Segregation and Childhood Blood Lead Levels in North Carolina

Blood lead testing data were obtained from the Childhood Lead Poisoning Prevention Program of the NC Department of Health and Human Services, Division of Public Health, Environmental Health Section. In NC, all children enrolled in Medicaid or the Women, Infants, and Children (WIC) Program are required to be tested at 1 and 2 years of age. Beyond children in these programs, health care providers would typically use the Center for Disease Control and Prevention’s lead risk screening questionnaire to determine if a non-Medicaid/non-WIC child should be tested. In theory, all Medicaid and WIC enrollees, as well as all children whose parents responded “yes” or “don’t know” during any pediatric or health department visit to any of the 3 questions on the Center for Disease Control and Prevention Lead Risk Assessment Questionnaire,²²  should have been tested for lead. However, it is difficult to ascertain true practice at the time.

The blood lead testing data include child name, birth date, test date, blood lead level, type of test (venous or capillary), and home address and were provided pursuant to a data use agreement. The NC State Laboratory of Public Health analyzed 90% of the blood samples. During the study period, all children with BLLs <1 μg/dL or below the level of detection were assigned a value of 1 μg/dL. BLLs were recorded and stored in the state database as integer values.

The initial lead test dataset spanned 1992 to 2015 and contained 3 129 048 records. We restricted the dataset to geocoded records (n = 2 497 427) and children with BLLs between 1 μg/dL and 300 μg/dL (n = 2 497 365). For children with multiple test results, we retained the record with the highest value (n = 1 656 309). The dataset was restricted to children who were <7 years of age at the time of their lead test (n = 1 648 144; state requirements for blood lead testing and follow-up testing for elevated BLLs are specified for children <7 years of age) and who were non-Hispanic Black (NHB) or non-Hispanic White (NHW; n = 1 244 992) due to small cell sizes for other racial/ethnic groups. To simplify the analysis, we restricted to blood lead tests administered in 2 date ranges, specifically: 1992 to 1996 (n = 154 366) and 2013 to 2015 (n = 166 550). This resulted in an analysis dataset of 320 916 lead test results from all 100 NC counties.

We previously developed a local, spatial measure of RI, which ranges from 0 to 1 and represents a weighted average proportion of the NHB population in the local neighborhood environment.¹⁵  Although our index is a pure measure of isolation, it has to be constructed with reference to a particular group. Given the disproportionate exposure to and impact of lead on NHB, we made NHB the referent group. A census tract with a neighborhood environment (ie, the tract itself and any tracts adjacent to it) that is predominantly composed of non-NHB individuals will have an RI value close to 0. In contrast, a census tract with a neighborhood environment that is nearly all NHB individuals will have an RI value close to 1.

We calculate our previously developed local, spatial measure of racial isolation of non-Hispanic Blacks (RI_NHB; compared with all other racial/ethnic groups) using 1990 decennial Census data and 2015 American Community Survey data for each NC census tract. For information on the RI calculation, see the Supplemental Information, Appendix A.

US census boundaries may change over time. The 2015 American Community Survey data are reported at 2010 US census tract boundary definitions. We relate the 1990 Census data to 2010 US Census tract boundary definitions using the National Historical Geographic Information System time series tables²³  as described in the Supplemental Information, Appendix B.

Children tested for lead between 1992 and 1996 were assigned tract-level RI_NHB calculated using 1990 Census data, and children tested for lead between 2013 and 2015 were assigned tract-level RI_NHB calculated using the 2015 American Community Survey data. To control for socioeconomic contributors to blood lead levels, children tested for lead between 1992 and 1996 were assigned census tract level percentage of population with a college degree (or higher) and percentage receiving public assistance variables from the 1990 Census, and children tested for lead in 2013 to 2015 were assigned these variables from the 2015 American Community Survey.

Models regressing individual-level BLL on a continuous measure of census-tract level RI_NHB were fit at the state level to estimate the statewide association between BLL and RI_NHB. Models were stratified by child’s race (NHB and NHW) and year (1990 and 2015) and adjusted for season during which the child was tested for lead (winter, spring, summer, fall), tract-level percentage of population with a college degree (or higher), and tract-level percentage of the population receiving public assistance. A random effect at the census tract level was also included.

All statistical analyses were performed by using R, version 4.2.2. Access to the lead vital statistics data described in this research is restricted and governed by data use agreements with NC DHHS and protocols reviewed and approved by the Institutional Review Board at the University of Notre Dame.

Over 25 years, RI_NHB patterns differed across regions of the state (Fig 1). In both the top (1990) and bottom (2015) maps in Fig 1, RI_NHB ranges from 0 (blue) to 1 (yellow), with higher values indicating higher levels of RI_NHB. Comparing the top and bottom maps, we see a swath of yellow concentrated in the eastern part of the state in 1990. This pattern persists in 2015, with some areas exhibiting lower RI_NHB, evidenced by “dimming” of the yellow. In the Supplemental Information, Appendix C, we provide an equivalent map for the entire United States. Figure 1 reveals that although some areas have seen increases in RI_NHB and other areas have seen decreases, the overall distribution of RI_NHB has been relatively stable over the 25-year study period.

We subtract 2015 RI_NHB values from 1990 RI_NHB values to calculate the change in RI_NHB. Negative values indicate an increase in RI_NHB (trending orange/brown); positive values indicate a decrease in RI_NHB (trending dark blue). We see hotspots of increasing RI_NHB in places like Charlotte, Greensboro, Raleigh-Durham, and Greenville. An equivalent map for the entire United States is provided in the Supplemental Information, Appendix C.

Table 1 reveals how RI_NHB has changed across NC and in the 3 standard regions of Mountains, Piedmont, and Coastal Plain. We somewhat arbitrarily define “relatively stable” as changes that range from −0.025 to 0.025. Overall, 38.7% of NC census tracts exhibit stable levels of RI_NHB between 1990 and 2015. Of the remaining tracts, 30.5% exhibit decreasing levels of RI_NHB, and 30.7% of tracts exhibit increasing levels of RI_NHB over the same 25-year period. The greater increases in RI_NHB in the Piedmont are especially noteworthy given that significant economic growth experienced by North Carolina during the study period was concentrated in the Piedmont.²⁴  A histogram showing the distribution of the magnitude of change in RI_NHB for NC between 1990 and 2015 is provided in Fig 2.

We also categorize the changes in RI_NHB using Rural Urban Commuting Area codes, which are developed by the US Census Bureau on the basis of population density, urbanization, and daily commuting flows.²⁵  RI_NHB is more stable in suburban and rural tracts, with the biggest percentage of NC tracts with increasing and decreasing levels of RI_NHB in urban areas (Supplemental Tables 3 & 4). RI_NHB increased in 37.1% of urban tracts and decreased in 28.4% of urban tracts.

In some senses, the most important issue is what is happening at the top end of the RI_NHB distribution. Movement from an RI_NHB value of 0.3 to 0.4 is likely less concerning than moving from 0.7 to 0.8. In 2015, 8.1% of NC Blacks lived in tracts with RI_NHB ≥0.6, 2% of NC Blacks lived in tracts with RI_NHB ≥0.7, 0.3% of NC Blacks lived in tracts with RI_NHB ≥0.8, and 0% of NC Blacks lived in tracts with RI_NHB ≥0.9 (Supplemental Tables 5 & 6). In 1990, the corresponding values were 15.0%, 8.8%, 4.3%, and 0.9%. Thus, although the average RI_NHB has been relatively stable over time, the percentage of NC Blacks living in the tracts with the highest levels of RI_NHB decreased between 1990 and 2015.

The BLL analysis dataset consisted of 320 916 children aged <7 years with a blood lead test result between 1992 and 1996 or 2013 and 2015 and residing in all 100 NC counties. Summary statistics of the study population are provided in Supplemental Table 7.

Average BLLs declined between the early 1990s and 2015, which is evident in the distributions of BLLs by year and race/ethnicity (Fig 3). Mean (median) BLLs among NHB children were 6.7 µg/dL (6 µg/dL) in 1992 to 1996 compared with 1.6 µg/dL (1 µg/dL) in 2013 to 2015. Mean (median) BLLs among NHW children were 4.9 µg/dL (4 µg/dL) in 1992 to 1996 and 1.5 µg/dL (1 µg/dL) in 2013 to 2015.

Preliminary analysis revealed that BLL data were right-skewed. Thus, individual-level BLL (the outcome variable) was natural log-transformed before fitting statistical models. Our 4 models (NHB-1990, NHW-1990, NHB-2015, NHW-2015) regress childhood BLLs on RI_NHB while controlling for season of BLL test, percentage college educated, and percentage receiving public assistance. We included an interaction term between tract-level variables RI_NHB and percent receiving public assistance to help differentiate between race-based and economically-based patterns. Because we include an interaction term between RI_NHB and percentage receiving public assistance, associations between RI_NHB and BLL are reported holding percentage receiving public assistance at its mean. The variables for which we adjusted all behaved in ways consistent with previous research (Table 2).

In the early 1990s, a 1-standard-devation (SD) higher value of RI_NHB was associated with a 2.86% (95% confidence interval: 0.96%–4.81%) greater value in BLL among NHB and a 2.44% (1.34%–3.56%) greater value in BLL among NHW (Table 2). In 2013 to 2015, while attenuated, higher tract-level RI_NHB was still associated with higher BLL for both NHB and NHW children: a 1-SD higher value of RI_NHB was associated with a 1.59% (0.50%–2.70%) greater value in BLL among NHB and a 0.76% (0.08%–1.45%) greater value in BLL among NHW (Table 2).^26,27 

In a sensitivity analysis, we subset the study sample into children who were enrolled in Medicaid versus those who were not and fit separate models for each subsample. Results from these models were similar to the original main analysis results described above (ie, RI_NHB was associated with lead exposure in 1990 and 2015, and the magnitude of the association was smaller in 2015). One exception was observed in models of Medicaid enrollees only, in which RI_NHB was only marginally associated with lead levels among NHW children in 1990 or 2015 (P = .01 in 1990, P = .09 in 2015).

There is no “optimal” value for RI_NHB; the index should be interpreted within the context of overall racial demographics, the associated distribution of social and environmental (dis)amenities, and cultural preferences and attitudes. For example, an RI_NHB value of 0.9 might be considered high, but if it is not accompanied by differences in social or environmental exposures or in health, educational, economic, or developmental outcomes, it could be viewed neutrally. The structural racism drivers, which lead to the cumulation of adverse exposures, shape the impact of RI on outcomes.

Given the relationship between structural racism and racial isolation, it is not surprising that RI is correlated with measures of socioeconomic status.^28,29  Its validity and importance as an independent predictor results from several factors: (1) researchers have shown that disparities exist between Whites and Blacks who experience similar socioeconomic circumstances;^30,31  so despite the correlation, there is something more that RI_NHB is contributing to health outcomes, (2) the models presented in this paper reveal that RI_NHB is an independent predictor above and beyond the measures of SES included in our models, and perhaps most importantly, (3) using racial isolation, rather than measures of SES, directly implicates structural racism in disparate outcomes. The greater increases in RI_NHB in urban areas are notable because ∼80% of tracts in the United States are classified as urban, and the United States is becoming increasingly urban.³² 

Importantly, BLLs are declining over time in North Carolina and across the United States.^1,33  Nonetheless, census tract level RI_NHB is associated with higher BLLs, even as recently as 2015. Measured BLLs reflect exposures to various sources of lead. The leftward shift of the BLL distribution over time, from higher BLLs in the early 1990s to lower BLLs in 2013 to 2015, is unsurprising. The accelerated phaseout of leaded gasoline in motor vehicles, implemented as part of the 1990 Clean Air Act Amendments,³⁴  accompanied by additional restrictions on the use of lead in industrial operations and commercial products, dramatically reduced exposure to lead.^27,35–37  The contribution of airborne lead to BLLs is substantially diminished by the end of the study period.³⁸  Because of the widespread decline in ambient air concentrations of lead after the phaseout of leaded gasoline, children’s BLLs at the end of the study period may be a better marker of exposures sustained in their residential environment. In future work, we plan to investigate whether the level and rate of decline in blood lead levels are related to the rate of change in racial isolation.

Prominent scholars have argued that the RRS of Blacks is a fundamental cause of health disparities among Blacks and NHW in the United States.²⁰  Increasingly, evidence suggests that segregation is associated with health and health disparities.³⁹  The association of RI_NHB with BLL illuminates another pathway through which structural racism and discrimination, and RRS, negatively affect health and development.

The RI_NHB index presented here, available for the entire United States and at multiple points in time, at a highly resolved spatial scale, and across urban, suburban, and rural US communities, provides a powerful tool for assessing how segregation relates to health and what policies or approaches will be most effective at reducing segregation-driven disparities. In the interest of advancing this research, we invite policy researchers interested in how particular policies might be driving (or changing) segregation patterns, as well as outcomes researchers examining RI over time, health, and health disparities, to use the RI index, which is freely available through our Web site (cehi.nd.edu). We note that we have also developed a parallel educational isolation index, which measures the extent to which non-college-educated individuals live in neighborhoods consisting of primarily other non-college-educated individuals. The educational isolation index is also freely available through our Web site.

Structural racism is embedded in historical, societal, institutional, and governmental structures through formal and informal processes and practices that limit opportunities and resources available to specific population groups. Racial residential segregation, produced by a housing system that deliberately sorts people of different races/ethnicities into separate, unequal neighborhoods, is one consequence of structural racism. This research, alongside work by others,⁴⁰  reveals that US neighborhoods remain racially segregated even in the present day. It further demonstrates that racial residential segregation continues to drive higher levels of childhood lead exposure among Black children in the United States. This enduring relationship is one of many detrimental legacies deriving from structural racism. By fostering residential environments inimical to health, RRS serves as a foundation of structural racism and a key contributor to preventable racial/ethnic health disparities in the United States.

The RI_NHB index and measures of its change over time and geography are meant to shift the debate around disparities from using race, a nonmodifiable factor, as an explanatory variable, to the experience of racial minorities, which is modifiable, as a key driver of disparities. Areas with high levels of RI likely experience more adverse social and environmental exposures and may represent the most basic measure of the structural injustices experienced in the United States. This work is timely as the United States wrestles with its history of racial inequalities, which have long been present, but whose fault lines have been starkly revealed through coronavirus disease 2019, protests around unjust policing practices, and a dislocated economy.

We thank Ed Norman for his contribution to earlier versions of the manuscript.

BLLs

blood lead levels

NHB

non-Hispanic Black

NHW

non-Hispanic White

RI

racial isolation

RI_NHB

racial isolation of non-Hispanic Blacks

RRS

racial residential segregation

SD

standard deviation

WIC

Women, Infants, and Children Program