Dynamic Changes in Place-Based Measures of Structural Racism and Preterm Birth in the US

Abstract

Background:

Structurally racist systems, ideologies, and processes generate and reinforce inequities among minoritized racial/ethnic groups. Prior cross-sectional literature finds that place-based structural racism, such as the Index of Concentration at the Extremes (ICE), correlates with higher infant morbidity and mortality. We move beyond cross-sectional approaches and examine whether a decline in place-based structural racism over time coincides with a reduced risk of preterm birth across the US.

Methods:

We used as the outcome count of preterm births overall and among non-Hispanic (NH) Black and NH white populations across three epochs (1998-2002, 2006-2010, 2014-2018) in 1,160 US counties. For our measure of structural racism, we utilized ICE race/income county measures from the US Census Bureau. County-level fixed effects Poisson models include a population offset (number of live births) and adjust for epoch indicators, percent poverty, and mean maternal age within counties.

Results:

A standard deviation increase in ICE (0.11) over time corresponds with a 0.6% reduced risk of preterm birth overall (incidence rate ratio [IRR]: 0.994, 95% CI: 0.990, 0.998), a 0.6% decrease in preterm risk among NH Black births (IRR: 0.994, 95% CI: 0.989, 0.999), and a 0.4% decrease among NH white births (IRR: 0.996, 95% CI: 0.992, 0.999).

Conclusions:

Movement away from county-level concentrated NH Black poverty preceded reductions in preterm risk, especially among NH Black populations. Our longitudinal design strengthens inference that place-based reductions in structural racism may improve perinatal health. These improvements, however, do not appear sufficient to redress large disparities.

INTRODUCTION

Preterm birth (delivery of a live birth at less than 37 weeks of gestation) increases the risk of infant death, developmental delays, and low educational attainment [1]. In the US, non-Hispanic (NH) Black birthing persons have a 1.5-fold increased risk of delivering preterm as opposed to NH whites (i.e., 13.0 vs 8.9 per 100 live births) [2]. Scholars contend that both macro-level (e.g., structural racism) and micro-level (e.g., access to health care) factors drive this disparity [1,3].

Structural racism is defined as the systems, social forces, institutions, ideologies, and processes that interact with one another to generate and reinforce inequities among racial and ethnic groups [4]. This term emphasizes the most influential levels at which racism affects health disparities [5]. Manifestations of structural racism include, but are not limited to, systems and policies related to employment, education, residential segregation, housing, health care, and the criminal legal system. These macro-level factors introduce, and reproduce, an unequal distribution of health-promoting resources (e.g., educational and economic opportunities) across places by race/ethnicity.

Cross-sectional literature finds that several proxies of place-based structural racism, such as the Index of Concentration at the Extremes (ICE), correlate with higher infant morbidity and mortality, especially among NH Black births [6–16]. The ICE measure draws from a multitude of scholars including but not limited to Massey and Denton’s seminal work on residential segregation by race and class [17–23], as well as Western’s work on racebased disparities in the criminal legal system [24]. ICE reveals how groups spatially concentrate by race and socioeconomic deprivation and privilege [10,25]. Researchers routinely assign ICE measures to specific places such as census tracts and counties, given the recognition of distinct variation across place in the unequal distribution of resources by race.

The above literature on place-based structural racism and birth outcomes, however, remains limited for three, key reasons. First, the work is largely cross-sectional in that it measures place-based structural racism and health outcomes at one point in time [10,7,8,26,14]. This circumstance makes it challenging to draw causal claims or to inform policy change. As much spatial work has demonstrated, places are not exchangeable and therefore not directly comparable “holding constant” all factors except for structural racism [27]. A more rigorous alternative design would involve examination of changes in place-based structural racism over time within a community. Such a longitudinal focus would not only establish temporal precedence between structural racism and birth outcomes, but also determine whether reductions in historical structural racism have the potential to reduce subsequent racial/ethnic health disparities. This longitudinal approach, moreover, may assist with identifying places in which measures of deprivation from place-based structural racism appear to decline over time.

Second, prior work on ICE and birth outcomes uses mostly contemporary (2010 and later) data [7,8,10,14,16,28]. Given the dynamic nature of places, a historical perspective which uses >20 years of data could help identify shifts over time in structural racism. In addition, use of a larger time span covers an epoch in which the US continued its “hollowing out” of the middle class and a substantial widening between the rich and poor—a phenomenon which may adversely affect racial/ethnic disparities in preterm birth [29].

Third, previous work evaluates changes in infant morbidity and mortality as a function of ICE within neighborhoods in individual cities or states in the US including, Boston, Detroit, New York City, and California [7,8,10–12,14,26]. The dynamic changes in racial composition and socioeconomic opportunity in other metropolitan locations over time may also warrant evaluation across the US.

Prior work reports that the incidence of preterm among NH Black births is associated with ICE race/income [7,8,10,14]. Building on this work, we evaluate the role of changing ICE race/income over time on preterm among NH Black births. We incorporate nationwide longitudinal, county-level ICE measures, dating back to 1998, with the potential to uncover important contributors to how changes in place-based structural racism affect changes in NH Black and NH white preterm birth over a 21-year period in the US. We hypothesize that the risk of NH Black preterm birth will decline over time following place-based increases in ICE (i.e., moving away from spatially concentrated NH Black poverty). We base this hypothesis on the cross-sectional literature which finds greater infant morbidity and mortality in deprived neighborhoods [10,7,8,26,14,13,9,11,16].

METHODS

Data

We retrieved data on all live births in the US, from 1998 to 2018, from natality files assembled by the National Center of Health Statistics (NCHS), Division of Vital Statistics. These files include the longest series of natality data, with individual-level variables, available to us at the time of our inquiry. We excluded birth records missing gestational age (<3% of all records) and restricted the analyses to live birth records with plausible gestational age (i.e., 22 to 45 weeks). The files contain month and year of birth, as well as gestational age (GA) in weeks. We used the NCHS combined LMP (based on date of last menstrual period or, if missing or improbable, obstetric estimate) estimate of GA. As the only consistent measure of GA during the study period, we classified preterm birth as any live birth with GA <37 weeks, as done in previous literature[3,7–9,12]. We used preterm birth as the outcome for two reasons. First, prior literature reports cross-sectional associations between ICE race/income and this outcome [7,8,10,14]. Our analysis may therefore permit comparisons with earlier work. Second, unlike infant and fetal deaths, preterm deliveries are sufficiently common to permit stable estimation of “rates” for less populated counties with relatively few births (described in more detail in “Analysis”, below). The University of California, Irvine Committee for the Protection of Human Subjects approved this study (Protocol # 20195444).

We focused on NH Black and NH white births given the persistently large disparity in preterm birth between these two groups [30]. We classified race/ethnicity of the birth according to the race/ethnicity of the birthing person. The restricted-use US natality file has a geographic resolution at the county level. Given our focus on place-based changes in ICE at higher ecological levels, we aggregated race-specific counts of preterm birth to the county-level for each year. We then (for reasons described below) aggregated race-specific preterm births for three epochs: 1) 1998-2002; 2) 2006-2010; and 3) 2014-2018. We utilized race-specific live births to identify the population at risk.

We compiled data needed to construct our dynamic ICE race/income for all US counties from the 2000 US Decennial Census, as well as the 2006-2010 and 2014-2018 American Community Survey (ACS) 5-year estimates for household income by race/ethnicity [31–33]. These three time periods compelled us to align the preterm birth data in the US for these years. We could not obtain data to calculate ICE race/income for 1998, 1999, 2001, and 2002 and instead utilized the publicly available 2000 US Decennial Census given that it serves as a midpoint. As described by Krieger and colleagues [10], we calculated ICE for each county by subtracting the number of NH Black households with an annual income equal to or below the 20^th percentile from the number of NH white households with an annual income equal to or above the 80^th percentile. Because our analysis spanned the entire US, we benchmarked our percentiles to each county’s state-specific thresholds for income levels. The difference is then divided by the total number of county households, bounding our measure from −1 (indicating that all county households are NH Black with annual incomes below the 20^th percentile) and 1 (indicating that all households are NH white with annual incomes at or above the 80^th percentile) [10]. We then created a z-score using our continuous ICE measure for better interpretation of our findings since counties average a 0.3 change in ICE over time rather than a one-unit change in our study sample. Z-scores therefore measure changes in preterm birth as a function of a standard deviation of ICE within our specific sample.

Analysis

We grouped several years of birth data, by county, into three discrete epochs to ensure stable rates of preterm by county-epoch. Each of these epochs (1998-2002, 2006-2010, and 2014-2018) overlap closely with the timing of census-based estimates of ICE race/income and provide a sufficiently large denominator of race-specific births (n>100 per county) to permit stable estimates of preterm. We further restricted our analysis to urban counties as identified by the Rural-Urban Continuum Codes from the US Department of Agriculture, given that Black segregation is exceedingly high in large, urban areas with concentrated poverty [20,34]. This process yielded 1,160 counties, which comprise an estimated 83% of total US births during the study period.

We used a fixed effects Poisson regression to quantify the relation between changes in ICE (from epoch to epoch) and changes in the counts of preterm overall and by race/ethnicity. When using fixed effects, Poisson regression remains preferred over negative binomial models regardless of overdispersion [35]. Importantly, unlike earlier work, the county serves as the “fixed effect”, such that the mean level of preterm birth for each county is removed to measure within county changes over time.

The race-specific count of preterm in each county-epoch serves as the dependent variable. Our models include a population offset (number of race-specific live births), which allows us to interpret the outcome as county-level rates. We then include our z-score measure of ICE by county-epoch, which is the independent variable of interest. In this specification, we approximate the change in the incidence rate of preterm as a function of a standard deviation change in ICE over time, net of time-invariant county-level differences. In addition, we include binary indicator variables for each of the three epochs to control for time-dependent factors that affect the risk of preterm or ICE across the US. We also adjust for the percent of the population below the federal poverty line in each county to differentiate between concentration of racial/economic segregation and poverty. Lastly, we adjust for mean maternal age within counties to account for differing maternal age composition across counties as birthing persons of older ages show a greater likelihood of preterm birth [36].

As a sensitivity check, we then re-ran all analyses but categorized ICE into quintiles to permit comparability to previous studies and assess whether counties that change to higher or lower quintiles of privilege show changes in preterm birth [7,8,10]. We utilized clustered, robust standard errors to account for heteroskedasticity and correlations within clusters (e.g., county).

RESULTS

Table 1 provides descriptive characteristics of NH Black and NH white preterm births and ICE across 1,160 counties included in analyses. Of the urban counties with more than 100 live births per epoch, NH Black preterm birth prevalence exceeds that of NH whites (16.42 vs. 11.05 per 100 live births). These estimates in addition to overall preterm births (11.97 per 100 live births) cohere with national trends, although slightly higher given our focus on urban counties [30][37] (Table 1). ICE averages 0.12 across all epoch-counties with a standard deviation of 0.11 in our study sample. ICE values for counties in our sample range from −0.33 to 0.92 (Table 1).

Table 1.

Preterm birth, Dynamic Index of the Concentration of the Extremes (ICE) race/income, sociodemographic, and maternal characteristics in 1,160 urban counties in the United States with more than 100 NH Black and NH white live births in 1998-2002, 2006-2010, and 2014-2018.

Characteristics	Mean (SD)
Index of the Concentration of the Extremes (ICE)	0.12 (0.11)
Q1	−0.13 (0.06)
Q2	0.07 (0.01)
Q3	0.10 (0.01)
Q4	0.15 (0.01)
Q5	0.27 (0.15)
All preterm births (per 100 live births)	11.97 (2.38)
Preterm births among NH Black (per 100 live births)a	16.42 (3.27)
Preterm births among NH white (per 100 live births)a	11.05 (1.88)
Below the federal poverty line (%)	12.85 (5.29)
Maternal age (mean years)
Overall	26.74 (1.76)
NH Blacka	25.86 (1.76)
NH whitea	27.72 (1.73)

^aLive births specific to race/ethnicity

In Figure 1, overall, NH Black and NH white preterm births increase during epoch 2 (2006-2010) and then decline in epoch 3 (2014-2018). The choropleth map of ICE by quintile (Figure 2) shows variation across place. The southeastern US, as well as some urban centers (e.g., Chicago, Houston, Dallas) and regions of California show persistently high concentrations of NH Black poverty (i.e., lowest quintile of ICE). Overall, mean ICE increases slightly over time (i.e., less concentrated NH Black poverty), as does its variability (Figure 3).

Figure 1.

Preterm births (per 100 live births) among all, non-Hispanic Black, and non-Hispanic white mothers from 1,160 urban counties in the United States withmore than 100 live births^a in 1998-2002, 2006-2010, and 2014-2018.

^aLive births specific to race/ethnicity

Figure 2.

Quintiles of the Index of the Concentration of the Extremes (ICE) race/income for 1,160 urban counties in the United States with more than 100 live births in 2000 (A), 2006-2010 (B), and 2014-2018 (C).

A. 2000

B. 2006-2010

C. 2014-2018

Figure 3.

Mean, standard deviation, and range (blue barbell) of the Index of the Concentration of the Extremes (ICE) race/income for 1,160 urban counties in the United States with more than 100 live births in 1998-2002, 2006-2010, and 2014-2018.

Fixed effects Poisson regression results (Table 2) show that a standard deviation increase in ICE (0.11) corresponds with a 0.6% decrease in risk of preterm birth overall (Incidence Rate Ratio [IRR]: 0.994, 95% Confidence Interval [CI]: 0.990, 0.998). We find a 0.6% decrease in preterm incidence among NH Black and a 0.4% decrease in NH white preterm births as ICE increases (moving away from concentrated NH Black poverty) (NH Black - IRR: 0.994, 95% CI: 0.989 – 0.999; NH white – IRR: 0.996, 95% CI: 0.992 – 0.999).

Table 2.

Fixed effects Poisson regression results predicting change in preterm births (with a population offset) as a function of change in z-score of ICE race/income (standard deviation) across urban counties with greater than 100 live births^a in the US, between 1998-2002, 2006-2010, and 2014-2018.

	All preterm births	NH Black preterm births	NH white preterm births
Covariates	IRR (95% CI)	IRR (95% CI)	IRR (95% CI)
ICE race/income	0.994** (0.990 – 0.998)	0.994** (0.0.989 – 0.999)	0.996** (0.992 – 0.999)
Percent poverty	1.008** (1.004 – 1.011)	1.004* (0.999 – 1.009)	1.004** (1.001 – 1.006)
Mean maternal age	0.993 (0.984 – 1.002)	0.993 (0.973 – 1.014)	0.999 (0.983 – 1.017)
N (# of counties)	3,387 (1,141)	2,300 (783)	3,460 (1,154)

^*p<0.1;

^**p<0.05;

^***p<0.001

^aLive births specific to race/ethnicity

Epoch and county indicators included but not shown

Sensitivity results (Table A1), in which we converted ICE to quintiles and used the highest quintile of ICE (i.e., Q5) as the referent exposure, yielded similar inference to the original test. In Table A1, the relation between lower ICE values and greater risk of preterm concentrates in the 1st ICE quintile for both NH Black and NH white births, as well as overall preterm births (Table A1). Counties in Q1 (more deprived) see a 2.6% increase in risk of preterm per county-epoch relative to a prior Q5 ICE measure (more privileged) in that same county-epoch (IRR: 1.026 CI: 0.995 – 1.058). The magnitude of the result is slightly larger for NH Black births when compared to NH white births (NH Black – IRR: 1.030, 95% CI: 1.003, 1.058; NH white – IRR: 1.025, CI: 0.997, 1.953). Table A2 shows that, whereas most counties undergo modest ICE changes over time, almost all transitional possibilities (i.e., for combinations of Q1 to Q5, Q2 to Q5, Q3 to Q5, etc.) occur in our dataset. In Table A3, we show that our results remain robust to very preterm births (GA <28 weeks) among all, NH white, and NH Black mothers. We also find no relation between ICE race/income and Hispanic preterm births (Table A4).

To estimate preterm births statistically averted by a one standard deviation increase in ICE race/income, we applied the 0.6% reduction to the number of NH Black and NH White preterm births in the study. An estimated 1,285,085 preterm births occurred during epoch 3 (2014-2018) in our 1,141 test counties. Application of the ICE race/income coefficient to this value indicates an estimated 7,710 preterm births statistically averted during a five-year period if ICE race/income rose by one standard deviation in all study counties. Despite the potentially minimal clinical relevance, the population relevance remains substantial given preterm birth’s association with infant mortality, development, and educational attainment [2].

DISCUSSION

Extensive work supports the use of Index of Concentration at the Extremes (ICE) race/income as an important indicator of structural racism [5–7,12,16]. That literature, however, remains cross-sectional and uses populations with unknown external validity. We used data on a 21-year period in the US to quantify the extent to which movements of ICE away from concentrated NH Black poverty precede reductions in the risk of preterm among NH Black births. Results suggest that a standard deviation increase in ICE over time corresponds to a 0.6% reduction in NH Black preterm birth. Although not expected positive changes in county-level ICE also vary with a reduced risk of preterm among NH white births.

These findings cohere with cross-sectional and individual-level studies identifying higher risks of adverse pregnancy and birth outcomes in areas with lower ICE scores [6–8,14]. Mothers living in racially segregated and relatively low-income communities are often exposed to chronic and acute stressors that physically manifest in ways that harm maternal health [16] and increase risk of early deliveries [25,28]. Similar processes likely explain excess counts of preterm births at the county level, such that increases in ICE may indicate fewer exposures to stressors and reduced rates of adverse outcomes. Still, our findings indicate relatively small effects of ICE changes compared to cross-sectional studies [6,8,14]. It is possible that while ICE increases suggest reductions in relative deprivation, the changing racial-income structure may also indicate disruptions to neighborhoods. For example, clustering is a component of racial segregation that identifies the geographic proximity of Black neighborhoods and may reflect political empowerment, social support, and cohesion [28]. Prior work suggests that higher levels of clustering are associated with higher birth weights and reduced risk of preterm birth [28]. Thus, if county ICE is changing because of or alongside significant neighborhood changes, then the protective birth effects of higher ICE scores may be muted by disruptions to political power, social support, and cohesion. Our results suggest that any benefits conferred from reductions in ICE-based structural racism extend to both NH Black and NH white births, joining related work in demonstrating how macro-level structural inequalities may harm both oppressed and privileged groups [38]. Taken together, although increases in ICE values over time precede county-level improvements in perinatal health, they do not appear sufficiently large to substantially reduce rates of preterm birth or related racial/ethnic disparities.

The fact that racial and economic segregation changes over time means that other research examining historical processes that gauge place-based structural racism may consider including repeated measures for geographical units of interest. Future studies may also wish to look at unique employment, residential, or community characteristics that initiate increases in ICE values. Higher ICE or less concentrated NH Black poverty, could, for example, indicate higher incomes and improved living conditions for Black populations residing in a specific area or, alternatively, gentrification where NH Black individuals reluctantly or unwillingly relocate [39]. Whereas we do not conduct a case study for specific counties that undergo increases in ICE over time, we encourage future research in this area as it may reveal the mechanisms that specifically enhance birth outcomes in minoritized populations.

Strengths of our approach include the use of the universe of birth data in the US over a 21-year period for all counties with a sufficient number of births to permit estimation of race-specific risk of preterm birth. This approach supports population-based generalizability of results. In addition, county fixed-effects methods focus on changes within a county over time in structural racism, which enhances internal validity relative to cross-sectional work that compares distinct counties and assumes exchangeability despite much epidemiologic and sociologic evidence that indicates otherwise. The within-county approach, moreover, holds relevance to place-based policy interventions aimed to improve community health in specific places over time.

Limitations of our study include reliance on NCHS birth records, which do not distinguish between spontaneous vs. induced preterm births, and the lack of annual ICE measures between 1998-2018 for the entire US given the availability of data from the US Census outside of the three epochs we utilized. Neighborhood-level changes in racial composition and economic opportunity such as gentrification or large-scale layoffs may also coincide with changes in preterm birth among NH Black populations across the US. Changes in ICE may also co-occur with changes in environmental quality, such as levels of air pollution or greenspace that may be related to preterm birth [26]. Additionally, Krieger and colleagues report that local, neighborhood measures of ICE avoid underestimating the extent to which segregation affects health when compared to city-level measures [11]. Although our analyses focus on counties, more localized research using neighborhood-level data appears warranted over time for the entire US. Finally, we do not include rural counties due to low birth counts of NH Black persons, so our results may not generalize to rural populations.

Whereas our results reach conventional levels of statistical detection and show stronger associations among NH Black (vs. NH white) births, we cannot rule out the possibility of a chance finding. Next, our county fixed effects study design, while precluding confounding by time-invariant factors that are unique to specific counties, remains open to bias by time-varying factors that correlate with, but are not caused by, ICE race/income. For instance, gentrification (i.e., in-migration of wealthier NH white persons, along with outmigration of lower-income NH Black persons) could alter the composition of the community and affect risk of preterm birth [40]. Such mobility processes, and their relation to ICE race/income, warrant more careful scrutiny and would benefit from detailed data on across-county migration patterns, by race/ethnicity, over time from the American Community Survey [41].

Measurement of structural racism in epidemiology is relatively new [5]. While many researchers continue to create new measures of racism, the field has not reached a consensus on how best to capture the complexity of this phenomenon. However, scholars agree on the need for theoretically informed measurement and modelling. Our work seeks to build upon this literature by directly modeling the dynamic nature of structural racism that has been elucidated by multiple critical race theorists [42].

Whereas scholars have used ICE as a proxy for structural racism, we do not aim to imply that it encapsulates all forms and dimensions of structural racism and its analogues (e.g., cultural racism, structural sexism). Our analysis may therefore underestimate the relation between racism and risk of preterm birth. However, our work suggests the need for longitudinal research to better capture the changing nature of place-based racism, as well as racism in other social systems. Given that changes in ICE modestly reduced racial/ethnic disparities in preterm birth, future work must also consider the role of other place-based factors that serve to reinforce racial inequities over time. Equitable access to healthenhancing resources within communities such as health care, employment, and housing may further contribute towards racial health equity in birth outcomes.

KEY MESSAGES.

What is already known on this topic –

Cross-sectional literature finds that greater place-based structural racism (ICE) correlates with adverse birth outcomes, especially among NH Black populations. Previous work evaluates this relation within neighborhoods in individual cities or states after 2010.

What this study adds –

We use data from 1998–2018 and examine longitudinal relationships between declines in place-based structural racism and risk of preterm birth among NH Black populations across more than 1,000 counties in the US. These counties comprise ~83% of total US births.

How this study might affect research, practice, or policy –

Our study strengthens internal validity of prior correlational work in that reduced racial and economic segregation, as well as movement away from concentrated NH Black poverty, may benefit birth outcomes over time.

Data Availability:

Aggregated datasets and statistical code are available from the corresponding author upon request.