A decomposition analysis to evaluate the role of social, chemical, and neighborhood factors on the Black/White disparity in preterm birth in Greater Houston, Texas

Abstract

Background:

Non-Hispanic Black women have almost double the risk of preterm birth compared with non-Hispanic White women in the United States, and the disparity is not fully explained by sociodemographic factors, health behaviors, or medical factors. Thus, it is critical to look beyond these “traditional” risk factors and evaluate contributions from psychosocial stressors, chemical exposures, and features of residential environments where pregnant women live.

Methods:

This study was conducted among 1,847 Black and White women recruited at the time of delivery between 2021 and 2023 in greater Houston, Texas to examine the contribution of individual- and neighborhood-level factors to the Black/White disparity in preterm birth using two-fold Blinder–Oaxaca decomposition. We evaluated the contribution from variables within the following domains: sociodemographics; perinatal history; comorbidities and health behaviors; chemical exposures; social factors; and neighborhood characteristics. We present the results as percent contributions to the disparity for each covariate and by domain.

Results:

With a 6.6% higher prevalence of preterm birth among Black mothers compared with White mothers, variables included in our decomposition analysis explained 63.0% of the preterm birth disparity, with perinatal history explaining the highest proportion (36.3%), followed by neighborhood characteristics (32.8%). Individual variables with the highest contributions included residing in a more racially minoritized and economically disadvantaged neighborhood (19.2%) and having a previous pregnancy complication (18.5%), a hypertensive disorder in the index pregnancy (16.7%), or chronic hypertension before pregnancy (16.8%).

Conclusion:

Our study underscores the importance of perinatal history and neighborhood factors in explaining disparities in preterm birth, which should be further explored in other populations.

What this study adds

Previous studies decomposing the Black/White disparity in preterm birth among women in the United States have largely analyzed contributions from sociodemographic factors and comorbidities and have only explained up to 40% of the disparity. We evaluated the contributions from less explored social, perinatal, and neighborhood factors and chemical exposures (in addition to traditional factors) to the Black/White preterm birth disparity among women in greater Houston, Texas. Our study explained 63.0% of the disparity (considerably more than previous decomposition studies of preterm birth), with perinatal history (36.3%) and neighborhood factors (32.8%) explaining the highest proportion of the disparity, underscoring their importance and the need for study in other populations.

Introduction

Preterm birth, that is, delivering an infant <37 completed weeks of gestation, is a leading cause of infant mortality in the United States (US).¹ Preterm birth also increases the risk of serious comorbidities both during infancy and childhood² as well as through adulthood,^3,4 and imposes a significant economic burden.⁵ In 2022, the rate of preterm birth in the US was 10.4%, with an even higher rate among Black (14.6%) compared with White women (9.4%).^6,7 Some areas in the nation, however, experience an even higher burden of preterm birth, such as in Harris County, Texas, where the prevalence in 2022 was 12.1%.⁸

A number of medical, behavioral, and sociodemographic risk factors for preterm birth have been identified (e.g., previous preterm birth,^9–11 multifetal gestation,^9–11 infection during pregnancy,^9–11 hypertension or diabetes,^9,10 a body mass index (BMI) ≤ 18.5 kg/m² or ≥ 30 kg/m² ,¹¹ late or no prenatal care,⁹ substance use,⁹ younger (≤24 years) or older (≥35 years) maternal age,¹² low socioeconomic status,^10,11 and lacking a partner¹⁰). Beyond these established risk factors, a growing body of literature has identified associations between psychosocial stressors (e.g., antenatal stress¹³ and discrimination¹⁴), environmental exposures (e.g., ambient fine particulate matter^15,16 and metals¹⁷), as well as neighborhood deprivation¹⁸ and segregation¹⁹ with preterm birth. Experiences of stress, discrimination, and living in under-resourced neighborhoods may activate stress-related biological responses,¹⁸ and exposure to air pollution or metals may induce inflammation and oxidative stress pathways,^16,17 thereby contributing to preterm birth.¹⁵ Further, Black mothers may experience higher levels of psychosocial,^20–22 neighborhood,^23,24 and environmental stressors,^25,26 which suggests a role for these factors as potential drivers of the Black/White disparity in preterm birth.

Blinder–Oaxaca decomposition, a statistical method that quantifies the proportion of the difference in the prevalence of an outcome between two groups that can be explained by a set of covariates,^27,28 has been applied in a handful of studies evaluating the disparity in preterm birth between Black and White mothers.^29–34 Three of these studies, each using US national data for singleton births for a single year (2013,³¹ 2016,³² or 2019³⁴), evaluated the contribution of sociodemographic variables (i.e., maternal age, education, insurance, and marital status or paternal acknowledgment [except for Mehta et al^34]), perinatal risk factors (i.e., prenatal care, parity, gestational hypertension and diabetes [except for Mehta et al^34]), and maternal comorbidities and health behaviors (i.e., hypertension, diabetes, prepregnancy body mass index, and smoking) to racial disparities in preterm birth. In two studies, approximately 22%³⁴ and 27%³¹ of the disparity was explained, with sociodemographic factors explaining the largest proportion at 11.1%³⁴ and 20.6%.³¹ Thoma et al³² included additional medical risk factors (i.e., interpregnancy interval and infection) as well as geographic characteristics (i.e., state of residence, county population, and a county-level measure of racialized and economic segregation) in their decomposition analysis and explained ~38% of the Black/White disparity in preterm birth.³²

Two additional state-based studies used Blinder–Oaxaca decomposition to evaluate factors that explained differences in preterm birth between Black and White women.^30,33 Su et al³³ examined birth records among primiparous women who delivered in Nebraska between 2005 and 2014 and, in addition to individual-level covariates, evaluated urban/rural classification of the mother’s county of residence. Overall, the study explained 29.5% of the preterm birth disparity, with sociodemographic factors explaining the largest proportion at ~29%.³³ Benmarhnia et al³⁰ used California natality data from 2005 to 2010 to decompose the Black/White disparity in preterm birth and included metrics for unemployment, poverty, educational attainment, language proficiency, and air pollution (fine particulate matter (PM_2.5) and nitrogen dioxide (NO₂)) at the zip-code level based on a mother’s zip code of residence at delivery.³⁰ Despite not evaluating comorbidities or perinatal factors, the study explained 39.3% of the disparity, with sociodemographic (17.5%) and neighborhood (16.1%) factors being the largest contributors.³⁰

Owing to gaps in the prior literature that have attempted to explain disparities in preterm birth, we applied Blinder–Oaxaca decomposition to quantify contributions of a host of factors not previously evaluated among non-Hispanic Black and non-Hispanic White mothers who delivered their infants at large obstetric hospitals serving the greater Houston area. Drawing on information gathered during surveys, through electronic health record (EHR) abstraction and chemical analysis of urinary samples, we expanded upon the suite of individual-level factors examined previously by including employment status, previous pregnancy complications, antibiotic use in the past month, secondhand smoke exposure, biomarkers of toxic metals exposure, experiences of stress while pregnant and discrimination over a lifetime. Additionally, our study is the first to examine the impact of the neighborhood where a pregnant woman lived at a localized scale and expanded the evaluation of features that may play a role in explaining racial disparities in preterm birth to include food accessibility and walkability.

Methods

Study population

We conducted an analysis of data collected from 1,847 mothers recruited into the Maternal and Infant Environmental Health Riskscape (MIEHR) study from April 2021 to December 2023.³⁵ Women were recruited from three obstetric hospitals in the Texas Medical Center that provide maternal care to women and serve as a hub for referrals for complicated pregnancies. Research coordinators approached potential participants in ante-/postpartum wards within 24–72 hours after giving birth to confirm eligibility. Among 4,812 eligible women, 1,847 (38.4%) agreed to participate. Eligibility criteria included: ≥18 years old; residing within the greater Houston, Texas area (Brazoria, Chambers, Fort Bend, Galveston, Harris, Liberty, Montgomery, or Waller Counties); self-identifying as non-Hispanic Black or non-Hispanic White (hereafter referred to as Black and White women, respectively); having a singleton delivery without major congenital anomalies; and being able to converse in English. Participants completed interviewer-administered questionnaires; data were abstracted from EHRs; and spot urine samples were collected, aliquoted, and stored at −80°C. The study was reviewed and approved by the Baylor College of Medicine Institutional Review Board (protocols: H-48962 and H-47983) and the University of Texas Health Science Center at Houston Institutional Review Board under a reliance agreement. All participants provided written informed consent before enrollment.

Preterm birth

Based on the clinical estimate of gestational age from the EHR, participants were classified as having a preterm birth if their infant was delivered before 37 completed weeks of gestation.

Individual-level covariates

Individual-level covariates were grouped into the following domains: sociodemographic factors; perinatal history; comorbidities and health behaviors; chemical exposures; and social factors. The following self-reported sociodemographic factors were assessed via questionnaire: maternal age (years), education level (high school or general educational development or less, college degree or higher), annual household income (<$35,000, $35,000–$74,999, ≥$75,000), nativity (born in the US, born outside the US), whether participants were married/living with a partner (yes, no), and employed (yes, no). Sociodemographic variables extracted from the EHR included infant sex and the insurance carrier (subsequently categorized as public or no insurance vs. private or self-pay).

Perinatal history variables obtained from the EHR included parity, gestational diabetes, and hypertensive disorders of the index pregnancy (HDP) that included hypertension in pregnancy, preeclampsia, eclampsia, or hemolysis, elevated liver enzymes, and low platelet count (HELLP) syndrome. Women were also classified as having a previous pregnancy complication if their EHR indicated a history of preterm birth, stillbirth, gestational diabetes, blood transfusion in pregnancy, intrauterine growth restriction, placental abruption, placenta previa, macrosomia, chorioamnionitis, premature rupture of the membranes, HDP, hemorrhage, or having a child with a major congenital anomaly. Finally, the self-reported date of a woman’s first prenatal care visit, assessed via questionnaire, was used to determine the gestational age of prenatal care initiation (<13 weeks, ≥13 weeks, or no prenatal care).

For the comorbidities and health behaviors domain, we extracted the following data from the EHR: diagnosis of type 2 diabetes or chronic hypertension, maternal height (m) and prepregnancy weight (kg), which were used to calculate prepregnancy BMI and categorized based on World Health Organization guidelines: underweight (<18.5 kg/m²), healthy weight (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), and obesity (≥30 kg/m²).³⁶ While we retained a discrete category for missing BMI values in subsequent analyses to preserve sample size, results for this category should not be interpreted substantively. We obtained information on alcohol use during pregnancy (yes, no), antibiotic (yes, no) or probiotic (yes, no) use in the prior month to delivery, and history of COVID-19 positivity (ever, never) from the survey.

Self-reported variables comprising the chemical exposures domain included smoking in the 3 months before pregnancy or during pregnancy (yes, no), secondhand smoke exposure during pregnancy (yes, no), and pesticide use (any use of insecticides, herbicides, fungicides, rodenticides, or antimold/mildew products, none) during pregnancy. Although smoking may also be considered a health behavior, due to the thousands of chemicals within tobacco smoke,³⁷ we chose to classify it within the chemical exposure domain. In addition, we measured urinary concentrations of metals, as described elsewhere.³⁸ Briefly, laboratory analyses of urine for trace metal concentrations were performed at the Wadsworth Center (Albany, NY) using inductively coupled plasma-tandem mass spectrometry/mass spectrometry (Agilent Model 8900, Agilent Technologies, Santa Clara, CA). Before analysis, urine samples were thawed to room temperature, mixed on a rocker for ≥45 minutes, then diluted to 5 ml using a reagent solution containing 2% (v/v) double distilled HNO₃, 0.5% (v/v) veritas ultimate grade HCl (GFS Chemicals, Powell, OH), 1,000 µg/L Au (High Purity Standards, Charleston, SC), and 0.005% (v/v) Triton X-100 (Sigma Aldrich Co., St. Louis, MO). To account for urinary dilution, metal concentrations were normalized to urinary creatinine concentrations, which were measured via high-performance liquid chromatography-tandem mass spectrometry³⁸ and expressed in µg/g of creatinine. For this analysis, we focused on cadmium, nickel, and lead, which were dichotomized at the 75th percentile (0.22, 4.32, and 0.39 µg/g creatinine, respectively) based on their distribution in the study population.

Regarding social factors, MIEHR participants were asked to self-rate their level of stress (not stressful or average stress, very stressful) and paternal support (none or a little, good or excellent) during their pregnancy, concern regarding COVID-19 (none, a little, moderate, and very), and to indicate if they encountered economic or logistical difficulties due to the pandemic (i.e., reduced or lost employment, housing or transportation issues, paying for medical expenses, or accessing child care, food, medication, or healthcare) (yes to any, no to all). Krieger’s Experiences of Discrimination Scale (EOD), a validated assessment of racial discrimination, was also administered; participants responded (yes, no) to whether they had experienced discrimination on the basis of their race, ethnicity, or color in each of nine situations (at school, work, in public, within the legal system, or when obtaining employment, housing, bank loans, credit, or medical care).³⁹ We summed responses to create a composite “EOD situations score,” ranging from 0 to 9.

Characterization of participants’ neighborhood environments

We obtained each participant’s address at delivery from the questionnaire or EHR (in the latter case, participants verbally confirmed that the address was correct), which was geocoded using ArcGIS Pro (version 3.1, ESRI, Redlands, CA) with StreetMap Premium North America 2023 Release 3 basemaps. We calculated the Euclidean distance in meters from the location of a participant’s residence to the nearest major roadway. Major roadways were identified based on the 2020 road data from the Bureau of Transportation Statistics and defined as those with a class 1–3 designation (i.e., freeways, primary or secondary provincial or territorial highways, or municipal arterials).^40,41 Distance to the nearest major roadway was dichotomized at the 25th percentile (300.8 meters) of the distribution in the study population. We also obtained daily estimates of ambient PM_2.5 concentrations (µg/m³) at locations of census tract centroids generated from a hybrid spatiotemporal model^42–44 during the study period (2020–2023). To estimate daily, high-resolution (100 m) exposure to PM_2.5 in our study area, we utilized a Convolutional Recurrent Neural Network (CoReL) framework. The framework works by coupling a convolutional neural network for spatial feature extraction with a long short-term memory network for temporal modeling. In brief, the CoReL framework utilizes a three-step modeling process: first, it reconstructs gap-free, high-resolution satellite-based aerosol optical depth surfaces; second, it integrates these surfaces with ground-based air quality system measurements alongside a parsimonious set of geographic and meteorological predictors (including land-use, traffic, and atmospheric data); and third, it generates exposure surfaces using a super-learner ensemble that combines multiple machine learning algorithms (e.g., LightGBM, Random Forest, and XGBoost) with a geospatial Gaussian Markov Random Field algorithm, optimizing all hyperparameters via 10-fold cross-validation. Similar to methods applied in a previous decomposition analysis,³⁰ for each census tract, PM_2.5 estimates were averaged across the entire study period and assigned to MIEHR participants based on the census tract of their delivery address. This variable was then dichotomized at the 75th percentile (10.1 µg/m³) based on its distribution in the study population.

Using the National Land Cover Database 2021 data,⁴⁵ we calculated both the percentage of tree canopy cover within 100-meter buffers of participants’ addresses and classified census tracts as having high heat risk if they met the following criteria: no bodies of water within their geographic boundary; <5% tree cover; and >60% impervious ground cover.⁴⁶ We characterized walkability at the census block group level by using the US Environmental Protection Agency’s National Walkability Index,⁴⁷ a scale ranging from 1 to 20, derived by ranking and dividing census block groups into 20 quantiles based on four features of the built environment—intersection density, proximity to transit, and two measures of land diversity—which are weighted in the final formula.⁴⁸ We assessed food access using a variable within the US Department of Agriculture’s 2019 Food Access Research Atlas that measures the percentage of households in each census tract without a vehicle and located >½ of a mile from a supermarket.⁴⁹

We calculated the Area Deprivation Index (ADI)⁵⁰ and Index of Concentration at the Extremes (ICE) for race and income⁵¹ using data from the 2022 US Census American Community Survey. We used the “sociome” Package in R to construct census tract-level ADI.⁵² The ICE metric was calculated as the difference in the number of non-Hispanic White residents with a household income > the US 80th percentile (i.e., $100,000) and the number of Black residents in a given census tract with a household income ≤the 20th percentile (i.e., $20,000), divided by the total population in the census tract.⁵¹ The ICE metric ranges from −1 to 1, which represents areas comprised entirely of low-income Black persons or high-income, non-Hispanic White persons, respectively. We dichotomized ICE based on the 75th percentile (0.279) of its distribution in the study population. Finally, we used the Environmental Systems Research Institute crime index, a relative measure of seven tracked crimes (i.e., murder, rape, assault, robbery, burglary, larceny, and motor vehicle theft) occurring in a census tract as compared with the national average, with a score >100 indicating above-average risk.⁵³ We categorized this variable into the following crime risk categories: low (≤100), above average (>100–<200), and high (≥200).

Statistical analysis

We conducted descriptive statistics for the total population (n = 1,847) and separately for Black (n = 1,148) and White (n = 699) women. To be included in the decomposition analysis, we required (1) categorical variables to have ≥10 cases of preterm birth per level when stratified by race to avoid issues related to small cell sizes and (2) variables to be associated with preterm birth in univariable logistic regression analyses at an alpha level of <0.1. Logistic regression was conducted in SAS version 9.4 (SAS Institute, Cary, NC).

To explore the contributions of covariates described above to the Black/White disparity in preterm birth, we performed two-fold Blinder–Oaxaca decomposition with a nonlinear extension to facilitate comparisons with prior studies that used the same³⁰ or a similar (Fairlie) decomposition method.^32,33 Within a regression model framework, two-fold decomposition partitions the disparity into two components: (1) the endowment, or portion explained by group differences in the observed mean levels of covariates, and (2) the unexplained component, or portion reflecting unobserved variables not included in the model (i.e., differences in the intercepts) and differential effects of variables (i.e., differences in regression coefficients between groups).^27,28 This partitioning of the disparity represents the primary advantage of decomposition over other approaches used in disparities research to understand drivers of health disparities (e.g., stratification⁵⁴ or nested regression models^22,55). Decomposition was performed in Stata version 18.5 (StataCorp, College Station, Texas) using the Oaxaca command with the logit extension. For the endowment, percent contributions are reported for individual covariates and for covariates grouped by domain. Positive values indicate covariates that widen the preterm birth disparity, while negative values indicate covariates that narrow the disparity.

Results

In total, six variables failed to satisfy one of the criteria for inclusion in the decomposition model because they had at least one level with fewer than 10 cases of preterm birth when stratified by race (i.e., income; nativity; chronic diabetes; paternal support in pregnancy; concern regarding COVID-19; heat risk). An additional 12 variables were excluded because they had P values ≥0.1 in the univariable logistic regression analyses (i.e., age; infant sex; education; initiation of prenatal care; history of COVID-19 positivity; alcohol use in pregnancy; probiotics used in past month; insecticide use; crime index; distance to nearest major road; urinary lead levels; and tree canopy coverage). Supplemental Table 1 (https://links.lww.com/EE/A464) provides results from the univariable logistic regression analyses for the 22 variables that met both inclusion criteria.

Tables 1 and 2 present descriptive statistics for individual- or neighborhood-level covariates, respectively, that satisfied inclusion criteria, grouped by domain. White participants generally had more advantaged sociodemographic and perinatal and comorbidity profiles (Table 1). A greater proportion of Black participants, as compared with White participants, had a diagnosis of HDP (37.5% vs. 26.0%), a previous pregnancy complication (35.3% vs. 24.5%), chronic hypertension (15.2% vs. 6.0%), were obese (34.0% vs. 18.9%), and used antibiotics in the prior month (22.0% vs. 14.7%). In terms of chemical exposures, Black women reported higher levels of secondhand smoke exposure (23.9% vs. 10.0%), but a higher proportion of White women had urinary nickel concentrations in the highest quartile (31.8% vs. 20.6%). In terms of social factors, on average, Black participants reported experiencing discrimination in more situations (mean EOD situations score = 1.7, SD = 2.3 vs. 0.3, SD = 0.8) and more economic or logistical difficulties from COVID-19 (57.1% vs. 45.2%) compared with their White counterparts. When examining neighborhood characteristics (Table 2), Black participants generally lived in more deprived neighborhoods, reflected by a higher median ADI (110.9 vs. 85.8) and a slightly higher proportion of residents with low food access (3.3% vs. 1.3%). Additionally, a greater proportion of Black participants resided in census tracts with ICE scores in the lower three quartiles (91.2% vs. 48.1%) when compared with White participants. More White women, however, resided in census tracts with higher PM_2.5 exposure (≥75th percentile) compared with Black women (30.3% vs. 21.7%).

Table 1.

Distribution of individual-level factors included in the decomposition analysis of preterm birth among MIEHR study participants in greater Houston, Texas, 2021–2023

	All women n = 1,847	Black women n = 1,148	White women n = 699
Sociodemographic factors
Not married/living with a partner, n (%)	763 (41.3)	694 (60.5)	69 (9.9)
Public or no insurance, n (%)	1006 (54.5)	857 (74.7)	149 (21.3)
Missing	12 (0.7)	5 (0.4)	7 (1.0)
Not employed, n (%)	734 (39.7)	553 (48.2)	181 (25.9)
Perinatal history
Parity, mean (SD)	1.2 (1.4)	1.4 (1.5)	0.9 (1.0)
Previous pregnancy complication, n (%)	576 (31.2)	405 (35.3)	171 (24.5)
HDP, n (%)	612 (33.1)	430 (37.5)	182 (26.0)
Gestational diabetes, n (%)	114 (6.2)	65 (5.7)	49 (7.0)
Comorbidities and health behaviors
Prepregnancy BMI, n (%)
Underweight	24 (1.3)	17 (1.5)	7 (1.0)
Healthy weight	420 (22.7)	206 (17.9)	214 (30.6)
Overweight	308 (16.7)	172 (15.0)	136 (19.5)
Obese	522 (28.3)	390 (34.0)	132 (18.9)
Missing	573 (31.0)	363 (31.6)	210 (30.0)
Chronic hypertension, n (%)	217 (11.8)	175 (15.2)	42 (6.0)
Antibiotics used in past month, n (%)	355 (19.2)	252 (22.0)	103 (14.7)
Missing	5 (0.3)	3 (0.3)	2 (0.3)
Chemical exposures
Smoking in the 3 months before and/or during pregnancy, n (%)	172 (9.3)	117 (10.2)	55 (7.9)
Missing	1 (0.1)	0 (0)	1 (0.1)
Secondhand smoke exposure in pregnancy, n (%)	344 (18.6)	274 (23.9)	70 (10.0)
Missing	2 (0.1)	0 (0)	1 (0.1)
Cadmium ≥75th percentile, n (%)	458 (24.8)	291 (25.4)	167 (23.9)
Missing	13 (0.7)	9 (0.8)	4 (0.6)
Nickel ≥75th percentile, n (%)	458 (24.8)	236 (20.6)	222 (31.8)
Missing	13 (0.7)	9 (0.8)	4 (0.6)
Social factors
Very stressful pregnancy, n (%)	466 (25.2)	271 (23.6)	195 (27.9)
Missing	5 (0.3)	3 (0.3)	2 (0.3)
Economic/logistical difficulties from COVID-19, n (%)	972 (52.6)	656 (57.1)	316 (45.2)
Missing	10 (0.5)	5 (0.4)	5 (0.7)
EOD situations score, mean (SD)	1.2 (2.0)	1.7 (2.3)	0.3 (0.8)
Missing	3 (0.2)	1 (0.1)	2 (0.3)

BMI indicates body mass index; COVID-19, Coronavirus disease 19; EOD, Experiences of Discrimination; HDP, hypertensive disorders of pregnancy; MIEHR, Maternal and Infant Environmental Health Riskcape; SD, standard deviation.

Table 2.

Distribution of neighborhood-level factors included in the decomposition analysis of preterm birth among MIEHR study participants in greater Houston, Texas, 2021–2023

	All women n = 1,847	Black women n = 1,148	White women n = 699
Neighborhood characteristics
ADI, median (Q1, Q3)	100.9 (85.7, 116.5)	110.9 (95.7, 123.4)	85.8 (75.4, 98.9)
ICE <75th percentile, n (%)	1383 (74.9)	1047 (91.2)	336 (48.1)
Low food access (%), median (Q1, Q3)	2.13 (0.8, 5.0)	3.3 (1.2, 6.4)	1.3 (0.5, 3.0)
PM2.5 ≥75th percentile, n (%)	461 (25.0)	249 (21.7)	212 (30.3)
Walkability index, median (Q1, Q3)	11.2 (7.8, 14.3)	11.3 (8.3, 14.0)	10.3 (7.0, 15.0)

ADI indicates Area Deprivation Index; ICE, Index of Concentration at the Extremes; Q1, first quartile; Q3, third quartile; MIEHR, Maternal and Infant Environmental Health Riskcape, PM_2.5, fine particulate matter.

Black/White disparity in preterm birth

Black mothers had a preterm birth prevalence of 23.7%, compared with 17.1% in White mothers, reflecting a 6.6% disparity and, in total, the decomposition model explained 63.0% of the disparity in preterm birth (Table 3). When we evaluated contributions by domain, perinatal history and neighborhood characteristics explained the largest portions of the disparity at 36.3% and 32.8%, respectively, followed by social factors at 6.0%. Factors with relatively large individual contributions included having chronic hypertension (16.8%, 95% CI = 5.2%, 28.4%), a previous pregnancy complication (18.5%, 95% CI = 5.7%, 31.3%), or HDP (16.7%, 95% CI = 4.6%, 28.7%), being unmarried or living without a partner (13.0%, 95% CI = −28.7%, 54.7%), residing in census tracts with ICE scores in the lower 3 quartiles (19.2%, 95% CI = −47.8%, 86.1%) or with lower food access (15.8%, 95% CI = −3.9%, 35.6%), and more experiences of discrimination 10.3%, 95% CI = −14.2%, 34.8%).

Table 3.

Decomposition of the Black/White disparity in preterm birth among participants in the MIEHR study in greater Houston, Texas, 2021–2023 (n = 1,802).

Preterm birth prevalence	% (95% CI)	-
Black	23.7 (21.2, 26.2)	-
White	17.1 (14.3, 19.8)	-
Black/White difference	6.6	-
Difference explained
Total explained	63.0 (−11.8, 137.8)	-
Total unexplained	37.0 (−52.1, 126.1)	-
	% Contribution explained by:
	Individual factors (95% CI)	Domain
Sociodemographic factors		−11.9
Not married/living with a partner	13.0 (−28.7, 54.7)	-
Public or no insurance	−22.8 (−78.1, 32.5)	-
Not employed	−2.1 (−20.4, 16.2)	-
Perinatal history		36.3
Parity	2.3 (−13.0, 17.6)	-
Previous pregnancy complication	18.5 (5.7, 31.3)	-
HDP	16.7 (4.6, 28.7)	-
Gestational diabetes	−1.2 (−4.2, 1.8)	-
Comorbidities and health behaviors		11.9
Prepregnancy BMI		-
Overweight	1.1 (−5.6, 7.8)	-
Obese, ≥30	−7.1 (−25.3, 11.0)	-
Missing	−0.5 (−3.2, 2.1)	-
Chronic hypertension	16.8 (5.2, 28.4)	-
Antibiotics used in past month	1.7 (−5.2, 8.5)	-
Chemical exposures		−12.1
Smoking in the 3 months before and/or during pregnancy	0.6 (−2.2, 3.4)	-
Secondhand smoke exposure in pregnancy	1.9 (−11.4, 15.2)	-
Cadmium ≥ 75th percentile	1.1 (−1.9, 4.0)	-
Nickel ≥ 75th percentile	−15.6, (−28.3, −2.9)	-
Social factors		6.0
Very stressful pregnancy	−4.2 (−10.1, 1.7)	-
Economic/logistical difficulties from COVID-19	−0.1 (−10.1, 9.1)	-
EOD situations score (per 1 unit increase)	10.3 (−14.2, 34.8)	-
Neighborhood characteristics		32.8
ADI (per IQR increase)	−6.8 (−71.6, 57.9)	-
Low food access (per IQR increase)	15.8 (−3.9, 35.6)	-
ICE <75th percentile	19.2 (−47.8, 86.1)	-
Walkability index (per IQR increase)	0.7 (−4.2, 5.6)	-
PM2.5 ≥ 75th percentile	3.9 (−5.4, 13.1)	-

ADI indicates Area Deprivation Index; ICE, Index of Concentration at the Extremes; BMI, body mass index; CI, confidence interval; IQR, interquartile range; MIEHR, Maternal and Infant Environmental Health Riskcape, PM_2.5, fine particulate matter.

Discussion

After applying decomposition methods to understand the contributions of “traditional” and more novel factors to disparities in preterm birth among women in the MIEHR study, our model explained 63.0% of the disparity, a considerably larger proportion compared with previous decomposition analyses.^30–34 Notably, we found perinatal history (36.3%) and neighborhood characteristics (32.8%) explained the largest proportion of differences in the prevalence of preterm birth. Additionally, factors not previously evaluated in prior decomposition studies (i.e., previous pregnancy complications, food accessibility, and discrimination) contributed meaningfully to the disparity as well.

Within the perinatal history group, having a history of pregnancy complications or HDP were the most important individual variables. The pathophysiology of HDP involves improper vascular remodeling of the spiral arteries, leading to hypoperfusion of the placenta⁵⁶ and subsequent fetal growth restriction, oligohydramnios, and placental abruption, which increase the risk for preterm delivery.⁵⁷ Given that Black women have a higher prevalence of pregnancy-associated hypertension compared with women of other races or ethnicities,⁵⁸ it is not surprising that HDP was a large contributor to the Black/White preterm birth disparity in our study. Although the literature provides links between previous pregnancy complications^10,59 with preterm birth, and the prevalence of these factors is higher among US Black versus White women,^60,61 previous decomposition studies have not evaluated this factor. Previous studies have also not evaluated contributions of HDP as a composite outcome, but Su et al³³ and Thoma et al,³² evaluated gestational hypertension and found it explained 2% and 9.9% of the endowment, respectively. These estimates are considerably lower than what we found for HDP (16.7%), which included gestational hypertension as well as higher acuity conditions (e.g., preeclampsia).

Few studies have evaluated neighborhood factors when decomposing the Black/White disparity in preterm birth despite consistent associations between neighborhood disadvantage and adverse pregnancy outcomes.^18,62 Furthermore, prior investigations^30,32,33 assessed neighborhood factors at relatively large geographic scales (e.g., county or zip code level), likely obscuring variation at more localized levels. Although together, the neighborhood factors included in our decomposition model explained over a third of the disparity, this was largely due to living in a neighborhood characterized by racialized and economic segregation (ICE) or low food access. Although the study by Thoma et al³² also found that ICE contributed positively to the Black/White preterm birth disparity, the magnitude of the contribution was quite small (i.e., 1.8%). However, in that study, ICE was evaluated at the county level, which points to the importance of characterizing neighborhoods on relatively small spatial scales. Although not examined in previous decomposition analyses, we found low food access was important in explaining the Black/White preterm birth disparity among women in our study. Neighborhood-level food access could be a marker of food insecurity,⁶³ which has been previously associated with preterm birth^64,65 and also disproportionately impacts Black households,⁶⁶ providing further support of our findings.

Within the “comorbidities and health behaviors” domain, chronic hypertension emerged as one of the largest individual contributors at 16.8%. As chronic hypertension is associated with placental hypoperfusion phenotypes^67,68 and superimposed preeclampsia,^57,67,69 having this comorbidity increases the risk of preterm delivery.^67,69,70 As with HDP, Black women experience a higher burden of chronic hypertension than White women.⁷¹ Previous studies also found chronic hypertension contributed to the Black/White preterm birth disparity at lower magnitudes (endowments ≤3.6%)^33,34 as compared with our finding, possibly due to smaller Black/White differences in the prevalence of hypertension than in our study (≤2.1% vs. 9.2%) and potential underreporting of maternal chronic illnesses on birth certificates, particularly among racial and ethnic minorities.^72–74 Among the sociodemographic factors we evaluated, being unmarried or living without a partner explained the largest proportion (13.0%) of Black/White differences in preterm birth. Our finding aligns with results from prior studies, which reported contributions of 10.1% to 13.8%,^30–33 except for Benmarhnia et al³⁰ (2.7%). In contrast to previous studies that found public insurance explained the disparity in preterm birth,^30–34 our results were in the opposite direction.

To the best of our knowledge, we are the first study to evaluate chemical exposures including (self-reported) secondhand smoke and urinary concentrations of several toxic metals (i.e., cadmium and nickel), as well as social stressors, as contributors to the Black/White disparity in preterm birth in the context of a decomposition analysis. Regarding stress, we found that racial discrimination positively contributed to the Black/White preterm birth disparity. Experiencing discrimination can induce stress responses, including increased cortisol and catecholamine release, both of which cause conditions that increase the risk of preterm delivery.¹⁸ Although prior decomposition studies have not included measures of discrimination, the epidemiologic literature supports not only an association between discrimination and preterm birth,¹⁴ but also provides evidence that Black women experience a higher prevalence of discrimination as compared with White women.^20–22 On the other hand, there was little contribution due to smoking during pregnancy, secondhand smoke exposure, or urinary cadmium levels to the Black/White disparity in preterm birth. However, nickel exposure had a relatively strong negative contribution due to higher levels among White women, which we cannot explain as we do not have information on potential sources of this metal in our study population.

Strengths and limitations

Our study expands upon prior decomposition analyses by incorporating novel factors—such as pregnancy complications, chemical exposures, social stressors, and localized neighborhood factors. Notably, several of these factors contributed meaningfully to explaining the preterm birth disparity, including previous pregnancy complications, EOD, and food accessibility. We also utilized a method well-suited for health disparities research—Blinder–Oaxaca decomposition, which directly partitions the observed disparity based on unequal exposures and differential variable effects between groups.⁵⁴ We evaluated neighborhood metrics at relatively small geographic areas (i.e., census tract and block group), improving neighborhood characterizations. Although, where possible, we relied on documented information from participants’ EHRs (i.e., medical diagnoses) to minimize bias, we also collected data via survey at the time of enrollment; thus, we cannot establish temporality between some risk factors and preterm birth, and like all studies relying on retrospectively collected data, our study could be subject to recall bias. Additionally, we did not capture information on maternal diet or occupation, two factors that may contribute to preterm birth. Also, we relied on participants’ address at delivery to characterize the neighborhood and did not account for potential residential mobility. Finally, notwithstanding the advantage of incorporating variables ascertained by questionnaire or EHR abstraction (or through chemical analyses of biological samples), we were limited by sample size, particularly in comparisons to other studies that applied decomposition methods to national or state-wide data.

In conclusion, our study expands knowledge on factors that may contribute to the Black/White disparity in preterm birth among participants living in a diverse, urban landscape. We found prior and current perinatal complications, racialized and economic segregation, and low food access comprised most of the disparity, along with chronic hypertension. Neighborhood-based factors should be further explored in larger datasets and across different populations. Findings from this analysis can help inform policy and interventions to address this critical gap in perinatal health outcomes.

Conflicts of interest statement

The authors declare that they have no conflicts of interest with regard to the content of this report.

ACKNOWLEDGMENTS

We would like to acknowledge Dr. Syam Andra from the Institute for Exposomic Research, Department of Environmental Medicine, Icahn School of Medicine at Mount Sinai and Dr. Kurunthachalam Kannan from the Division of Environmental Health Sciences, Wadsworth Center, New York State Department of Health, for their contributions in measuring urinary creatinine levels.