Air pollution has been associated with adverse birth outcomes, with variation by socioeconomic position (SEP). However, it remains unknown which aspects of lower SEP – comprised of myriad physical and psychosocial stressors – may best explain observed pollution susceptibilities. Building upon previous studies that estimated joint associations of air pollution and socioeconomic deprivation on term birth weight in New York City (NYC), this study seeks to identify specific social stressors underlying that relationship.

We examined records for 243,853 term births in NYC from 2007–2010. Residence-specific pregnancy-average NO₂ was estimated using NYC Community Air Survey (NYCCAS) and EPA regulatory data. Twenty-six community social stressor indicators were tested as modifiers of NO₂-birthweight associations in linear mixed-effects models, adjusting for particulate matter (PM_2.5), individual-level maternal characteristics, and other covariates. In sensitivity analyses, we also examined non-linear interactions between continuous NO₂ and census-tract level violence and deprivation terms.

Consistent with previous work, a 1-IQR (6.2 ppb) increase in average prenatal NO₂ exposure was associated with a 12.6 (SE = 2.7)-gram decrease in term birthweight.We observed similar values in independent models for most stressors related to violent crime or SEP and significantly lower birthweights with higher stressor exposures.

In models of effect modification, however, we found that – despite lower average birthweights in high-stressor communities – NO₂-birthweight associations were weaker in higher-stressor communities, particularly for violence-related stressors. For example, in the highest-quartile communities for assault, a 1-IQR increase in NO₂ exhibited a decrement of only 7.3 g, on average, compared to 16.9 g in the lowest-assault quartile (p = .01 trend across quartiles). Exposures to non-violent stressors were not significantly associated with lower birthweights, nor modified observed NO₂-birthweight associations.

We found significantly lower term-infant birthweights with higher NO₂ or community stressors. Counter to hypotheses, however, in communities with very high stressor exposures (esp. violent crimes), despite lower overall birthweights, associations for NO₂ were weaker than in low-stressor communities.

Our results suggest a possible saturation effect in stress-pollution interactions, wherein very high stressor exposures appear to overwhelm any effects of pollution. In addition, we observed stronger effects for violent crimes, in relation to other social stressors.

Urban air pollution has been associated with negative birth outcomes including lower birthweights [1, 2], higher rates of preterm birth [3], neurocognitive impairments [4], and infant mortality [5]. Lower socioeconomic position (SEP) has also been associated with both adverse birth outcomes [6] and greater susceptibility to pollution [7]. It remains unknown, however, which aspects of lower SEP – a construct that includes a host of physical and psychosocial stressors (e.g., housing quality, crime, food insecurity, etc.) – may best explain observed variation in pollution susceptibility by SEP.

Associations between urban air pollution and birth outcomes may be mediated, in part, via systemic oxidative stress [8, 9], or inflammation [10], leading to placental insufficiency. There has been substantial variation in observed relationships between air pollution and fetal growth, however, potentially attributable to differences in exposure measurement or misclassification [11], co-pollutant adjustments [12], incomplete adjustments for confounding, or differential exposure–response relationships across populations. Appropriately accounting for SEP is a particular challenge, as SEP is often spatially correlated with air pollution [13, 14], and thus may confound measures of association, while common biologic pathways related to inflammation, immune, and neuroendocrine dysregulation may facilitate synergistic effects, and variation in true effects of pollution across SEP subgroups [15, 16].

Low area-level SEP has been associated with adverse pregnancy outcomes after accounting for individual-level SEP. Separately, studies have reported greater susceptibility to air pollution in communities of lower SEP, with effects on respiratory disease and asthma [17, 18], cardiovascular disease [19, 20], birthweight [21, 22], and neurocognitive development [23, 24]. Notably, the key “causal constituents” underlying SEP-related susceptibility remain undefined [25], as SEP includes a host of chemical and non-chemical exposures accumulating and interacting over the life course. Growing evidence suggests that chronic stressors, which may explain a substantial portion of this SEP-related susceptibility [26], given evidence of immune, endocrine, and metabolic dysregulation associated with chronic psychosocial stress (i.e., allostatic load) [27].

Racial disparities in birth outcomes including preterm birth and low birthweight are well-documented, and may be attributable to substantial differences in social and environmental exposures between non-Hispanic Black and non-Hispanic White individuals [28]. In a commentary discussing the future of perinatal epidemiology and pre-term birth cohorts [29], Ness and colleagues described the need to disentangle complex interactions among social and physical exposures—which requires large and diverse samples, detailed information on environmental exposures and community stressors, and innovative analytic strategies to address persistent spatial confounding among correlated exposures [30].

In this study, we explored a more comprehensive set of community stressors associated with lower SEP, to consider their independent and synergistic associations, with full-gestation nitrogen dioxide (NO₂) exposures, on term birthweight, in an urban setting. We focused on NO₂ as a marker of traffic-related air pollution, with substantial spatial variation within New York City (NYC) [29]. As our health outcome, we focused on birthweight among term births, as an indicator of fetal growth with important life-course and population health implications [31]. Here, we build on a prior study of fine-scale spatio-temporal air pollution exposures and birth outcomes in NYC [32, 33], which reported significant associations between fine-scale NO₂ with term birthweight [32], and stronger NO₂-birthweight associations in census tracts in the highest and lowest quartile of a socioeconomic deprivation index (SDI) [7].This work incorporated previously-generated data on area-level air quality [32] and disaggregated community stressors that were used to develop composite social indicators [7] to measure their association with birth weight in NYC. We focused on analyzing the main effect and modifying role of individual community stressors (n = 26) in the association between NO₂ and birthweight.

We examined records for approximately 380,000 live births among residents of the five boroughs of NYC in 2008–2010, merged with patient-level data from the New York State Department of Health Statewide Planning and Research Cooperative System (SPARCS), which includes all licensed healthcare facilities. Data were restricted to full-term (37—42 weeks gestation) singleton births with no congenital anomalies, no self-reported maternal smoking during pregnancy, and with complete residential address and covariate data. Exclusion criteria for implausible clinical values and fixed cohort bias [34] in this population are previously detailed [32, 35].

Residence-specific nitrogen dioxide (NO₂) and fine particulate matter (PM_2.5) exposure estimates were created using 100-m resolution spatial surfaces from the NYC Community Air Survey (NYCCAS) and daily EPA regulatory data and averaged each live birth based on the gestation period. We estimated near-residence (300-m) air pollution exposures between December 2008 and December 2010. NYCCAS monitoring and modeling methods are detailed elsewhere [29, 36]. Briefly, NYCCAS used spatial saturation monitoring to estimate multiple combustion-related pollutants across 150 locations in four seasons over two years. Also, two-week integrated samples were collected at street-level (10–12 feet) each season.

Exposure estimates were temporally adjusted using regulatory monitoring data to match individual-level gestation periods. Births were geocoded to the mother’s residential address at delivery. NYCCAS pollution concentration surfaces were used to estimate near-residence exposure as the mean concentration within a 300 m radial buffer [33]. In keeping with previous analyses, we examined full-gestation average NO₂, as modified by social stressors, and adjusted for spatio-temporal PM_2.5 as a co-pollutant [7].

Birth records were linked through NYC’s Bureau of Vital Records to NYC residents. After applying the exclusion criteria, we restricted our sample population to 243,345 births. Term birthweight as an outcome was examined as a continuous variable. We adjusted for individual covariates associated with fetal growth, including averaged prenatal PM_2.5 exposure, maternal age, race/ethnicity [White, Black, Hispanic, or Asian, cross-classified by United States (US)- vs. foreign-born status], maternal age, education (< 9, 9–11, 12, 13–15, 16, or > 16 years), Medicaid status (yes/no), pre-pregnancy body mass index (BMI), receipt of prenatal care (yes/no), year of conception, parity (number of prior live births), infant sex, and gestational age (in weeks). To adjust for temporal trends in pollution, we also adjusted for conception year and season, as previous [32].

We linked individual maternal residences to administrative data representing 23 individual community social stressors, derived from federal, state, and city data sources, and broadly representing stressors in six domains [i.e., crime and violence; physical disorder, noise disruption, SEP, and access to health and mental health care]. We then re-aggregated indicators from multiple geographies (e.g., police precinct, census tract) to the United Hospital Fund (UHF) area (n = 34), as in Shmool et al. [37]. Three additional indicators are composite variables from a factor analysis of the 23 stressors, which revealed highly correlated factors characterized by: violence and physical disorder (Factor 1), crowding and poor resource access (Factor 2), and complaints related to air pollution or noise (Factor 3). Each indicator is shown in Table 1.

For comparability, all stressors were coded such that higher values were hypothesized to result in greater risk. We examined variation in birthweight by interquartile range (IQR) of each stressor, testing each as a direct predictor of birthweight (continuous variables). To test as potential modifiers of NO₂-birthweight associations (interaction analyses), we categorized stressors into quartiles.

Mixed effects regression

We used a linear mixed-effects approach to estimate the independent and interactive effects for NO₂ and each social stressor on birthweight. We adjusted for spatio-temporal gestation-averaged PM_2.5 and covariates, and used a random effect to account for participants within UHFs. The model takes the form shown in Equation 1, where \({Y}_{ij}\) is the birthweight outcome for the jth person in the ith UHF spatial unit, \({\beta }_{0}\) is the intercept and grand mean, \({b}_{i}\) is a random intercept change for each UHF, gestation-mean NO₂ concentration is denoted as \({X}_{ij1}\), \({X}_{ij2}\) is an categorical social stressor, \({X}_{ijp}\) include covariates, and \({\varepsilon }_{ij}\) is the error term.

Equation 1:   

Each social stressor was tested separately as a main effect (continuous variable) adjusted for NO₂ and covariates, and as a categorical modifier of NO₂-birthweight associations in interaction models, using the lowest quartile of each stressor as the referent category. We tested for linear trends across quartiles, using two-sided tests and alpha of 0.05.

Assumption of linearity, and model specification

Before finalizing the modeling approach and testing interactions, we tested for non-linearity in the NO₂-birthweight relationship by fitting an array of models using non-linear formulations of NO₂ [i.e., natural splines with varying numbers of knot points (1 to 5 knots, and unspecified)], both overall and within quartiles of each social stressor. We also examined non-linear model formulations within varying ranges (subsets) of NO₂ exposures (i.e., including and excluding outliers, 10—90 th percentile, 25 th—75 th percentile, by stressor quartile) using generalized additive mixed models (GAMM) with a random effect for level of aggregation (UHF or census tract).

We tested the robustness of our results to additional specifications in our modeling methods – including the exclusion of a random effect for UHF in generalized additive models (GAM) models with and without interactions by community social stressors, and various categorizations for stressor modifiers (tertiles, quartiles).

We explored non-linear interactions between continuous NO₂ and census tract-level indicators of violent crime (assault and total violent crime per 10,000 residential population) and SEP (poverty rate and a socioeconomic deprivation index (SDI)—a composite indicator of tract-level material deprivation [7, 54].

Finally, we examined the robustness of model adjustment by SDI, to account for effects of other co-occurring stressors.

Among 243,345 live births in NYC, the majority (90.3%) of measured birthweights ranged between 2,500 to 3,999 g. Average prenatal NO₂ was 26.8 ppb (SD = 5.3 ppb) (Table 2). The majority (64.1%) of births occurred at 39 to 40 weeks gestation to individuals between the ages of 25 and 35 (53%), with normal pre-pregnancy BMI (54.3%). The plurality of individuals reported their race and ethnicity as self-reported foreign-born Hispanic (21.9%) followed by US-born White (19.4%). Summary statistics for social stressor indicators are presented in Supplemental Table 1.

On average, we observed a 12.9 (SE = 2.7)-gram decrease in newborn birthweight per IQR increase in NO₂ (IQR = 6.2 ppb) in models adjusting for all covariates except social stressors.

In models adjusted for each social stressor (Fig. 1), we observed that NO₂-birthweight associations were generally robust, though ranged from −14. 7 to −11.4 g per IQR NO₂, suggesting that NO₂-birthweight associations were not overly confounded by social stressors. NO₂-birthweight associations were most attenuated when adjusted for community poverty rates.

Estimated decrease in term birthweight (g) per IQR NO₂, after adjusting for covariates and each stressor individually (listed across x-axis). Estimate without any stressor adjustment denoted in black. Models were adjusted for averaged prenatal PM_2.5 exposure, maternal age, race/ethnicity, maternal age, education, Medicaid status (yes/no), pre-pregnancy body mass index, receipt of prenatal care, year of conception, parity, infant sex, gestational age (in weeks), conception year, and season

Full size image

Observing stressor-birthweight associations across the six domains (Fig. 2), we found negative associations with birthweight for most variables related to violent crime (child abuse, assault, murder, robbery, and low perceived safety) or lower SEP (delayed rent, food stamps, poverty, and unemployment), as a main effect. Associations for stressors in all other categories were mixed.

Estimated difference in birthweight per IQR increase in each social stressor, by stressor domain, adjusted for NO₂, as main effects, and not accounting for interactions. Models were also adjusted for averaged prenatal PM2.5 exposure, maternal age, race/ethnicity, maternal age, education, Medicaid status (yes/no), pre-pregnancy body mass index, receipt of prenatal care, year of conception, parity, infant sex, gestational age (in weeks), conception year, and season

Full size image

In linear models testing the independent associations between each continuous social stressor and birthweight, adjusted for NO₂ and covariates, we found a statistically significant decrease in birthweight per IQR increase in UHF-area reported child abuse [β = 12.1 g (95% CI = 1.2–23.0 g)] and the composite Factor 1 (violent crime and physical disorder) [β = 13.3 g (95% CI = 0.7–26.0 g)]. We also observed significantly lower birthweights with poverty [β = 10.8 g (95% CI = 6.8–14.8 g)], unemployment [β = 11.2 g (95% CI = 0.7–21.7 g)], and delayed rent [β = 1.3 g (95% CI = 0.3–2.3 g)]. Finally, we observed significantly lower birthweights with more local unclean parks [β = 7.7 g (95% CI = 0.3–15.1 g)]. When adjusting for NO₂ and covariates, we observed decreasing birthweights across increasing quartiles of each stressor listed above, and for assault, murder, robbery, housing violations, and reliance on food stamps (p-trend < 0.05) (Supplemental Table S3). Residence in the highest quartile for murder, compared to the lowest, adjusting for NO₂ and covariates, was associated with a decrement of 22.1 g, on average, with consistent decreases in mean birthweight across quartiles (p-trend = 0.02) (Supplemental Table S3).

In models examining variation in NO₂-birthweight associations by each categorical social stressor (interactions), nine of the twenty-six tested stressor indicators (child abuse, assault, food stamps, murder, unclean parks, poverty, robbery, perceived lack of safety, and unemployment) demonstrated potential interactive effects in one more stress or quartiles and five (assault, murder, robbery, perceived lack of safety, unclean parks) displayed consistent NO₂-birthweight associations, across stressor quartiles (Table 3). In each case, average birthweights were lowest in the highest-stressor quartile (Supplemental Table 2) but observed effects of NO₂ were weaker with greater stressor intensity.

For example, the average observed birthweight was 35 g lower in communities in the highest quartile for murder rates (mean = 3318 g, SD = 455 g), compared to the lowest quartile (mean = 3353 g, SD = 437 g) (Supplemental Table 2). However, when considering interactions between NO₂ levels and social factors, in communities with high murder rates, a 1-IQR increase in NO₂ resulted in only a 3.3-g reduction in birthweight, compared to a 15.1-g reduction in the lowest-murder communities (Table 3). Similarly, while birthweight generally decreased across quartiles of increasing assault (p-trend ≤ 0.001; Supplemental Table 2), the observed effect of NO₂ was dampened from −16.9 g to −7.3 across quartiles.

As part of a sensitivity analysis for models that exhibited significant interactions with categorized social stressors as reported in Fig. 3, we mutually adjusted each stressor-NO₂ interaction model with significant social stressors (child abuse, assault, food stamps, murder, unclean parks, poverty, robbery, perceived lack of safety, and unemployment) as covariates. Overall, we observed point estimate changes across quartiles for each stressor (Supplemental Table S-4). For example, the highest quartile for categorized Murder shifted from 3.3 g reduction to 71.0 g reduction in birth weight per 1-IQR increase in NO₂.

Predicted birthweights and NO₂-birthweight associations (slopes) across selected community stressor quartiles (1 = lowest stressor exposure; 4 = highest) based on NO₂ (continuous)-social stressor (categorical) interaction linear mixed effects models. Estimates are shown for the five stressors displaying consistent change in the observed effect of NO₂ on birthweight, across stressor quartiles, shown for the 10 th to 90 th percentile of NO₂ exposures

Full size image

In all cases, the net effect of pollution and stressors on birthweight was negative. Residents in the highest-stressor quartiles consistently had the smallest infants and the lowest predicted birthweights (Supplemental Table S5. Associations between NO₂ and birthweight, however, were often dampened in high-stressor quartiles.

We confirmed that NO₂-birthweight associations were linear overall, and within each quartile of each social stressor, by examining GAMM models stratified by quartile, allowing for varying numbers of knot points. For illustration, Supplemental Fig. 1 depicts the observed linear relationship between NO₂ and birthweight by quartile of the composite Factor 1 (violence and physical disorder) using natural splines allowing for 5 knots as a sensitivity analysis.

In contrast, excluding UHF as a random effect in GAM models revealed some non-linearity, for a subset of social stressors. For example, in Supplemental Fig. 2, we observe non-linear relationships between NO₂ and birthweight in the first and second quartile of Factor 1. To account for clustering with UHF areas, we opted to retain the random effect for UHF, supporting a linear analysis.

We found mixed evidence for non-linear modification between continuous NO₂ and tract-level modifiers. In all cases, however, we did find significant decrements in birthweight with higher violence or SEP.

Adjusting for SDI did not substantially alter coefficient estimates or standard errors for NO₂ or chronic stressors. We conducted this test to determine if SDI had a more influential role on the modification of the NO₂-birthweight relationship after adjustments as shown in the Shmool et al. study. The adjustment also did not alter tests for statistical significance in NO₂–birthweight associations (not shown).

In this work, we conducted a follow-up study that focused on the modifying role of individual social stressors on the association between NO₂ and birthweight in NYC [7]. There is substantial prior evidence linking air pollution to adverse prenatal and perinatal health [3, 39], and several studies have examined modification by SEP. Though, it remains unknown, however, which specific social stressors explain SEP-related susceptibility. Among singleton births in NYC, we found significantly lower birthweights in higher-traffic pollution, higher-stressor communities. Contrary to hypotheses, however, NO₂-birthweight associations were weaker in high-stressor communities, particularly for stressors related to violent crime (the composite Factor 1, assault, murder, robbery, and low perceived safety). This result suggests that, although birthweights were lower in high-stressor communities, those low birthweights were not attributable to NO₂, but were rather more directly related to social conditions themselves. At particularly high levels of some chronic stressors, some ‘saturation’ effect may occur, wherein birthweights are already very low, due to social conditions themselves, and marginal effects of air pollution are no longer observable [15]. In other words, very high stressor exposures may overwhelm any potential effects of pollution, rendering air pollution impacts less evident at very high stressor exposures.

This work provides two key contributions to the literature. First, SEP is a complex construct, including a broad range of social stressors which may act very differently in shaping health outcomes, and disentangling these components is key to identifying effective interventions. Second, interactions are not necessarily linear, nor uniform, as threshold and saturation effects – combined with unique joint distributions between social and environmental exposures in each setting – may lead to very different pollutant-outcome associations along the continuum of any given chronic stressor.

Our results suggest a predominant role for community violence in shaping SEP-related susceptibility, in this urban setting, in keeping with much of our prior work [38, 40,41,42]. Both qualitative [43, 44] and quantitative [6, 45] work suggest that community violence, and fear thereof, is a critical component of urban SEP-related susceptibility. Prior studies have documented negative impacts of community violence on term birthweight; Masi et al. found up to an average reduction of 43 g in birthweight for a one-unit increase violent crime rate in Chicago, using census tract-level measures of violent crime [6]. Evidence has also linked community violence exposures during pregnancy to greater perceived stress [46], higher preterm birth rates [47], smaller gestational size [6], and pregnancy complications [48].

We also found that one indicator of neighborhood built environment quality (unclean parks) significantly modified NO₂-birthweight associations. Degraded built environments have been associated with lower birthweights [49] and worse birth outcomes [50], and evidence has shown that, particularly among women, neighborhood physical environments sends important signals regarding safety [51, 52]. As such, this finding may plausibly further support a central role for perceived safety in explaining SEP-related susceptibility among pregnant urban women.

These findings were made possible by revisiting the Shmool et al. study and applying similar methods to characterize the role of individual community stressors as risk factors and effect modifiers. In the Schmool study, a principal component analysis identified seven highly-correlated social stressors: the population percentage of residents with a college degree, unemployment rates, residential crowding, management or professional occupations, households below 200% of the Federal Poverty Level (FPL), households receiving public assistance, and non-White racial composition. These stressors were used to develop an area-level composite index for socio-economic deprivation. In this study, we adopted a more comprehensive and expansive framework to identify individual stressors that significantly influence the relationship between nitrogen dioxide (NO₂) levels and birthweight. These findings may have implications for developing more targeted and focused interventions that address domains related to socio-economic position, crime and violence, and physical disorder. A major advantage of this work is our access to exhaustive births data throughout NYC, capturing neighborhoods greatly varying in air pollution and social stressor exposures. In addition, we used highly spatially-resolved air pollution exposure estimates, reducing exposure misclassification. Also, multiple spatial predictors for chronic stressors were accessed through publicly available sources specific to a densely populated city. We were able to geocode and link detailed birth data across the study area. Also, we were able to adjust for several traditionally used maternal risk factors related to inter-uterine growth restriction.

There are also some limitations to this work. UHF, as a spatial unit (n = 34 across NYC), does not capture very fine-scale neighborhood influences. In tract-level sensitivity analyses, we found significant main effects for violent crime or SDI terms, though mixed evidence for non-linear interactions.

We did not simultaneously adjust for multiple social factors (except for sensitivity-testing using SDI), nor incorporate measures of race-based racial segregation, as doing so could plausibly dampen observed effects of specific social stressors, given the role of historical redlining and segregation in shaping current-day spatial distributions in social stressors. Gestation-averaged NO₂ concentration was used as an indicator of NO₂ exposure throughout pregnancy, following prior findings in this data that trimester-specific exposures did not yield meaningfully different results [7]. The analytic dataset is restricted to term infants, as such, associations of exposures with preterm birth could not be assessed. We speculate that restricting term births may bias findings toward the null since severely growth restricted fetuses are often born preterm [53]. Finally, the data is over a decade old, though the observed associations between pollution and birth outcomes, as well as modification by social stressors, remain salient. Future analyses will address several of the limitations above, focusing on that subset of social stressors available at finer geographic scales.

Our results suggest a predominant role for violence-related stressors in explaining SEP-related susceptibility to pollution—in this urban U.S. setting—and point to complex non-linearities in interactions among social stressors and pollutants. Further research in large datasets is needed to elucidate non-linear interactions among social stressors and pollution exposures.

Download references

RR and JC developed the analysis methodology. JC curated the exposure and outcome data. RR performed the analyses, wrote the original draft of the manuscript, and generated all figures and tables. JC provided funding and supervision throughout the project. JC and HB critically revised the manuscript for intellectual content. All authors read and approved the manuscript.

Correspondence to Richard V. Remigio.

The authors declare no competing interests.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions