation (RMSEA) less than 0.05 and comparative fit index (CFI) greater than 0.9 (34). RMSEA can range from 0 to infinity, with values less than 0.05 indicating good fit and 0 indicating very good fit. CFI can range from 0 to 1, with values greater than 0.9 indicating good fit and 1 indicating very good fit (34). Standardized coefficients that were statistically significant (P < 0.05) were included in the final model, with the strength of the relationship based on comparison of the standardized estimates (i.e., higher estimates indicated stronger relationships). Though structural equation modeling requires a larger sample size to ensure stable estimates without oversaturating the model, the sample size of 11,375 U.S. census tracts is sufficient for the best practice of a 20:1 ratio (subject to variable) to maintain 80% power.

Table 1 provides summary statistics for variables used in the analysis. The mean prevalence of diabetes was 11.8% (SD 4.9). Figure 2 shows the final model, and Table 2 presents the standardized direct, indirect, and total effects for the relationship between historic redlining and diabetes prevalence. Standardized estimates can be interpreted as a change in SD of the outcome resulting from 1 SD of the predictor. Therefore, higher estimates indicate more change in the outcome and, therefore, a stronger relationship between the predictor and outcome. Consistent with our conceptual framework, in the final model, redlining has significant direct and indirect relationships with diabetes prevalence. Redlining was directly associated with higher crude prevalence of diabetes within a census tract (r = 0.01; P = 0.008) after adjusting for the 2010 population (χ²₍₆₅₎ = 96,537.39; P < 0.001; RMSEA = 0; CFI = 1). Redlining was also directly associated with incarceration (r = 0.27; P < 0.001); poverty (r = 0.35; P < 0.001); discrimination (r = 0.30; P < 0.001); substance use measured by binge drinking (r = −0.09; P < 0.001); substance use measured by smoking (r = 0.28; P < 0.001); housing instability (r = 0.28; P < 0.001); education (r = 0.26; P < 0.001); unemployment (r = −0.36; P < 0.001); and food access (r = 0.27; P < 0.001) after adjusting for the 2010 population. Redlining was indirectly associated with diabetes prevalence via incarceration (r = 0.06; P < 0.001), poverty (r = −0.10; P < 0.001), discrimination (r = 0.14; P < 0.001); substance use (measured by binge drinking: r = −0.65, P < 0.001; and smoking: r = 0.35, P < 0.001), housing instability (r = 0.06; P < 0.001), education (r = −0.17; P < 0.001), unemployment (r = −0.17; P < 0.001), and food access (r = 0.14; P < 0.001) after adjusting for the 2010 population.

A graphical flowchart overview of our analytic sample creation is provided in Supplementary Fig. 1; diabetes prevalence distribution by census tract is shown in Supplementary Fig. 2; descriptive statistics across census tracts with both diabetes prevalence and HOLC score data are reported in Supplementary Table 1; descriptive statistics for our complete analytic sample (i.e., across tracts with no missing variables across all of our measures) are reported in Supplementary Table 2; and pairwise correlations for historic redlining, hypothesized pathways, and diabetes prevalence are listed in Supplementary Table 3.

Using national data measured at the census tract level, this study found that structural racism, as measured by historic redlining, has significant direct and indirect relationships with diabetes prevalence. On the basis of these findings, incarceration, poverty, discrimination, substance use, housing, education, unemployment, and food access may be possible targets for interventions aiming to mitigate the impact of structural racism on diabetes.

These findings contribute to a growing literature of studies of the direct pathways between structural racism and population health and provide additional information regarding possible indirect pathways for targeting interventions. In line with prior work examining negative associations between historic redlining exposure and population health outcomes, we found in this study a significant direct pathway between historic redlining and diabetes prevalence (11,24,35). Prior work identified a relationship between structural racism and diabetes prevalence, as well as relationships between structural racism and constructs identified as possible pathways in this study, including substance use, poverty, home ownership, home evictions, and education (29,36–38). This study is unique, to our knowledge, in its testing of these measures not as indicators associated with structural racism but as possible indirect pathways between structural racism and diabetes prevalence. For example, concentrated poverty has been identified as a present-day consequence of structural racism, with evidence supporting the significant association between structural racism and poverty across historically redlined areas in the U.S. (11).

The findings of this study are consistent with existing evidence and offer new insight by showing that when accounting for multiple factors together, poverty remains a significant indirect pathway through which structural racism increases diabetes prevalence. As a result, although poverty remains a critical component in the pathway between structural racism and diabetes prevalence, it may not be the strongest driver, and as such, multifaceted approaches that include poverty in addition to substance use, home ownership, home evictions, and education should be further explored for informing policy-level and/or community-level interventions that may reduce the impact of these area-level exposures on diabetes prevalence.

More work is needed to identify additional pathways between structural racism and present-day population health, and in developing and testing interventions that target pathways. Because structural racism results in differential access to wealth and resources, work is needed to understand how societal sectors continue to reinforce historic laws. For example, though historic redlining ended in 1968, this study highlights that multiple pathways still exist through which it exerts influence on present-day diabetes prevalence. In addition, work is needed to understand how to address social determinants of health, which encompass many of the pathways identified in this study, in a way that addresses the role of structural factors on differences. Researchers and policymakers may need to turn their attention toward the study of policies that can help ameliorate exposure to social risk factors, rather than focusing primarily on individual-level interventions to address social risk. Finally, work is needed across multiple sectors and multiple levels of influence to make a long-lasting change in population health (10,39). Structural racism is a mutually reinforcing and multisectoral influence, which exerts its impact on multiple domains, as seen in multiple significant pathways identified in this study (7). Instead of focusing on only one sector or one disease type, a concerted effort across disciplines is necessary to test strategies and facilitate change.

Despite the novel investigation using structural equation modeling and national data, this study has several limitations worth noting. First, given the observational study design, estimates should be interpreted as associations and not as casual effects. Findings may help inform policy-level and/or community-level interventions that may reduce the impact of these area-level exposures on diabetes prevalence, but more work is needed using longitudinal or experimental designs to validate findings. Second, estimates were obtained on the basis of an ecological analysis whereby associations were identified using community-level, rather than individual-level, data. Third, health outcomes data are based on model estimates from the Centers for Disease Control and Prevention’s PLACES database. Although prior work validated these small area estimates and used them in other recent ecological studies (22,29,38), we note that the quality of these data depends on the appropriateness of the underlying calculations.

In conclusion, using national data, we found that historic redlining has significant direct and indirect relationships with current-day diabetes prevalence. Incarceration, poverty, discrimination, substance use, housing, education, unemployment, and food access may be possible targets for interventions aiming to mitigate the impact of structural racism on diabetes. Future work should focus on identifying additional pathways, investigating policies that can ameliorate exposure to social risk and coordinating efforts across disciplines and sectors to address the impact of structural racism on individuals with diabetes.

Funding. This study was partially supported by National Institute of Diabetes and Digestive Kidney Disease (grants R01DK118038, R01DK120861; principal investigator [PI], L.E.E.), National Institute for Minority Health and Health Disparities (grants R01MD013826, PIs, L.E.E. and R.J.W.; and R01MD018012 and R01MD017574, PIs, L.E.E. and S.L.).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. L.E.E. conceptualized the study and performed the statistical analyses. J.A.C., R.J.W., and S.L. drafted the manuscript. All authors were involved in critical revision of this manuscript content and approved the final manuscript. L.E.E. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Elizabeth Selvin and Meghana D. Gadgil.