Structural barriers to health care as risk factors for preterm and small-for-gestational-age birth among US-born Black and White mothers

Abstract

We develop county-level measures of structural and institutional barriers to care, and test associations between these barriers and birth outcomes for US-born Black and White mothers using national birth records for 2014–2017. Results indicate elevated odds of greater preterm birth severity for Black mothers in counties with higher uninsurance rates among Black adults, fewer Black physicians per Black residents, and fewer publicly-funded contraceptive services. Most structural barriers were not associated with small-for-gestational-age birth, and barriers defined for Black residents were not associated with birth outcomes for White mothers, with the exception of Black uninsurance rate. Structural determinants of care may influence preterm birth risk for Black Americans.

Black mothers in the US have an elevated risk of preterm birth (PTB) and small-for-gestational-age birth (SGA) relative to White mothers (Martin et al., 2018). Structural racism represents a fundamental cause of Black-White disparities in birth outcomes, as racism is embedded in durable social and institutional structures that maintain an unequal distribution of diverse resources and harms based on the prevailing racial hierarchy—with far-reaching health impacts (Hailu et al., 2022; Wallace et al., 2017). For example, Black Americans disproportionately experience inferior access to quality health care across the lifespan relative to Whites (Lu et al., 2010; Smedley et al., 2003). Disparities in care result partly from racial differences in financial resources, insurance coverage, residence in medically underserved areas, and likelihood of experiencing ineffective patient-physician relationships (Greenwood et al., 2020; McMorrow et al., 2015). These examples of barriers to care are historically rooted and maintained by the social structure in the form of intergenerational wealth disparities, employment discrimination, residential segregation, and prevalent racial bias (Gaskin et al., 2012; Pager and Shepherd, 2008).

Prior research has documented harms from interpersonal racism in reproductive care and the segregation of hospitals (Almond et al., 2006; Howell et al., 2018), as well as racialized institutional and interpersonal influences on quality of care (Hall et al., 2015; Howell and Zeitlin, 2017; Smedley et al., 2003). Less understood is whether structural racism in health care contributes to racial disparities in PTB and SGA, as few studies have examined associations between structural and institutional determinants of care and birth outcomes among Black mothers (Brown et al., 2019; Greenwood et al., 2020). One challenge is an absence of established measures of structural barriers in healthcare systems and area-level health care resources (Groos et al., 2018; Howell and Zeitlin, 2017). This stands in contrast to measures of residential segregation and socioeconomic inequities where data are readily available, and numerous studies have examined their influence on racial disparities in birth outcomes (Alson et al., 2021; Ncube et al., 2016; Wallace et al., 2017). Examples of area-level measures of structural and institutional barriers to health care include: (low) insurance rates; residence in underserved areas; absence of providers sharing racial, ethnic, or cultural backgrounds; scarce public health services; and anti-abortion legislation. Comparison of racial differences in exposure to such barriers can quantify structural racism in healthcare and its link with disparities in reproductive outcomes. Hereafter, we use the term structural barriers to care when referring to structural and institutional factors that may prevent or restrict access to health care. This study seeks to establish measures of structural barriers to care and test associations between barriers to care and PTB and SGA for Black mothers.

Background

To ensure optimal reproductive outcomes, access to health care is needed across the life course, especially during reproductive years (Lu et al., 2010). Yet, structural barriers may disproportionately prevent Black relative to White Americans from obtaining quality care. Below, we summarize evidence from structural interventions in the US and observational research indicating plausible barriers to care that may influence risk of adverse birth outcomes for Black mothers.

Health insurance coverage is critical to accessing health care in the US, but pregnancy-limited coverage is likely insufficient to prevent prematurity and fetal growth restriction. For instance, when Medicaid income eligibility thresholds were increased in the late 1980s-1990 to include more pregnant mothers, insurance coverage during pregnancy and prenatal care utilization increased for Black mothers but rates of PTB and low birth weight (LBW) were not affected (Howell, 2001; Krans and Davis, 2012). In contrast, state Medicaid expansions in 2014–2016 through the ACA (Patient Protection and Affordable Care Act) may have reduced Black-White disparities in PTB and LBW (Brown et al., 2019). Recent Medicaid expansion was distinct in that eligibility increased independent of pregnancy status, such that low-income women of reproductive age gained insurance coverage while racial disparities in insurance decreased (Johnston et al., 2019); use of preconception services and postpartum contraception also increased (Myerson et al., 2020). Moreover, research indicates that uninsurance prior to pregnancy may lead to delayed initiation of prenatal care. In one study of Medicaid-covered deliveries, lack of pre-pregnancy Medicaid coverage was associated with 170% higher odds of delayed prenatal care initiation (Rosenberg et al., 2007). Thus, lack of insurance preconception and postpartum likely increases future pregnancy risks even when pregnancy-limited insurance programs exist, such that insurance rates among the reproductive-aged population may more meaningfully capture health care access for the population at risk of pregnancy compared to insurance rates among pregnant individuals.

The availability of a comprehensive range of contraceptive services is a key component of equitable reproductive health care. Moderately/highly effective contraceptives enable control of pregnancy timing and spacing, such that expanding access to contraceptives may improve birth outcomes by reducing unintended and rapid repeat pregnancies (Hogue et al., 2011; Shah et al., 2011). In recent decades, 35–45% of births in the US were from unintended pregnancies, with higher rates for Black relative to White women (Hayford and Guzzo, 2016). The unequal rate of unintended pregnancies is influenced by disparities in uninsurance which limits access to contraceptives (Dehlendorf et al., 2014; Geiger et al., 2021). In the US, uninsured individuals can obtain publicly funded contraceptive services at community health centers that receive funding through the federal Title X program (Frost et al., 2017), such that the availability of publicly funded contraceptives is an important measure of reproductive care access.

Racial disparities in care are reflected not only in unequal insurance coverage and access to services, but in the quality of care received. Across a range of settings and health conditions, Black patients receive lower quality care relative to Whites (Smedley et al., 2003). Black women commonly report feeling disrespected by providers and stressed by their prenatal care, and describe racial stereotypes and racism as undermining influences on their relationships with providers (McLemore et al., 2018). Anti-Black bias among physicians, more common among non-Black physicians, adversely affects physician-patient communication as well as the experiences of Black patients (Hall et al., 2015; Penner et al., 2010). A relevant avenue for improving preconception and prenatal care for Black women is thus to increase the supply of Black physicians and health professionals. Prior literature highlights patient-physician communication as a contributor to worse outcomes for Black patients with a White physician, who receive less health information, are less involved in joint decision-making, and are more dissatisfied with care relative to Black patients with Black physicians (Shen et al., 2018). For prenatal care, where physician-patient interactions occur regularly and health education is a key component, physician-patient communication is crucial. Increasing the supply of Black physicians, who also are more likely to treat low-income patients relative to White physicians, could increase rates of patient-physician racial concordance while improving availability of care for Medicaid patients (Smedley et al., 2003).

Another example of a structural barrier is whether care is spatially accessible and convenient (Charreire and Combier, 2009). Predominantly Black zip codes are more likely to have a shortage of primary care providers, especially in segregated areas (Gaskin et al., 2012). However, evidence for differential spatial access to clinics and hospitals influencing disparities in pregnancy outcomes is weak (White et al., 2012; Yin, 2018). In particular, although proximity to providers is associated with higher use of preventive services (White et al., 2012), this pattern for prenatal care utilization has not been established. One study reported that greater spatial accessibility predicted lower prenatal care utilization (Yin, 2018), while another found residence in a census tract with a prenatal care provider was not associated with SGA (Heck et al., 2002). Spatial proximity to hospitals also was examined as an explanation for the racially segregated care of very LBW neonates, and White mothers in New York City were more likely to reside in neighborhoods with top-tier hospitals than Black mothers but this pattern was reversed for proximity to nearest top-tier hospital (Hebert et al., 2011). Thus, although Black Americans are more likely to reside in primary care shortage areas, whether this factor influences pregnancy outcomes is unknown.

Current Study

This study 1) establishes new county-level measures of structural barriers to health care and documents racial inequities in such barriers; and 2) tests whether barriers to care are associated with elevated PTB and SGA odds among Black mothers. Measures of county structural barriers to care include: uninsurance rates among reproductive-aged Black adults; low supply of Black physicians; share of Black adults residing in primary care shortage areas; insufficiency of publicly funded contraceptives; and few public health expenditures. These measures capture diverse determinants of access to care and may influence quality of care, but are not exhaustive and lack patient experiences. This study addresses gaps in prior literature by documenting specific ways structural racism manifests within healthcare and whether barriers to care are related to risk of adverse birth outcomes.

Using natality records from 2014–2017 for all counties meeting study criteria, the sample includes singleton births to US-born non-Hispanic White and Black mothers. We tested the following hypotheses: 1) Black women experience higher rates of county-level structural barriers to health care relative to White women, with the exception of public health expenditures and contraceptive service availability as the included measures are not race-specific; and 2) structural barriers to care are associated with higher odds of PTB severity and SGA for Black mothers.

Methods

Data

The sample of live births came from restricted access national birth records for years 2014 through 2017 (National Center for Health Statistics, n.d.). Inclusion criteria were implemented to refine the sample to: 1) singleton births; 2) to US-born mothers identifying as non-Hispanic Black (Black) or non-Hispanic White (White); 3) who resided in a county with a population of at least 20,000 and a metropolitan statistical area (MSA) with at least a total population of 100,000 and Black population of 10,000. Area-based inclusion criteria were implemented to ensure adequate sample sizes for valid measurement of structural barriers and because the meaning of structural barriers and their influence on health outcomes might operate differently in rural areas or areas with a small Black population. More specifically, we included a minimum population threshold for counties and minimum Black population in MSAs to reduce error in measuring structural barriers and prevent sample size-related outliers (e.g., the supply of same-race physicians per capita could appear large in counties with a single Black physician if the Black population was small). Moreover, we included only MSAs and applied minimum population thresholds at this level, in line with prior research (Kramer et al., 2010), as we expected segregation patterns and the placement of health clinics to operate within regional housing markets. Yet, even when applying area inclusion criteria, we retained the large majority of births to US-born non-Hispanic Black mothers, 88.8% of total births during the study years (i.e., 1,584,411 of a total 1,784,302 births). The sample of births to US-born non-Hispanic White mothers in analytic counties includes a relatively smaller share of total births in this population (68.8% of a total 7,338,369). Preterm birth rates for Black mothers were comparable but slightly lower in analytic counties (12.1%) relative to excluded counties (12.5%; t = 4.20, p <.001). These differences warrant further research but are beyond the scope of this study.

When modeling PTB severity, we included all births with estimated gestation from 22 to 42 weeks. This is consistent with commonly used lower cutpoints for extreme preterm birth (i.e., 22 weeks) and excludes post-term births that have higher clinical risks (Moutquin, 2003). When modeling SGA, we included births with estimated gestation from 23 to 41 weeks and available birth weight measurements due to our use of established cutpoints for birth weight at given gestational weeks (Olsen et al., 2010). These criteria resulted in loss of only 0.34% and 0.63% of Black births for PTB and SGA, respectively. Births were also excluded if missing area-level measures, resulting in further loss of 2.22% of total births in the eligible areas (n = 35,067); these births occurred in 69 counties that had missing area data relative to the total of 785 counties meeting area inclusion criteria. Thus, the analytic sample for PTB severity included 1,543,961 births to US-born non-Hispanic Black mothers in 716 US counties, covering 86.5% of the total births among this population. The corresponding percentage of total births to US-born non-Hispanic White mothers when modeling PTB was 66.4%, with a sample of 4,869,255.

Data for county-level measures came from multiple sources, described below.

Measures

Measures are summarized in Table S1 of Supplemental Material.

Adverse birth outcomes

Gestational age at birth, recorded on birth certificates, was based on the best obstetric estimate (Martin et al., 2015). We coded PTB severity using the following gestational age categories: term (37 to 42 weeks), mild (34–36 weeks), moderate (32–33 weeks), very (28–31 weeks), and extreme (22–27 weeks) (Moutquin, 2003).

Birth weight was directly measured and listed on nearly all records (Northam and Knapp, 2006). SGA was coded using sex-specific cut points to represent neonates in the lowest 10^th percentile of birth weight for gestational age (Olsen et al., 2010).

Structural and institutional barriers to health care

To assess insurance coverage among the reproductive-aged population, we used the county-level rate of Black Americans between 18 and 44 years without insurance coverage. This estimate came from the average of American Community Survey 2013–2017 single-year estimates; the use of single-year estimates was due to a five-year estimate being unavailable (US Census Bureau, n.d.). Comparable estimates of uninsurance for White Americans were prepared. Missing data for county uninsurance rates were common, with only 433 of the 716 counties having available data. We used two strategies to handle missingness. First, we imputed missing values using state-level estimates. Second, we used state estimates when observations were either missing or had high sampling error based on coefficients of variation exceeding 0.12 (Folch et al., 2016). Although we initially planned to use the second approach, high sampling error was common as only 70 counties had a coefficient of variation less than 0.12—yet the majority of births in the analytic sample were in these counties (i.e., 56.9% of births to Black mothers). Thus, we used both approaches, as county measures were conceptually favored because the smaller scale may more accurately capture exposures and resources, albeit with sampling error, whereas state measures likely have less error.

To proxy for spatial access to primary care, we estimated the percent of adult Black residents who lived in a primary care shortage area within counties. Practice locations for nearly all health care providers in the US came from the Centers for Medicare and Medicaid Services NPI (National Provider Identifier) registry (National Plan & Provider Enumeration System, n.d.). Restricting to individual primary care providers with specialty in family practice, general practice, internal medicine, pediatrics, adolescent medicine, or obstetrics and gynecology, we computed the number of physicians practicing in each ZIP code. Next, similar to prior research (Gaskin et al., 2012), matching ZIP codes to ZCTAs, we calculated the ratio of persons to physicians and coded primary care shortage areas as ZCTAs with >3500 persons: 1 primary care physician. Finally, using the Census 2010 ZCTA to County Relationship File, we aggregated data to calculate the county-level share of Black residents in a primary care shortage area, and created comparable estimates for White residents. Although Health Professional Shortage Areas are identified by the US Health Resources and Services Administration, we elected to derive our own measure. This decision was motivated by inconsistencies in the geographic scale of federally-defined shortage areas (i.e., reported as whole counties in many cases and tracts within counties otherwise) that would not allow use of population counts by race in sub-county geographies to estimate race-specific rates of residence in primary care shortage areas.

The number of county direct patient care Black and White physicians in 2014 came from the Association of American Medical Colleges (Association of American Medical Colleges, 2017). To estimate relative supply of Black physicians, we computed the ratio of 1000 Black county residents per Black direct care physicians, and a comparable estimate for White residents per White physicians.

To assess the sufficiency of available publicly funded contraceptive clinics, we used data from the Guttmacher Institute (Frost et al., 2019, 2017). Based on the estimated number of women aged 13–44 who likely needed public support for contraceptive services in 2016, we computed the share of this need met based on the number of female contraceptive clients served at publicly funded clinics in 2015. Described in detail elsewhere (Frost et al., 2019), estimates of the number of women likely needing public support for contraceptives incorporated population estimates for different groups and their likely need for contraceptives based on sexual activity, fecundity, and pregnancy status. Need for publicly funded contraceptives was based on income level (i.e., under 250% of federal poverty level) for women aged 20 or older, and included all women under 20-years-old with a likely need for contraceptives as youth have higher confidentiality needs and may seek services separate from family resources or insurance. Number of contraceptive clients came from extensive primary data collection efforts with publicly funded clinics and captured unduplicated female patients receiving contraceptive services or supplies (Frost et al., 2017).

Counties with higher public health expenditures, especially for maternal and child health, have lower rates of adverse birth outcomes and smaller Black-White disparities (Bekemeier et al., 2014; Curtis et al., 2019). We used local government health expenditures per capita, assessed for county and sub-county governments via the quinquennial US Census of Governments (Pierson et al., 2015). Health expenditures included costs for public health administration, community health care, health education, mental health services, and regulation of air/water, while excluding hospital expenditures (US Census Bureau, n.d.). We restricted expenditures to operational costs (excluding capital outlay), which covered direct employee compensations, costs for supplies and materials, and contractual services; intergovernmental transfers to other local governments were excluded to avoid double counting. Health expenditures per capita, adjusted for cost-of-living using the Regional Price Parity index, were summed across county and sub-county general and special purpose governments and averaged across years of 2012 and 2017.

Covariates

Birth-level covariates included maternal age, parity (1, 2, 3–5, 6+), and infant sex. Although other sociodemographic and health risks were available in natality records, such as education and partner involvement, these variables may mediate the effects of structural barriers and were thus omitted from our initial models. For instance, high rates of uninsurance and weak public health systems, including funding for contraceptive services, could differentially influence fertility rates across sociodemographic groups, with socioeconomically disadvantaged individuals most affected by barriers to contraceptives, while this group also has elevated PTB risk (Lindrooth and McCullough, 2007; Myerson et al., 2020). In sensitivity tests, we included adjustment for maternal education (categories of no high school diploma; high school diploma or GED; some college, no Bachelor’s degree; Bachelor’s degree or higher; unknown) and partner involvement (married; nonmarried, paternity acknowledged; nonmarried, paternity unacknowledged) (Curtis et al., 2022).

For area-level covariates, we included county NCHS urban-rural classifications due to expected differences in health care resources based on the size and density of metropolitan areas (Caldwell et al., 2016). Percent of total population identifying as Black and Black-White residential segregation were included as the size and spatial distribution of the Black population may bias measures of structural barriers to health care. Residential segregation was assessed using the spatial exposure/isolation index (Reardon and O’Sullivan, 2004). As potential socioeconomic confounders, we included county poverty rate among Black persons aged 15 to 44-years-old and county median household income for the county population (a general measure of economic circumstances). Socioeconomic covariates came from the American Community Survey 2013–2017 five-year estimate. Finally, as measures of overall county health care resources, we included per capita primary care physicians and hospital beds for 2015 from the Area Health Resources File (U.S. Health Resources and Service Administration, n.d.). These are conceptually distinct from race-specific measures of supply of same race physicians and share of Black residents in primary care shortage areas.

Analysis Plan

We used two-level models to account for clustering of births in counties, with ordered and binary logistic regression for PTB severity and SGA, respectively (Hedeker, 2015). Ordered logistic regression allows for modeling PTB severity and produces more robust estimates relative to binary logistic models (Sankey and Weissfeld, 1998). Shown as a cumulative logit model, the equation for PTB severity is: $\text{ln}\left(\frac{\pi {(c)}_{ij}}{1-\pi {(c)}_{ij}}\right)=\gamma (c)-{\mu }_j$, where $\text{ln}\left(\frac{\pi {(c)}_{ij}}{1-\pi {(c)}_{ij}}\right)$ refers to cumulative log-odds of PTB severity categories c for birth i in county j, γ(c) are c - 1 increasing thresholds when random effects are 0, and μ_j is the random effect for county j; μ_j is assumed to follow a normal distribution: $N(0,{\sigma }_{\mu }^2)$. The model was similar in form for SGA, yet logistic regression was used for this binary outcome given lack of clear ordinal cut-points for SGA. The meologit and melogit commands were used in Stata 15.1.

In hypothesis tests, we separately modeled the association between each county-level barrier and adverse birth outcomes. These models included birth- and county-level covariates. Next, health care barriers were jointly modeled to determine their independent effects, retaining all covariates. We reported the county-level variance in PTB severity explained by health care barriers as a measure of the effect size. We evaluated the proportional odds assumption for ordinal models where the relationship between predictors and PTB is assumed to be the same across severity categories; in particular, we used the Brant test (Brant, 1990) and examined differences in estimates when fitting binary logistic regressions with each PTB cut-point (Hedeker, 2015).

Additionally, we modeled B-W disparities in PTB severity and SGA using the sample of births to Black and White mothers. A term representing Black race (1=Black, 0=White) was included in these models, and a random race effect estimated across counties. County variance in the race term represents the county-specific deviation from the average county-level racial disparity in the outcome. After first estimating the racial disparity in PTB severity, race-specific measures of barriers to health care were added to capture average exposure per racial group in the county (i.e., Black and White mothers have unique values for each county-level barrier). We reported the difference in coefficients for the race term across models and the proportion of variance explained for the random effect of race and total county-level variance when adding the structural barriers to care. Covariates were similar in these models, but we included the race-specific county poverty rate instead of the Black poverty rate.

Finally, to test whether the pattern of associations for structural barriers among Black residents is unique to Black mothers (in contrast to White mothers), we modeled Black-specific structural barriers to health care as predictors of PTB severity and SGA risk for births to White mothers, with an expectation of null associations. The same set of covariates are included in these models as those with births to Black mothers to facilitate comparison of results.

Results

Descriptive statistics are shown in Table 1 and Table 2. Rates of PTB show large racial disparities, especially for more severe categories—in particular, mild PTB is 50% more common for Black relative to White mothers, moderate PTB 100% more common, very PTB 149% more common, and extreme PTB 277% more common. The rate of SGA among Black mothers is 129% higher than that among Whites mothers. Reported in Table 2, moderate-to-large racial inequities exist for race-specific measures of health care barriers. Among sample births, Black relative to White mothers had fewer same-race physicians, higher rates of residence in primary care shortage area, and higher rates of uninsurance. These measurements come from race-specific populations of county residents (e.g., reproductive-aged adults) rather than the population of pregnant individuals. In contrast, racial differences are very small for the county-wide measures of sufficiency of contraceptives and public health expenditures.

Table 1.

Descriptive statistics for births to Black and White mothers, National Vital Statistics System Natality Records, 2014–2017.

	Black births (n=1,543,961) Mean ±SD (%)	White births (n=4,869,255) Mean ±SD (%)
Maternal age	26.61 ±5.81	29.43 ±5.45
Parity	1.28 ±1.47	0.98 ±1.20
Infant sex (female)	(49.21)	(48.69)
Preterm birth severity
Mild	(7.96)	(5.31)
Moderate	(1.46)	(0.73)
Very	(1.37)	(0.55)
Extreme	(1.13)	(0.30)
SGA birth	(10.24)	(4.48)

SGA=small-for-gestational-age.

^adifferences between Black and White births were significant at p < .05 based on independent samples t-tests for continuous variables and chi square tests for categorical variables.

Table 2.

Descriptive statistics for measures of structural barriers to health care by maternal race, with tests of racial differences in exposure.

County measures	Black births (n=1,543,961) Mean ±SD	White births (n=4,869,255) Mean ±SD	Racial differenceb Cohen’s D
Same-race physicians (per 1000 residents)a	0.66 ±0.33	1.72 ±1.10	1.08
Residents in primary care shortage area (%)a	29.01 ±16.33	23.79 ±13.820	0.36
Uninsurance rate, adults aged 18–44 (%)a	20.07 ±5.78	14.74 ±6.43	0.85
Sufficiency of publicly-funded contraceptives (%)	27.00 ±16.74	26.67 ±17.89	0.02
Public health expenditures ($100s per resident)	1.70 ±2.19	1.49 ±1.92	0.10

^aRace-specific measurement.

^bAll racial differences significant at p < .05 based on independent

samples t-tests.

To investigate relationships between structural barriers and potential for multicollinearity, we examined correlations between county-level structural barriers to care (n = 716 counties), shown in Table S2. Correlations between the three structural barriers defined for Black residents were generally of a weak magnitude but statistically significant (around r = .20), public health expenditures were correlated around .10 or higher with other barriers, and sufficiency of contraceptives showed an inconsistent pattern of at most weak correlations with other barriers. We checked for problems with multicollinearity when jointly modeling structural barriers to care and area covariates, and VIF (variance inflation factor) coefficients for structural barriers were 1.6 or below, indicating multicollinearity problems are unlikely serious.

Sample of Black births.

In the unconditional means model for PTB severity, county-level variance in the intercept was .0146 (SE = .0016). The variance decreased to .0085 in a model that included adjustment for birth- and area-level covariates but omitted structural barriers to care; this variance is used when testing the explanatory power of structural barriers to care. For SGA, county-level variance was equal to .0168 (SE = .0021) in the unconditional means model and decreased to .0100 with adjustment for covariates.

Results for PTB severity among Black mothers using multilevel ordered logistic regression are shown in Table 3. All estimates are adjusted for birth- and area-level covariates. When structural barriers to care are modeled separately, odds of greater PTB severity were elevated for Black mothers in counties with fewer Black physicians per 1000 Black residents, higher uninsurance rates among Black reproductive-aged adults, and lower sufficiency of publicly-funded contraceptive services. To facilitate comparison across structural barriers, we also report ORs when standardizing structural barriers; expressed as 1 standard deviation unit increase for each structural measure, ORs were equal to 0.984; 1.023; and 0.982, respectively.

Table 3.

Multilevel ordered logistic regression results for preterm birth severity among Black mothers (n = 1,543,961 births).

County measures	Separately modeled OR (95% CI)	Jointly modeled OR (95% CI)
Black physicians (per 1000 Black residents)	0.95 (0.92, 0.98)	0.96 (0.93, 1.00)
Black residents in primary care shortage area (10%)	1.01 (1.00, 1.01)	1.00 (1.00, 1.01)
Black uninsurance rate, adults aged 18–44 (10%)	1.04 (1.02, 1.06)	1.03 (1.01, 1.05)
Sufficiency of publicly-funded contraceptives (10%)	0.99 (0.98, 1.00)	0.99 (0.99, 1.00)
Public health expenditures ($100 per resident)	1.00 (0.99, 1.00)	1.00 (0.99, 1.00)

Note. Bold font indicates statistical significance at p < .05. Estimates adjusted for maternal age, parity, infant sex, and the following county variables: urban-rural classifications, Black population share, Black-White spatial exposure/isolation index, poverty rate among Black persons aged 15 to 44, median household income for total population, per capita primary care physicians, and hospital beds.

The second column of estimates in Table 3 shows the independent associations of structural barriers with PTB severity for Black mothers. Estimates for the structural barriers are slightly attenuated when jointly modeled, although Black physicians per Black residents, uninsurance rate among Black reproductive-aged adults, and sufficiency of publicly-funded contraceptive services remained statistically significant predictors; ORs are equivalent to 0.987, 1.019, and 0.987 for a 1 SD unit increase in these respective structural measures. Simultaneous adjustment for structural barriers to health care explained 12% of remaining county variance after previously adjusting for all covariates (reducing the variance from .0093 to .0082). Results from the Brant test (Long and Freese, 2006) signified the proportional odds assumption was not violated for estimates of supply of Black physicians, uninsurance rates, and sufficiency of publicly-funded contraceptives, but was violated for primary care shortage and public health expenditures. Testing different thresholds for binary logit models, results indicated no significant associations for either residents in primary care shortage areas or public health expenditures when using the different PTB cutpoints.

Model estimates for SGA odds among births to Black mothers are shown in Table 4. Estimates were inconsistent with expectations of higher SGA odds in areas with more barriers to care. However, sufficiency of publicly-funded contraceptives was associated with SGA, such that odds increased as more of the need for public contraceptives was met. Simultaneous modeling of structural barriers did not substantively alter this association. Barriers to care explained only 3% of county variance beyond covariate adjustment (reducing variance from .0100 to .0097).

Table 4.

Multilevel logistic regression results for small-for-gestational-age birth among Black mothers (n = 1,539,517 births).

County measures	Separately modeled OR (95% CI)	Jointly modeled OR (95% CI)
Black physicians (per 1000 Black residents)	0.99 (0.95, 1.03)	0.98 (0.94, 1.03)
Black residents in primary care shortage area (10%)	1.00 (0.99, 1.01)	1.00 (1.00, 1.01)
Black uninsurance rate, adults aged 18–44 (10%)	0.99 (0.97, 1.01)	0.99 (0.97, 1.02)
Sufficiency of publicly-funded contraceptives (10%)	1.01 (1.00, 1.02)	1.01 (1.00, 1.02)
Public health expenditures ($100 per resident)	1.00 (1.00, 1.01)	1.00 (0.99, 1.01)

Note. Bold font indicates statistical significance at p < .05. Estimates adjusted for maternal age, parity, infant sex, and the following county variables: urban-rural classifications, Black population share, Black-White spatial exposure/isolation index, poverty rate among Black persons aged 15 to 44, median household income for total population, per capita primary care physicians, and hospital beds.

In sensitivity tests, we added maternal education and partner involvement to models with all structural barriers to care as predictors of PTB severity and SGA. Estimates for structural barriers were similar to models without these potential mediators. In particular, the Black uninsurance rate (OR = 1.04, 95% CI: 1.02, 1.06) and sufficiency of publicly funded contraceptives (OR = 0.99, 95% CI: 0.98, 0.99) were associated with PTB severity, but the number of Black physicians was no longer a significant predictor (OR = 0.97, 95% CI: 0.94, 1.01). Estimates were similar for SGA but sufficiency of publicly funded contraceptives was no longer significant (OR = 1.01, 95% CI: 1.00, 1.01). Next, we examined the Black uninsurance rate when imputing county estimates with high sampling error using state estimates. Results indicated that Black uninsurance was associated with greater PTB severity (OR = 1.06, 95% CI: 1.04, 1.09); estimates for other barriers and inferences were similar, with the exception of sufficiency of publicly funded contraceptives no longer being significant (OR 1.00, 95% CI: 0.99, 1.00).

Sample of Black and White births.

Results from mixed effects models including births to Black and White mothers are shown in Table 5. The first column of results shows fixed and random effects for maternal race when adjusting for birth- and area-level covariates. The second column shows results when modeling structural barriers to care, including the race-specific and county-wide measures; race-specific measures are defined separately for Black and White mothers within counties (e.g., Black [White] residents in primary care shortage areas). All included structural barriers were associated with greater PTB severity in the expected direction (i.e., fewer same-race physicians, more same-race residents in primary care shortage areas, higher same-race uninsurance rate, greater insufficiency of publicly funded contraceptives, and fewer public health expenditures predicting elevated). Adjusting for structural barriers to care reduced the regression coefficient for maternal race by 9.6% relative to the model without the barriers, and reduced the variance for maternal race (i.e., the race random effect) by 19.6% and total county-level variance by 13.4%.

Table 5.

Multilevel ordered logistic regression results for preterm birth severity among Black and White mothers (n = 6,413,216 births).

County measures	OR (95% CI)	OR (95% CI)
Black maternal race (White = reference)	1.76 (1.71, 1.82)	1.69 (1.64, 1.74)
Same-race physicians (per 1000 residents)		0.97 (0.96, 0.98)
Same-race residents in primary care shortage area (10%)		1.02 (1.01, 1.02)
Same-race uninsurance rate, adults aged 18–44 (10%)		1.02 (1.01, 1.04)
Sufficiency of publicly-funded contraceptives (10%)		0.99 (0.98, 1.00)
Public health expenditures ($100 per resident)		0.99 (0.98, 1.00)
Variance components	Variance (SE)	Variance (SE)
Random county effect	0.0172 (0.0015)	0.0149 (0.0015)
Random race effect	0.0149 (0.0024)	0.0120 (0.0023)

Note. Bold font indicates statistical significance at p < .05. Estimates adjusted for maternal age, parity, infant sex, and the following county variables: urban-rural classifications, Black population share, Black-White spatial exposure/isolation index, poverty rate among Black persons aged 15 to 44, median household income for total population, and hospital beds. Per capita primary care physicians is omitted due to conceptual overlap with same-race physicians for births to White mothers given the majority share of White physicians.

Model results for SGA odds using the sample of births to Black and White mothers are shown in Table S3. More same-race physicians per capita is associated with reduced SGA odds (OR = 0.98, 95% CI: 0.97, 0.99) while greater sufficiency of publicly-funded contraceptives is associated with increased SGA odds (OR = 1.01, 95% CI: 1.00, 1.02). Adjusting for structural barriers to care reduced the coefficient for maternal race by 3.3% and variance in racial differences by 8.0%.

Sample of White births.

We tested Black-specific structural barriers to care as predictors of PTB severity and SGA for births to White mothers, with an expectation of null associations; these tests are intended to demonstrate the specificity of same-race-derived measures of structural barriers. Results are shown in Table S4. The only Black-specific barrier associated with PTB severity for White mothers is the Black uninsurance rate. In contrast, both county-wide structural barriers were associated with PTB severity. To probe the unexpected finding for Black uninsurance rate, we included the White uninsurance rate in the model without other structural barriers. Here, the Black uninsurance rate was not associated with PTB severity for White births (OR = 1.02, 95% CI: 1.00, 1.05) and the estimate for White uninsurance rate was similar and nonsignificant (OR = 1.02, 95% CI: 1.00, 1.05).

Discussion

Structural racism is increasingly recognized as a cause of racial disparities in adverse birth outcomes (Alson et al., 2021), yet how these forces operate within the healthcare system is not well understood. Our study examines structural and institutional factors that may reduce access to effective health care, and whether such barriers are associated with greater PTB severity and SGA births among Black mothers. Consistent with our hypothesis, county-level higher uninsurance rates among reproductive-aged Black adults, insufficient publicly-funded contraceptives, and smaller supply of Black physicians per Black residents are associated with increased odds of more severe PTB among Black mothers; these associations are independent of established socio-economic structural determinants and other barriers to care. Although reported associations are of small magnitude, estimates refer to the increased risk of higher PTB severity across the population rather than increased risk for an individual experiencing a structural barrier. Given the large number of births affected, the included structural barriers are associated with a costly and tragic level of excess prematurity among Black infants in the US. Notably, increased risk occurs across severity categories, such that structural barriers likely contribute to elevated rates of very and extreme PTB for Black relative to White mothers. Extreme prematurity is an especially large risk factor for infant mortality and physical and developmental risks (MacDorman, 2011).

Uninsurance prevalence among reproductive-aged adults may influence PTB rates via more frequent unintended pregnancies (Johnston and McMorrow, 2020; Myerson et al., 2020), and through financial- and health-related pathways (Finkelstein et al., 2012; Rosenberg et al., 2007). The link between county publicly-funded contraceptives and PTB severity also is consistent with research showing PTB is more common among unintended pregnancies (Shah et al., 2011). When contraceptives are available at no cost, whether through insurance coverage or government programs, a higher proportion of individuals opt for highly effective contraceptive methods (Darney et al., 2022; Johnston and McMorrow, 2020). In the US, the ACA expanded Medicaid eligibility and mandated that insurers cover contraceptive at no out-of-pocket costs to consumers, leading to increased use of prescription contraceptives for Black women (Johnston and McMorrow, 2020). In sum, mechanisms to improve access to contraceptives through public and private channels (e.g., regulation of private insurers, expand eligibility for public insurance, public funding for contraceptive services) can help address the reproductive needs of individuals from diverse social circumstances and may prevent PTB. We note that greater access to contraceptives may achieve reproductive justice goals most effectively if paired with efforts to maintain patient autonomy and reduce coercion by health care providers.

We found that residing in a county with a smaller supply of Black physicians was positively associated with risk of greater PTB severity among Black mothers, suggesting that quality, regular care may influence birth outcomes. This is consistent with research showing potential health benefits for Black individuals living in areas with more Black physicians (Snyder et al., 2023). Additionally, physician-patient racial concordance has been linked with improved health outcomes, likely due to communication patterns and patient experiences (Shen et al., 2018). Physician-newborn racial concordance also has been linked with reduced mortality among Black newborns, possibly from underperformance by White physicians (Greenwood et al., 2020). Yet, to our knowledge, no studies have documented whether prenatal care delivered by Black physicians helps prevent PTB among Black mothers. Increases in the share of Black physicians in the US since 2009 are nonetheless encouraging and may influence maternal and infant health outcomes (Snyder et al., 2023). Efforts to increase the diversity of the US healthcare workforce are longstanding and have achieved some success (Institute of Medicine, 2001), with the US Health Resources and Services Administration administering multiple programs aimed at achieving this goal (Camacho et al., 2017).

Contrary to our hypothesis, the proportion of Black residents in primary care shortage areas and level of county health expenditures were not associated with PTB severity for Black mothers, although higher health expenditures predicted reduced PTB severity in the sample of births to White and Black mothers. Prior ecological studies provide mixed evidence for an association between county public health expenditures and birth outcomes. In a study of two states that included detailed expenditure types, maternal and child health services appeared to be more predictive of LBW rates than total health expenditures, but estimates were inconsistent across specifications and geographies (Bekemeier et al., 2014). Another study reported associations between changes in health department expenditures and concurrent changes in Black and White infant mortality (Grembowski et al., 2010). The inconsistency of our findings with other studies may be due to the ecological design of prior studies or the time-invariant and general nature of our measure that includes broad types of services.

In line with the concept of structural racism, our findings show racial patterning of structural barriers to care and race-specific associations for PTB severity. Consistent with our hypothesis, we found moderate to large racial disparities in structural barriers to care, with Black participants experiencing a lower supply of same-race physicians, higher proportion of residents in primary care shortage areas, and higher uninsurance rates. Contrary to findings for Black mothers, the supply of Black physicians was not associated with PTB severity for White mothers, indicating some specificity in who experiences this barrier to care and its link with PTB. In contrast, barriers defined using the full county population—insufficiency of publicly-funded contraceptives and fewer public health expenditures—were significantly associated with higher PTB severity for White mothers. These results document specific structural and institutional health care factors that disadvantage Black relative to White Americans. The sociohistorical and upstream causes of structural racism in health care are beyond the scope of the current study, but such causes are likely numerous and complex. For instance, ACA-linked Medicaid expansion is a recent structural intervention to improve access to care (Johnston et al., 2019), but its reach was limited by states who declined to expand, possibly due to racial attitudes among Whites (Grogan and Park, 2017).

Contrary to our hypothesis, none of the structural barriers was associated with higher odds of SGA birth among Black mothers, while less publicly-funded contraceptives predicted reduced SGA odds. These findings suggest access to health care may influence PTB to a greater degree than SGA, thus indicating a different set of mechanisms (e.g., stress-related and inflammatory relative to diet and cigarette smoking) (Kramer et al., 2000). Moreover, prior research suggests that SGA as compared to PTB may be less sensitive to racial discrimination and area-level socioeconomic disadvantage, based on the number of studies reporting a significant relationship (Blumenshine et al., 2010; Daalen et al., 2022). Yet, much of the research on structural racism and fetal growth restriction has relied on LBW, which stems from prematurity and restricted fetal growth, instead of the more specific measure of SGA, such that understanding of structural racism as a determinant of fetal growth is limited.

Strengths of the study include a large sample of live births to Black mothers and consideration of both PTB severity and SGA as outcomes. Moreover, we established novel measures of structural barriers to health care that reflect circumstances among the broader population of Black or White Americans. These include race-specific measures that allow for quantifying structural racism in health care. The hypotheses and study design were preregistered.

We note a few key limitations. First, this is an observational study and thus does not support causal inferences for the reported associations. Second, examination of mediating pathways was beyond the scope of this study, as our intent was to establish measures of structural barriers to care and describe associations with birth outcomes for Black mothers. Third, measures of structural barriers refer only to aggregate experiences or resources among county Black residents and lack participant experiences. If the effects are causal, personal experiences and disadvantages due to structural barriers would likely be associated with even higher risks. Finally, we used county as the unit of analysis due to available data and our preference for a smaller geography than states, allowing us to more closely assess local conditions. Yet, barriers to health care likely exist at multiple geographic levels, and our results may be sensitive to the geography selected—known as the scale problem of the modifiable areal unit problem.

Several study findings warrant further research. For example, the Black uninsurance rate was associated with PTB severity for births to both Black and White mothers, indicating this measure may be correlated with unobserved determinants of insurance coverage rates that increase risk across racial/ethnic groups. Moreover, the health care barriers we included are likely related to one another in complex ways, such that independent estimates for individual barriers on PTB may be conservative. Research is needed to consider cumulative and interactive effects and the mediating pathways. Future quasi-experimental studies are also needed to strengthen the credibility of causal effects.

Conclusion

Structural racism in healthcare refers to the totality of ways that social structures produce inequities in access to effective care, with financial, geographic, social, and political factors contributing (Hailu et al., 2022). Our findings provide only a partial description of how structural racism may manifest in healthcare, but we show that structural factors, such as public funding for contraceptives, low insurance coverage rates, and few Black physicians relative to Black residents, are associated with greater PTB severity for Black mothers. To reduce racial disparities in birth outcomes, structural and institutional interventions are likely needed to increase access to quality health care for Black Americans across the reproductive life course (Lu et al., 2010). Understanding the mechanisms of structural barriers to care, even as distal or fundamental causes remain, has potential to inform actionable policy and institutional responses to attenuate health disparities.

Highlights.

• Few area-level measures of structural barriers to health care are available

• Structural racism is evident in unequal insurance rates and access to care and same-race physicians

• Black mothers in areas with lower insurance rates or fewer Black physicians have elevated preterm birth

• Preterm birth is more common in areas with insufficient publicly funded contraceptives

• Structural barriers to health care likely contribute to racial disparities in preterm birth