Association of Historical Redlining With Gestational Diabetes Mellitus: The Mediating Role of BMI and Area Deprivation Index Free

Gestational diabetes mellitus (GDM) affects ∼5–9% of pregnancies annually in the U.S., posing serious health risks for both mothers and infants. For women, GDM increases the likelihood of cesarean delivery, preeclampsia, and future development of type 2 diabetes (1–3). Infants born to mothers with GDM face complications such as macrosomia, neonatal hypoglycemia, and increased adiposity, along with a higher risk of developing cardiometabolic disorders later in life (4,5). The etiology of GDM is multifactorial, involving genetic, hormonal, and environmental influences. Although precise mechanisms remain unclear, insufficient insulin production, delayed β-cell response to glucose levels, and pregnancy-related insulin resistance are believed to contribute to GDM (5). Notably, racial and ethnic disparities in GDM prevalence are well-documented, with higher rates observed among Asian, African American, and Hispanic populations compared with non-Hispanic Whites (6).

Structural inequities driven by institutional policies are well-established contributors to health disparities across the U.S. (7). An example of such inequity is historical redlining, a discriminatory practice that systematically denied financial services, including access to credit and insurance, to residents of predominantly minority and low-income neighborhoods. The term “redlining” originated in 1934 from Home Owners’ Loan Corporation (HOLC) security maps, which categorized neighborhoods by the perceived risk of extending mortgage loans, largely based on racial and ethnic composition. These maps divided neighborhoods into four color-coded categories: A (“best”), marked in green; B (“still desirable”), in blue; C (“definitely declining”), in yellow; and D (“hazardous”), in red (8). This grading system effectively segregated affluent, predominantly White neighborhoods (A zones) from poorer, minoritized racial and ethnic groups (D zones), systematically excluding people of color from home ownership and wealth-building opportunities.

Although redlining was abolished in 1968 through the Fair Housing Act (9), its legacy continues to shape present-day health disparities such as higher risk of diabetes, obesity, cardiovascular diseases, severe maternal morbidities, preterm birth, and worse coronavirus disease 2019–related outcomes (10–15). Recent studies have concluded that redlining contributes to higher diabetes prevalence and mortality through pathways such as limited health care access, food deserts, and chronic stress exposure. These disparities create a cumulative risk environment that exacerbates chronic diseases (11). Moreover, redlined neighborhoods often face lower socioeconomic status, limited opportunities for physical activity, and higher exposure to environmental pollutants, all of which are associated with increased diabetes risk and related morbidities and mortality (16,17).

Analyzing the link between residential segregation and health disparities requires understanding the complex interplay of multiple factors influencing maternal health outcomes. These factors, spanning both individual-level and area-level determinants, contribute to why some pregnant individuals are at higher risk for maternal comorbidities like GDM. Previous literature has identified key pathways linking structural racism to poor pregnancy outcomes, emphasizing individual-level factors such as BMI, stress, and physical activity, as well as area-level factors like socioeconomic deprivation, environmental quality, and neighborhood infrastructure (2). Maternal obesity, as reflected by pregestational BMI, is a well-established risk factor for GDM, as it influences metabolic health and insulin resistance during pregnancy. Additionally, neighborhood-level deprivation, captured by present-day area deprivation index (ADI), reflects long-term effects of redlining, including reduced access to health care, nutritious food, and safe living environments (18–20). These individual- and area-level factors act through mechanisms such as chronic stress, poor nutrition, limited health care access, and environmental exposures, all of which contribute to the pathophysiology of GDM (5,21).

Redlining, as a policy that historically segregated communities and restricted access to resources, has left a legacy of neighborhood deprivation, leading to higher levels of maternal obesity and increased ADI scores in affected areas. The mediation analysis in this study aims to examine these complex relationships, specifically exploring how redlining influences maternal obesity and neighborhood deprivation, which, in turn, may contribute to GDM risk. By understanding these pathways, we can better identify key areas for targeted public health interventions, addressing both individual and structural drivers of health disparities.

This study aims to investigate the association and underlying pathways between structural racism, measured by redlining and risk of GDM, using a large and diverse pregnancy cohort derived from Kaiser Permanente Southern California (KPSC) electronic health records (EHRs) data from 2008 to 2018. We hypothesize that residence in areas with worse HOLC grading is associated with a higher risk of GDM. Furthermore, we assess whether this association is mediated by maternal obesity and present-day neighborhood deprivation as measured by ADI.

This study included women residing in areas previously graded by HOLC (A [“best”], B [“still desirable”], C [“definitely declining”], and D [“hazardous”/redlined]), and gave birth to singleton neonates between 1 January 2008 and 31 December 2018 at KPSC facilities, a large integrated health care system in Southern California.

From 2008 to 2018, 439,254 KPSC mothers were preliminarily identified in this large pregnancy cohort after excluding multiple births (n = 7,454) and stillbirths (n = 1,961). The study excluded participants with preexisting diabetes (n = 5,518) and those with missing GDM laboratory test results (n = 7,626). We further excluded those living outside HOLC map coverage (n = 331,822), because HOLC maps were only created for two Southern California cities, Los Angeles and San Diego. Both cities are located on the southern Pacific coast of California, with populations of 3.85 million and 1.41 million, with an area of 501.6 square miles and 372.4 square miles, respectively (22,23). We extracted information regarding demographics, medical records, birth records, individual lifestyle, and residential history during pregnancy from KPSC EHRs. Further, we used self-reported information on birth certificates to determine maternal race and ethnicity categorized as non-Hispanic Black, non-Hispanic Asians, Hispanics, non-Hispanic Whites, and others (e.g., Pacific Islanders, Native American/Alaska Native and other racial and ethnic groups). Additionally, gestational age was estimated based on the date of last menstrual period and corroborated by first or second trimester ultrasonography.

The diagnosis of GDM for all participants followed standard guidelines and was obtained using KPSC laboratory glucose tolerance tests. Routine GDM screening for most pregnant women occurred between the 24th and 28th weeks of gestation, while those at elevated risk underwent earlier assessment. Approximately 78% of GDM case patients in our study were diagnosed during the late second trimester. One of two diagnostic criteria was employed for GDM diagnosis: the Carpenter-Coustan protocol, that is, a 1-h post–50-g glucose challenge result ≥200 mg/dL, or, following a 50-g glucose challenge result of 130–140 mg/dL to 199 mg/dL, the presence of two or more abnormal values during a 3-h 100-g oral glucose tolerance test (OGTT), with specific cutoffs (fasting ≥95, 1 h ≥180, 2 h ≥155, and 3 h ≥140 mg/dL) (24); or the International Association of Diabetes and Pregnancy Study Groups (IADPSG) criteria. All but one of the participating centers used the Carpenter-Coustan protocol, with 0.09% of women with GDM identified using the IADPSG protocol. The latter necessitated the identification of at least one abnormal value during a 2-h 75-g OGTT, with specific cutoffs (fasting ≥92, 1 h ≥180, and 2 h ≥153 mg/dL) (25).

Redlining was assessed using digitized HOLC maps from Mapping Inequality (8). All maternal residential addresses during pregnancy in KPSC EHRs were geocoded and assigned HOLC grades (A, B, C, or D) based on their geographic location within HOLC-graded zones. HOLC maps from Los Angeles and San Diego cities were merged into one shapefile and overlaid onto geocoded addresses. Births to women who resided in neighborhoods that fell outside the HOLC map boundaries were not included in the analysis; however, the demographic statistics for those births are attached in Supplementary Table 1. Births to women who lived within the HOLC boundaries between 2008 and 2018 were assigned corresponding HOLC grades (A, B, C, or D). For GDM case patients, exposure was assigned based on the patient’s residential address at the diagnosis date; for noncase patients, it was assigned based on the residential address during the 24th to 28th gestational week, the typical GDM testing window. To ensure robustness, we performed a sensitivity analysis using the patient’s residential address at the diagnosis date for case patients and address at the time of delivery for non–GDM case patients. The results are attached in Supplementary Table 2. The population in HOLC-graded A, the “best” zone, was kept as a reference zone and was further compared with the population residing in B-, C-, and D-graded zones.

Although other potential mediators, such as individual socioeconomic status, health care access, or physical inactivity, were considered, pregestational BMI (reflecting obesity) and ADI (capturing neighborhood deprivation) were selected based on the availability of data and their established roles in the pathway between historical redlining and GDM. Redlining contributes to higher rates of obesity by limiting access to opportunities for physical activity, healthy food, health care, and clean environment, which, in turn, may increase the risk of GDM. Similarly, redlining-driven neighborhood deprivation, as captured by ADI, reflects socioeconomic conditions that contribute to maternal health disparities, including risk of GDM.

We extracted information on prepregnancy BMI as measured in kg/m² that was categorized as underweight (<18.5), normal (18.5–24.9), overweight (25.0–29.9), and obese (≥30.0). Further, we used ADI as an indicator of neighborhood socioeconomic disadvantage. It is a composite measure of 17 unique characteristics derived from 2011–2015 American community survey data, including income, education, employment, and housing quality at census block group level, indicated by a numerical score from 1 to 100 (100 being most disadvantaged) nationwide, and from 1 to 10 (10 being most disadvantaged) statewide (26). Each census block group received two ADI scores: a national percentile ranking and a within-state decile ranking. We used the latter, aiming to reduce confounding from state-level variation. Similar to logistic models, for the mediation model, exposure was assigned based on a patient’s residential address at diagnosis date for case patients and residential address during the 24th to 28th gestational week for noncase patients. To match census block groups, these addresses were linked to their respective ADI scores.

Summary statistics were calculated for the characteristics of the study population. For continuous variables such as maternal age, an independent t test was used to compare means between GDM case patients and non-GDM pregnancies, while χ² tests were used for categorical demographic characteristics.

To assess associations between redlining and GDM risk, we kept participants residing in the A-graded HOLC zone as a reference and performed mixed-effect logistic regression models. We included covariates based on prior literature and our directed acyclic graph plot (Supplementary Fig. 1). In the main analysis, we adjusted for a minimal set of true confounders, including maternal age, race and ethnicity, income, education, and smoking status. To address potential spatial clustering, zip code was included as a random effect (for intercept) in the model. This approach accounts for unobserved local-level characteristics, such as environmental exposures or local health care access, that may influence GDM risk and are shared by individuals within the same residential zip code. Covariate information was gathered from KPSC EHRs. The model was further adjusted for maternal race and ethnicity (categorized as non-Hispanic Black, non-Hispanic Asian, Hispanic, non-Hispanic White, and others), maternal age, household income at block group level in 2013, and education. The results were represented by adjusted odds ratios (aORs) with corresponding 95% CIs.

We further examined whether ADI and maternal BMI mediated the association between historical redlining and GDM risk across different HOLC-graded zones, using the inverse odds ratio weight (IORW) mediation method. Mediation analysis involves estimating the total effect of living in HOLC-graded zones on GDM, and decomposing this association into its direct (unmediated) and indirect (mediated) association components. The natural direct effect refers to the relationship between HOLC-graded zones exposure and GDM, assuming the mediator is at the value it would naturally take for a given level of exposure. The natural indirect effect describes the relationship between exposure’s influence on a mediator and the mediator’s subsequent association with outcome. We examined mediation by neighborhood deprivation and BMI independently using the following approach. First, we applied logistic regression to estimate the total effect of HOLC-graded zones on GDM for each mediator, adjusting for potential confounders. Next, we calculated direct and indirect effects of HOLC-graded zones using the Tchetgen inverse odds weighting method (27). Unlike traditional methods, IORW does not rely on functional form assumptions, reducing the risk of bias from model misspecification, particularly when the relationship between the mediator and outcome is complex. Additionally, it adjusts for mediator-outcome confounders by applying weights based on inverse odds of the mediator, without requiring explicit modeling of these confounders. The IORW method also allows for inclusion of multiple mediators, regardless of their scale, making it more flexible in analyzing complex pathways. Overall, its nonparametric nature and ability to handle multiple mediators with fewer assumptions make it particularly suitable for our analysis (27). Further, interpretation of results follows a logic similar to traditional mediation methods: indirect effect represents the portion of the HOLC-GDM association mediated by BMI or ADI, and direct effect represents the HOLC-GDM association not mediated by BMI or ADI. Additionally, mediation models were adjusted for individual-level confounders including maternal age, race, ethnicity, block group level income, education, and smoking. Since the IORW method can only accommodate binary variables, separate analyses were carried out for A versus B, A versus C, and A- versus D-graded zones.

We conducted following sensitivity analyses by 1) assigning exposure based on patients’ residential addresses at GDM diagnosis for case patients and delivery for noncase patients (Supplementary Table 2); 2) including additional variables either clearly associated with the outcome or having unclear evidence of confounding (i.e., insurance type and year of childbirth) to account for potential confounding. Maternal country of birth and parity were added because of their associations with GDM risk through genetic predisposition, cultural practices, health care access, cumulative metabolic changes, and socioeconomic exposures across pregnancies (27) (Supplementary Table 3); and 3) evaluating the impact of maternal characteristics through subgroup analyses stratified by maternal race and ethnicity, country of birth, age, median household income, education, and BMI were conducted to identify their potential effect modification (6,28). We measured the heterogeneity among subgroups using the Cochran Q test (Supplementary Table 3).

All analyses were performed with R 4.1.3, and SAS software version 9.4 (SAS Institute, Cary, NC). The study was approved by KPSC (Pasadena, CA) and the University of California, Irvine Institutional Review Boards (Irvine, CA) with an exemption for informed consent.

A total of 86,834 pregnancies in 2008–2018 with 10,134 (11.67%) GDM case patients were included in the final analyses (Table 1). Approximately 55% of the mothers were residing in previously graded C zones. Hispanic, non-Hispanic White, and non-Hispanic Black mothers accounted for approximately 55%, 17%, and 12% of all mothers, respectively. Compared with the non-GDM group, pregnancies diagnosed with GDM were more likely to be older, non-Hispanic Asian and Hispanic, less educated, obese, and residing in neighborhoods with an ADI ranking between 5 and 8.

The aORs and corresponding CIs of GDM associated with lower HOLC-graded zones are presented in Table 2. Compared with unadjusted results, the associations were modestly attenuated in the adjusted model accounting for race and ethnicity, maternal age, block group-level median household income, and education level, although the results remained significant for grade B (OR = 1.20, 95% CI 0.99–1.44), grade C (OR = 1.22, 95% CI 1.02–1.47), and grade D (OR = 1.30, 95% CI 1.08–1.57) neighborhoods.

From the mediation analysis (Table 3), we found that the association between HOLC-graded neighborhoods and GDM was mediated by BMI and ADI. For maternal pregestational BMI, the total effect for B-, C-, and D-graded zones was 0.017 (95% CI 0.003–0.029), 0.036 (95% CI 0.023–0.050), and 0.035 (95% CI 0.021–0.043), with the proportion mediated being 27.9%, 32.2%, and 44.2%, respectively. For the ADI, the total effect for B-, C-, and D-graded zones was 0.019 (95% CI 0.007–0.031), 0.036 (95% CI 0.024–0.049), and 0.034 (95% CI 0.021–0.044), with the proportion mediated being 28.6%, 41.5%, and 64.5%, respectively.

We conducted several sensitivity analyses to assess the robustness of our findings. First, exposure was assigned based on the residential address at the time of GDM diagnosis for case patients and at the time of delivery for noncase patients. The results were consistent and further supported the validity of the findings (Supplementary Table 2). Second, after adjusting the main model with additional covariates (i.e., year of childbirth, parity, mother’s country of birth, and type of insurance), we found a slightly stronger association with the risk of GDM among mothers residing in grade C (OR = 1.24, 95% CI 1.03–1.49) and grade D (OR = 1.31, 95% CI 1.09–1.59) neighborhoods (Supplementary Table 3). Finally, in the subgroup analysis, we did not find a statistically significant effect modification by maternal race and ethnicity, age, BMI, education, and income. The ORs were higher among non-Hispanic Asian and Hispanic mothers, and mothers born outside the U.S. (Supplementary Table 4).

In this study of 86,834 pregnancies with detailed individual-level EHRs of women from 2008 to 2018, our results suggested that, compared with “best”-graded zones, living in lower-graded HOLC zones during pregnancy was significantly associated with increased risks of GDM. Further, our results support BMI and ADI as a potential partial mediator, explaining 27.9–64.5% of observed increases in GDM risk associated with residing in different redlined neighborhoods.

Our findings suggest that both maternal BMI and neighborhood deprivation (measured by ADI) are key pathways linking historical redlining to GDM, highlighting important implications for health care and community-level interventions. Health care providers should prioritize obesity prevention and consider neighborhood deprivation in GDM risk assessments, with interventions such as personalized diet counseling, prenatal support, and improved care access in economically deprived communities. Community efforts to address deprivation could include increasing walkability, improving access to education and employment, and encouraging grocery chains to stock affordable, nutritious, simple carbohydrate–limited foods in underserved areas to combat food insecurity. Establishing health care offices and pharmacies closer to and within disadvantaged neighborhoods, offering services like prenatal vitamin distribution, hypoglycemic medications, and social support, along with home visits and technology-based treatments (e.g., telemedicine), could further enhance care delivery, especially for women whose access to needed medical care is limited by affordable and accessible transportation (29). Future research is needed to evaluate the effectiveness of these interventions and their potential to mitigate GDM disparities associated with structural racism.

The study revealed a risk of GDM elevated by 20%, 22%, and 30% among women residing in HOLC-designated B-, C-, and D-graded zones, respectively, compared with the risk to those in A-graded zones. These findings add to existing literature on direct pathways between structural racism and perinatal and reproductive health outcomes (12,13), while also highlighting a potential indirect pathway through maternal obesity and area-level deprivation.

Our findings suggest a dose-response pattern across HOLC grades for both direct and mediating pathways, with a particularly strong effect in “yellow-lined” (C-graded) and “redlined” (D-graded) zones, both of which have historically suffered the highest levels of disinvestment and socioeconomic deprivation. Additionally, in our sensitivity analysis (Supplementary Table 3), we further adjusted the model for other GDM-related risk factors such as parity, mother’s country of birth, insurance type, and child’s year of birth. A similar dose-response pattern persisted, with results indicating a modest, borderline-significant association with GDM among pregnancies in B-graded zones (OR = 1.21; 95% CI 1.00–1.46), while the associations for grade C (OR = 1.24; 95% CI 1.03–1.49) and grade D (OR = 1.31; 95% CI 1.09–1.59) zones remained statistically significant. This suggests that, even when controlling additional individual-level maternal factors, risk of GDM increases as HOLC grades worsen, with the highest risk observed in grade D areas. The results are aligned with previous research indicating that the legacy of redlining continues to affect health outcomes in a gradient-like fashion, impacting areas historically graded as C and D most severely.

Previous studies have documented potential mechanisms that may support our findings, such as dysfunction of hypothalamic-pituitary-adrenal (HPA) axis, systemic inflammation, and oxidative stress, each of which may cause insulin resistance, and result in subsequent diabetes (30,31). For example, past research has shown residents of redlined neighborhoods may persistently experience racial discrimination, poor walkability, and a wide array of present-day environmental exposures (e.g., higher air pollution, noise pollution, and reduced greenspace) (32–35). These pathways may be biologically transduced by dysregulation of HPA axis, increased levels of proinflammatory cytokines, and generation of reactive oxygen species, leading to insulin resistance, and dysregulation of glucose metabolism (30,36).

The legacy of historic redlining has been tied to obesity as well as neighborhood deprivation (15). Factors such as restricted access to healthy food options and limited neighborhood walkability along with higher poverty rates and greater social vulnerability elevate the risk of obesity and neighborhood deprivation in racially and ethnically segregated communities (34,37,38). The mediation analysis suggests that the association between historical redlining and GDM risk may be partially mediated by maternal obesity and contemporary neighborhood-level socioeconomic factors.

The percentage mediated of 44.2% and 64.5% via BMI and ADI in D-graded zones highlights the substantial role of maternal obesity and neighborhood deprivation in driving redlining-GDM associations. As ADI explains disparities in housing quality, access to education, employment, criminal justice, economic opportunities, and health care, it can be an indicator of neighborhood exposures causing poor health outcomes (39,40). In a previous ecological analysis, Egede et al. (11) identified several mediating pathways linking redlining to diabetes prevalence in the U.S., including factors such as incarceration, poverty, discrimination, substance use, housing, education, unemployment, and food access. Building on this foundation, our study specifically focused on GDM and validated that maternal BMI and ADI serve as significant pathways between redlining and GDM, using individual-level patient data.

Our study has several unique strengths. First, to the best of our knowledge, this is the first study examining the association of redlining and GDM while considering the mediating role of BMI and ADI. Second, KPSC’s high-quality EHR database provided an opportunity to explore a large pregnancy cohort including mothers from different sociodemographic backgrounds. Our results highlight the need to account for residential context in addressing GDM disparities. When assessing the risk of developing GDM, health care providers should consider whether individuals live in historically redlined communities. Further, the findings can inform policy-level or community-level interventions in areas burdened by socioeconomic deprivation, particularly those with a history of redlining. Such interventions are not limited to better walkability and access to healthy foods, medical care, and medications but also to improving access to education, employment, and housing stability in disadvantaged areas. However, more work is needed using longitudinal or experimental designs to validate findings.

Some limitations should be acknowledged. First, given the observational study design, the results of this study should be interpreted as associations rather than causal effects. Second, there is a potential risk of misclassification as there is no set standard for defining HOLC grades using historical city maps that are imperfectly aligned with administrative boundaries such as census tracts. Besides, HOLC is considered a group-level and non–time-varying variable in our study.

Third, the study only included members of KPSC’s health care system in Southern California, which may limit the generalizability of the study to other geographic locations. Fourth, we did not account for potential environmental factors that could contribute to GDM, such as air pollution, access to food and health care, and climate variability. Finally, the mediation analyses reflect statistical associations rather than causal pathways, as assumptions such as “no unmeasured confounder” are unmet. Unmeasured variables, such as environmental exposures and a history of GDM, may act as confounders, potentially biasing the estimates.

In conclusion, the findings of this study in Southern California add to the literature on associations between living in previously poor-graded housing zones and GDM risk. Residents in B-graded (“still desirable”—blue), C-graded (“definitely declining” —yellow), and D-graded (“hazardous,” i.e., redlined—red) zones appeared to have a higher risk of GDM than individuals residing in graded A zones. Further, no effect modification on maternal demographic characteristics was observed. The study found that maternal BMI and neighborhood deprivation represented by ADI was an important pathway linking housing zones and GDM, and the highest proportion of mediation was observed for C- and D-graded HOLC zones. The study highlights the importance of understanding the influence of historically discriminatory housing policies on pregnancy complications among marginalized communities.

Funding. This study was supported by the National Institute of Environmental Health Sciences (R01ES030353).

The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.

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

Author Contributions. W.U. designed the methodology, conducted the formal analysis, and prepared the first draft of the manuscript. Y.S. contributed to the methodology design, data curation, and manuscript review and editing. A.J. contributed to data curation and manuscript review and editing. K.D.L., M.L., C.C.A., J.M.S., D.A.S., and J.M. contributed to manuscript review and editing. V.Y.C. contributed to data analysis through discussion and manuscript review and editing. T.B. contributed to methodology design, data analysis, and manuscript review and editing. J.-C.C. contributed to manuscript review and editing. D.G. supervised and contributed to conceptualization, project administration, funding acquisition, methodology design, data curation, and manuscript review and editing. J.W. supervised and contributed to conceptualization, project administration, funding acquisition, methodology design, data curation, and manuscript review and editing. J.W. and D.G. are the guarantors of this work and, as such, had full access to all the data in the study and take 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.