Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Jun 7.
Published in final edited form as: Int J Hyg Environ Health. 2022 Jun 7;243:113991. doi: 10.1016/j.ijheh.2022.113991

Air pollution, children’s academic achievement and the potential mediating role of preterm birth

Arin A Balalian a, Katharine H McVeigh b, Jeanette A Stingone a,*
PMCID: PMC9541921  NIHMSID: NIHMS1840372  PMID: 35688002

Abstract

Introduction

Previous research has observed relationships between higher prenatal exposure to air pollutants and neurodevelopmental and academic outcomes later in childhood. Identifying intermediate outcomes mediating this relationship would inform prevention and intervention efforts. We aimed to investigate if previously observed associations between prenatal exposure to common urban air pollutants, diesel and perchloroethylene, with performance on third grade standardized tests were mediated through increased risk of preterm birth.

Methods

Data from the 1994–1998 birth cohorts within the New York City Longitudinal Study of Early Development were included in this analysis. Exposure was determined by linking the mother’s residence at the time of delivery to the U.S. EPA’s 1996 National Air Toxic Assessment of estimated ambient concentrations of diesel and perchloroethylene. Children’s third grade standardized math and language tests were used as the markers for academic achievement. Missing data on covariates were imputed, while participants with missing information on gestational age and test scores were excluded. Linear regression models and causal mediation analysis were used to examine potential mediation by preterm birth.

Results

In total, 187,723 and 196,122 participants were included in language and math analyses, respectively. Children with exposure to the fourth quartile of diesel or perchloroethylene had approximately 0.03 (95%CI: 0.02, 0.04) lower math z-scores when compared to individuals with exposure in the first quartile, although there was no consistent decreasing trend in math z-scores over increasing quartiles of diesel or perchloroethylene. We did not find evidence of mediation by preterm birth or exposure–mediator interaction in our models.

Conclusion

We did not find evidence that observed relationships between exposure to common urban air pollutants and test z-scores in childhood were mediated through an increased risk of preterm birth. This suggests other pathways between early exposure to air pollution and neurodevelopment should be investigated with causal mediation approaches.

Keywords: neurodevelopment, air toxics, adverse birth outcomes, mediation

Introduction

Air quality in urban areas is affected by emissions from various human-made sources, including mobile, stationary, and indoor sources. These emissions include both traffic-related air pollutants (TRAP) and air pollutants from other sources. Prenatal exposures to high levels of criteria pollutants, both from TRAP and non-TRAP sources (Karner et al. 2010), have been associated with poorer neurodevelopmental outcomes (Guxens et al. 2014; Ha et al. 2019). While a number of studies have demonstrated that exposure to air pollutants in the prenatal period is negatively associated with neurodevelopment or early childhood behavior (Cowell et al. 2015; Guxens et al. 2014; Ha et al. 2019; Perera et al. 2006; Ren et al. 2019; Sentís et al. 2017; Xu et al. 2016), others have found null or weak associations with these outcomes (Harris Maria et al. 2015). Some differences in findings across the studies could be explained by differences in pollutants that were investigated, such as focusing only on TRAP or non-TRAP pollutants.

We previously explored whether combined prenatal exposure to both TRAP and non-TRAP pollutants was associated with children’s academic achievement in third grade (Stingone et al. 2016). In that work, diesel particulate matter (PM) was used as a marker of TRAP. Diesel fuel is used by a number of highway vehicles and engines, and its emissions are a mixture of solid and gaseous materials. The diesel particulate matter refers to the solid visible emissions, composed of black carbon and other organic compounds such as polycyclic aromatic hydrocarbons, benzene, formaldehyde, acetaldehyde, acrolein, and 1,3-butadiene.(Wichmann 2007) Perchloroethylene is a non-TRAP chlorinated solvent that is used in the dry-cleaning industry and is prevalent in urban communities. (EPA 2010; Stingone et al. 2016). When examining the combination of exposure to diesel PM and perchloroethylene, we observed that high exposure to both diesel PM and perchloroethylene was associated with lower math scores among third graders in New York City when compared to children with low exposure to both pollutants (Stingone et al. 2016). This was consistent with previous literature suggesting exposure to air pollutants can hinder neurodevelopment and subsequent academic outcomes. However, the mechanisms of these observed relationships have not been explored.

One possible mechanism for the association between prenatal exposure to air pollution and neurodevelopmental outcomes could be through the greater risk of adverse birth outcomes such as preterm birth (Darrow et al. 2009; Li et al. 2017; Liu et al. 2019). Preterm delivery, defined as a birth before the completion of 37 weeks of gestation, is a significant adverse birth outcome associated with both air pollution exposure in pregnancy and adverse neurodevelopmental outcomes. Preterm birth and other adverse birth outcomes have been shown to be associated with higher air pollution exposure during pregnancy (Darrow et al. 2009; Li et al. 2017; Rich et al. 2009; Tan et al. 2017). Previous studies have demonstrated that preterm birth is associated with lower intellectual (Chan et al. 2016) and educational outcomes in early childhood (Lipkind et al. 2012; Morse et al. 2009; Quigley et al. 2012; Williams et al. 2013). A previous study among a cohort of children living in New York City found that children born preterm had higher odds for needing special education and lower standardized test scores compared to full-term children (Lipkind et al. 2012). These associations are biologically plausible as it is hypothesized preterm birth can alter dynamic brain growth and development, which normally occurs during the last trimester (Baron et al. 2012).

Given the association between air pollutants and preterm birth, as well as preterm birth with intellectual and educational attainment, we aimed to investigate if the associations between high perinatal exposure to diesel emissions and perchloroethylene and standardized test scores in the third grade are mediated through preterm birth. We were specifically interested in assessing natural direct effects, characterized as the change in standardized test z-scores associated with changing the exposures levels of diesel PM and perchloroethylene from high to low, while controlling preterm birth at the levels naturally observed at high or low levels of diesel PM and perchloroethylene. We specifically chose to estimate natural direct/indirect effects to describe the mechanism of mediation, as opposed to hypothesizing intervening directly on the mediator for all participants (controlled direct effects) or hypothesize a stochastic intervention (stochastic effects).(Rudolph et al. 2019) Due to the skewed nature of air pollution exposures and to facilitate mediation analyses, we dichotomized our air pollution exposures as high/low. We were also interested in assessing natural indirect effects (NIE), characterized as change in standardized test z-scores while setting exposure to pollutants as exposed, and intervening to change preterm births to full-term births. There are multiple existing public health and clinical interventions to reduce preterm birth. (Ricklan et al. 2021) We aimed to investigate whether an intervention to reduce the risk of preterm birth would also impact the relationship between elevated exposure to air pollution and neurodevelopment, as interventions to reduce air pollution exposure are often difficult to make at the individual level and require societal intervention. Assessing the potential for mediation by preterm birth would inform if the potential harmful effects of exposure to diesel PM and perchloroethylene could be mitigated to some extent by interventions to reduce the risk of preterm births.

Methods

Study Population

The Longitudinal Study of Early Development is an administrative data linkage of public health records from children living in New York City in the United States (Pfeiffer et al. 2012). Researchers at the New York City Department of Health and Mental Hygiene (NYCDOHMH) linked data across NYCDOHMH’s birth and death certificate registries, the Early Intervention program, the Lead Poisoning Prevention Program registry, and the New York City Department of Education’s administrative, special education and testing databases (Pfeiffer et al. 2012). Our sample included children born from 1994–1998 (1) who were born in New York City and whose mother’s primary residence was in New York City; (2) whose address provided on the birth record was successfully geocoded; and (3) were enrolled in third grade at a public elementary school in New York City by 2007 (Stingone et al. 2016). The data systems to create LSED did not change over the study years. Thus, no harmonization across individuals from different birth years was required to create the LSED database. Children who had been born as part of multiple births were excluded, as were those who had congenital anomalies reported on their birth record. We also excluded children who did not have information on preterm birth, the mediator of interest in this study. Thus, a subset of the participants of the original study (Stingone et al. 2016) were included in the current analysis (Supplementary Figure 1). This project was reviewed and approved by the NYCDOHMH and the Columbia University Institutional Review Boards.

Assessment of Outcomes

We used math and English Language Arts (ELA) standardized test scores as a measure of children’s academic attainment and a distal marker of children’s neurodevelopment and cognition. The math and ELA standardized tests were administered to all children enrolled in third grade in public schools in New York City and were designed based on the third-grade academic curriculum to assess children’s performance. The tests’ scores range from 330 to 770 on a continuous scale. To use z-scores as our primary outcomes, we standardized test scores by birth year within the full New York City population as a proxy for the year the test was administered to ensure consistency in scores across the years. Because of the deidentification practices, the school year was removed from the LSED database. However, in New York City children are enrolled in school based on birth calendar year, thus we do not anticipate major differences between school year and birth year.

Assessment of Exposures

Details about the assessment of diesel PM and perchloroethylene were described previously (Stingone et al. 2016). Ambient estimated concentrations of diesel PM and perchloroethylene were obtained from the 1996 U.S. EPA’s National Air Toxic Assessment (NATA) (EPA 2016), the time period closest in time to our study population. NATA evaluates ambient air toxics in the United States periodically based on emissions inventories and simulation techniques (EPA 2016) These assessments are typically released every three years, with some exceptions. Estimates of diesel PM and perchloroethylene were assigned certainty ratings of “medium” or higher for both emissions certainty and modelling certainty by the NATA assessment. This suggests the modeled estimates are generally reflective of concentrations in the ambient air.(EPA 2016) We linked the 1996 NATA estimates of diesel PM and perchloroethylene concentrations in New York City census tracts and assigned them to each child based on the census tract corresponding to the address provided at delivery. We categorized the raw concentrations of diesel PM and perchloroethylene to quartiles to account for skewed distributions of chemicals and minimize the influence of outliers. The majority of exposure variability is within the highest quartile. The categorization of the exposure and comparison between highest and lowest quartiles provided meaningful contrast between high and low levels of exposure while facilitating the analysis of direct and indirect effects in mediation analysis.

Assessment of Mediator and Confounders

Preterm birth, the mediator of interest in our analysis, was defined as a birth that occurred before the 37th week of gestation. We used preterm birth as opposed to the continuous metric of gestational age as most clinical interventions focus on the prevention of preterm birth, not the general extension of gestation among all births. In order to assess the natural direct and indirect effects of diesel PM and perchloroethylene on math and language scores, strong assumptions about presence of confounding must be met (VanderWeele and Vansteelandt 2009). These assumptions include a) no unmeasured exposure–outcome confounders, b) no unmeasured mediator–outcome confounders, c) no unmeasured exposure–mediator confounders, and d) no mediator–outcome confounder affected by exposure (VanderWeele and Vansteelandt 2009). We identified the potential confounders through meticulous assessment of literature and used directed acyclic graphs (DAGs) to evaluate their role as potential confounding variables (Supplementary Figure 2, Supplementary Text Box 1). We included maternal education, occupation, marital status, race/ ethnicity, nativity, and season of birth as potential confounders of the diesel PM/perchloroethylene and preterm birth. The confounders for the exposure–mediator–outcome model were: maternal education, occupation, marital status, race/ethnicity, nativity, primary financial means for delivery charges, possession of medical risk factors (gestational or chronic diabetes, pregnancy-associated or chronic hypertension, preeclampsia, eclampsia, previous preterm or small-for-gestational-age delivery), tobacco or alcohol use, age, and total number of live births; and child’s lead exposure (represented by the maximum venous blood lead concentration recorded in the lead poisoning prevention registry). Race/ethnicity was considered a confounder as a proxy for racism and other social factors that lead to disproportional exposure to environmental air pollutants and disparities in both adverse birth outcomes and neurodevelopment.

Maternal education was operationalized as a continuous variable by the number of years of education received. The borough of residence at delivery was also considered to control for neighborhood specific characteristics. Season of birth was calculated based on the date of birth and divided into December through February, March through May, June through August and September through November. Alcohol and tobacco use were considered present if the mother reported any consumption of these products on the birth certificate.

The neighborhood deprivation index (NDI) was defined and calculated for each census tract to adjust for neighborhood-level socioeconomic factors that could co-vary with ambient levels of diesel PM, perchloroethylene, and preterm birth as well as neurodevelopmental outcomes. The method used to calculate the NDI has been described elsewhere (Messer et al. 2006; Savitz et al. 2013). Briefly, this variable is calculated based on a principal component analysis of variables within the US Census data including “percent with college degree, percent unemployment, percent management/professional occupation, percent residential crowding, percent below 200% of the federal poverty line, percent of households receiving public assistance and percent of individuals identifying as a non-White race, and percent of linguistically isolated households.” (Messer et al. 2006; Savitz et al. 2013).The resulting NDI is a standardized metric, with a mean of 0 and a standard deviation of 1. Higher values indicate greater neighborhood-level deprivation. (Messer et al. 2006; Savitz et al. 2013).

We created a joint variable representing maternal race/ethnicity and nativity comprising the following categories to ensure adequate sample size within the categories of this variable: US-born White non-Hispanics, US-born Black non-Hispanics, US-born Hispanics, foreign-born White non-Hispanics, foreign-born Black non-Hispanics, foreign-born Hispanics, and all Asians. We collapsed the category of foreign-born and US-born participants identifying as Asian or other race, due to the low number of observations in the US-born Asian and “other” race categories. We performed sensitivity analyses to assess the effect of this categorization choice by rerunning analyses including only foreign-born Asians and excluding the other, smaller groups.

Missing Data

We used the MICE package in R (version 3.6.0) to impute missing values for lead concentration (21.84%), mother’s total live births, and maternal occupation using single-chain MARKOV chain Monte Carlo methodology, as previously described (Yuan 2011). We used maternal nativity, maternal occupation, maternal race and ethnicity, marital status, maternal education, NDI, the season of birth, maternal age, age of residential building, borough of residence, and birthweight to impute the missing values using 20 iterations between each imputation to run five imputations.

Statistical Analysis

We assessed the effects of diesel PM and perchloroethylene on test scores separately to account for the potential of different mediating relationships with preterm birth. We fit general linear models to assess the relationship between exposure to quartiles of diesel PM and perchloroethylene and test z-scores. We further assessed the relationship between exposure to these pollutants and the odds of preterm birth using logistic regression models. Finally, we used the mediation package in R (Tingley et al. 2014) to assess whether the association between the pollutants and math and ELA test z-scores is mediated by preterm birth. Natural direct and indirect effects, were calculated by comparing the fourth quartile to the first quartile of exposure as the reference category. In addition, the proportion of the total effect of pollutants on z-scores mediated by the preterm birth, was calculated after 1000 Monte Carlo draws for quasi-Bayesian estimation. The robust standard errors and 95% confidence intervals for natural direct and indirect effects were estimated. We identified the confounders from the constructed DAGs and methods to balance bias and variance of subsequent effect estimates (Greenland et al. 2016). Finally, we assessed exposure–mediator interaction, as well as borough of mother’s residence during pregnancy in the pollutants-outcome model. An a priori alpha level of 0.05 was chosen to assess the presence of statistical interaction.

Our primary mediation analysis only includes children in the top and bottom quartiles of exposure. To use the full study population, we conducted a sensitivity analysis where we dichotomized exposures at the 75th percentile to examine if our models were sensitive to excluding individuals in the two middle quartiles. The 75th percentile was chosen due to the skew in the exposure data. All analyses were repeated using this new exposure contrast.

Results

Sociodemographic Characteristics of Study Population

Demographic characteristics of the study population, overall and stratified by birth outcome, are presented in Table 1. There were a number of demographic differences between children born term versus preterm. Unmarried women made up a greater proportion of women who delivered at full-term when compared to women who delivered pre-term. Children who were born preterm tended to have lower math and ELA test z-scores compared to those who were born to term. The neighborhood deprivation index score was also higher among children born preterm than their full-term counterparts. Pregnancy complications were more commonly reported for children born preterm than children born full-term. The mothers of participants with complete data on preterm birth and ELA and math scores tended to have a higher level of education (11.8 vs. 10.2), be exposed to lower concentrations of diesel PM and perchloroethylene and be more likely to work outside the home (Supplementary Table 1) than children with incomplete data. Distributions of diesel PM and perchloroethylene in the analytic population are provided in Supplementary Table 2. After imputation and excluding observations with missing values on mediators and outcomes, a total of 187,723 and 196,122 individuals were included in the ELA and math models respectively.

Table 1.

Sociodemographic characteristics of study participants before imputation of missing variables [n (%) or mean (std)], New York City Longitudinal Study of Early Development Birth Years 1994–1998

Overall (n = 187,401) Term Birth (≥37 weeks) (n = 171,828) Preterm birth (<37 weeks) (n = 15,573)
Maternal age at birth (years) 27.3 (6.42) 27.3 (6.38) 27.7 (6.82)
Total live deliveries including current pregnancy
Maternal education (years) 11.8 (2.6) 11.9 (2.7) 11.7 (2.5)
Neighborhood deprivation index 0.0 (1.0) 0.0 (1.0) 0.2 (1.0)
Perchloroethylene (μg/m3) 0.8 (0.4) 0.8 (0.4) 0.8(0.4)
Diesel particulate matter (μg/m3) 9.5 (8.7) 9.5 (8.7) 9.5 (8.2)
Math z-score 0.0 (1.0) 0.1 (1.0) −0.2 (1.0)
Language z-score 0.0 (1.0) 0.0 (1.0) −0.2 (1.0)
Highest venous blood lead level test (μg/dL) 4.7 (3.5) 4.7 (3.5) 4.8 (3.6)
Borough of residence (%)
 Manhattan 27174 (14.5) 24730 (14.4) 2444 (15.7)
 Bronx 37913 (20.2) 34523 (20.1) 3390 (21.8)
 Brooklyn 67310 (35.9) 61597 (35.8) 5713 (36.7)
 Queens 44947 (24.0) 41657 (24.2) 3290 (21.1)
 Staten Island 10057 (5.4) 9321 (5.4) 736 (4.7)
Maternal occupation (%)
 Works at home only 90578 (48.3) 82996 (48.3) 7582 (48.7)
 Student 16389 (8.7) 14982 (8.7) 1407 (9.0)
 Not employed 18392 (9.8) 16697 (9.7) 1695 (10.9)
 Other occupation 57592 (30.7) 53048 (30.9) 4544 (29.2)
 Missing 4450 (2.4) 4105 (2.4) 345 (2.2)
Maternal marital status—Unmarried 78563 (41.9) 73344 (42.7) 5219 (33.5)
Nativity & race
 Foreign-born Black non-Hispanic 22523 (12.0) 20491 (11.9) 2032 (13.0)
 Foreign-born White non-Hispanic 10051 (5.4) 9522 (5.5) 529 (3.4)
 Foreign-born Hispanic 39328 (21.0) 36818 (21.4) 2510 (16.1)
 US-born Black Hispanic 40282 (21.5) 35667 (20.8) 4615 (29.6)
 US-born White Hispanic 21796 (11.6) 20414 (11.9) 1382 ( 8.9)
 US-born Hispanic 32276 (17.2) 29140 (17.0) 3136 (20.1)
 Asian and Other (US- or foreign-born) 21145 (11.3) 19776 (11.5) 1369 (8.8)
Season of birth
 Winter 46199 (24.7) 42255 (24.6) 3944 (25.3)
 Spring 46594 (24.9) 42757 (24.9) 3837 (24.6)
 Summer 47918 (25.6) 43920 (25.6) 3998 (25.7)
 Fall 46690 (24.9) 42896 (25.0) 3794 (24.4)
Any complication during delivery—Yes (%) 65137 (34.8) 57634 (33.5) 7503 (48.2)
Alcohol use during pregnancy 1071 (0.6) 791 (0.5) 280 (1.8)
Yes (%)
Reported tobacco use during pregnancy 10727 (5.7) 9163 (5.3) 1564 (10.0)
Yes (%)
Quartiles of Diesel PM
 1st quartile (<6.35μg/m3) 46988 (25.1) 43412 (25.3) 3576 (23.0)
 2nd quartile (6.35–7.67 μg/m3) 46939 (25.0) 42717 (24.9) 4222 (27.1)
 3rd quartile (7.67–9.74 μg/m3) 46702 (24.9) 42853 (24.9) 3849 (24.7)
 4th quartile (9.74–140 μg/m3) 46772 (25.0) 42846 (24.9) 3926 (25.2)
Quartiles of perchloroethylene (%)
 1st quartile (<0.58 μg/m3) 46869 (25.0) 43345 (25.2) 3524 (22.6)
 2nd quartile (0.58–0.68 μg/m3) 46974 (25.1) 42764 (24.9) 4210 (27.0)
 3rd quartile (0.68–0.84 μg/m3) 46740 (24.9) 42972 (25.0) 3768 (24.2)
 4th quartile (0.84–9.20 μg/m3) 46818 (25.0) 42747 (24.9) 4071 (26.1)
Open in a new tab

Association of Pollutants with Children’s Test Z-Scores.

For both diesel PM and perchloroethylene, we found that individuals with exposure in the fourth quartile had 0.03 lower math z-scores (95%CI: 0.02, 0.04) than children in the first quartile of exposure, although there was no consistent decreasing trend in math z-scores over increasing quartiles of diesel or perchloroethylene. (Table 2) Results for ELA scores were also consistent across pollutants with a null relationship observed between higher exposure and ELA test scores. The results did not change when we repeated the analysis on complete cases (Supplementary Table 3), or when we excluded Asians born in the US and the participants identified as other races rather than combine them with Asians born outside the US into a single category (data not shown).

Table 2.

Estimated effect of neighborhood pollutant concentrations on math and language test z-scores, calculated from linear models New York City Longitudinal Study of Early Development Birth Years 1994–1998

Language z-score Math z-score
Coefficient ab Estimates (95% CI) Estimates (95% CI)
Model 1 c Diesel PM 1st quartile REF
Diesel PM 2nd quartile 0.02 (0.01,0.03) 0.01 (0.00, 0.02)
Diesel PM 3rd quartile 0.01 (0.00, 0.02) 0.01 (0.00, 0.02)
Diesel PM 4th quartile 0.01 (0.00, 0.02) −0.03 (−0.04, −0.02)
Model 2 c Perchloroethylene 1st quartile REF
Perchloroethylene 2nd quartile 0.04 (0.03, 0.05) 0.02 (0.01, 0.03)
Perchloroethylene 3rd quartile 0.03 (0.02, 0.04) 0.03 (0.02, 0.04)
Perchloroethylene 4th quartile 0.01 (0.00, 0.02) −0.03 (−0.04, −0.02)
Observations 187723 196122c
Open in a new tab

Abbreviations: CI (Confidence Interval); PM (Particulate Matter); REF (Reference Category)

a

Estimates represent change in the outcomes’ z-scores associated with exposure level compared to reference.

b

All the models were adjusted for neighborhood deprivation index, maternal education, child’s highest venous lead level, maternal occupation status, maternal marital status, and maternal nativity/race/ethnicity

c

Separate linear models were fit for diesel particulate matter and Perchloroethylene as exposure variables. The sample size for these models varied as the language and math test z-scores were not imputed in corresponding models.

Mediation Analysis.

The results of mediation analysis are illustrated in Figure 1. Final models did not include interaction terms, as we did not find any evidence of interaction between exposure and mediator. Within all mediation models, the exposure contrast examined was comparing the highest quartile to the lowest quartile of exposure. The natural indirect effects were consistently null when examining the different pollutants and both ELA and math test scores. Our findings did not change in complete case analysis (Supplementary Figure 3) or when we included only Asians born outside the US and excluded Asians born in the US and participants identified as other races from the combined category due to their small sample size (Supplementary Figure 4). The proportion mediated did not exceed 1% in any of the models.

Figure 1.

Figure 1.

Open in a new tab

Estimated natural direct, indirect and total effects and 95% confidence intervals for estimated effects of highest quartile of diesel PM and Perchloroethylene compared to lowest quartile on math and language test z-scores, New York City Longitudinal Study of Early Development Birth Years 1994–1998. Dotted line indicates null value. All models were adjusted for neighborhood deprivation index, maternal education, child’s highest venous lead level, maternal occupation status, maternal marital status, maternal nativity/race/ethnicity, season of birth, maternal age, complications during labor, maternal total live births. Abbreviations: PERC: Perchloroethylene

In a sensitivity analysis, we repeated the mediation analysis, excluding children who participated in the early intervention program in New York City. We hypothesized that children who were identified and subsequently received developmental interventions through the early intervention program would have higher academic test scores. However, the findings were not different from the results presented. Also similar to our primary results, we did not see evidence of mediation when we dichotomized exposures at the 75th percentile. (Supplementary Figure 5) However, the estimates for the direct effects were slightly larger in magnitude.

We further explored if exposure was different among the children with missing information on preterm birth (n = 1995; Supplementary Table 1) and with missing data on standardized test scores (Supplementary Table 1), suggesting the potential for selection bias. There were 11,729 children with missing data on ELA tests. These children were exposed to higher levels of air pollutants and were more likely to be born to a mother who was foreign-born Hispanic. (Supplementary Table 1).

Discussion

In this study, we sought to determine if the association between exposure to common urban air pollutants and standardized test scores is mediated by preterm birth. This research built upon existing literature documenting relationships between air pollution, preterm birth and neurodevelopment as measured by school achievement (Darrow et al. 2009; Lipkind et al. 2012; Stingone et al. 2016). However, we did not find any evidence of mediation by preterm birth in our analysis.

Consistent with the previous analysis(Stingone et al. 2016), we did observe small effects of exposure to the highest quartile of pollutants on the math z-scores. Given our sample size, we should have been able to see some level of mediation, even with the small main effect, if it were present in the data. As shown in in Table 1, we did not see a strong association between exposure to air pollutants and preterm birth in our sample. This likely explains the lack of mediation observed. Because of the right skewness of the exposure data, the majority of the variability of exposure was in the highest quartile. This could explain why a consistent decreasing trend in math z-scores was not observed over increasing quartiles of exposure. This observed association, although small in absolute magnitude, is likely meaningful at the population level, as it could shift the population distribution of the test outcomes resulting in a larger number of individuals within the lower tail (Rose 2001).

Our results are supported by a small number of studies exploring the potential mediation of the relationship between air pollutants and neurodevelopment through birth outcomes. Perera et al. (2006) found that the association of prenatal exposure to polycyclic aromatic hydrocarbons with neurodevelopmental outcomes was not mediated by weight or head circumference at birth. (Perera et al. 2006). Another study found that gestational age did not mediate the association between criteria air pollutants and autism spectrum disorders (Becerra et al. 2013). The studies mentioned above assessed mediation by a difference method, where the exposure coefficient in the model with the mediator is compared to the exposure coefficient in the model without the mediator and a reduction of the estimate in the former model compared to the latter indicates mediation (VanderWeele and Vansteelandt 2009; VanderWeele 2016). In contrast, we conducted a causal mediation analysis that allowed for exposure and mediator interaction and causal interpretation of the effect estimates (Tingley et al. 2014; VanderWeele and Vansteelandt 2009; VanderWeele 2016). Although using different methods, we also did not find evidence of mediation of the relationship between urban air pollutants and a marker of neurodevelopment through preterm birth. As acknowledged above, the relationship between the specific air pollutants studied and preterm birth was not strong in our sample. The use of study populations which experience greater risks of adverse birth outcomes associated with air pollutant exposure or a focus on pollutants that are more strongly associated with preterm birth may help address remaining questions about the role of adverse birth outcomes in neurodevelopmental deficits associated with air pollution.

Aside from triggering preterm birth, oxidative stress in the brain is another possible mechanism for the association of these pollutants with neurodevelopmental outcomes (Calderón-Garcidueñas et al. 2002). It has been suggested that the pollutants can translocate through the olfactory nerve to the brain (Oberdorster et al. 2004) and result in oxidative stress, as they do in other tissues (Beck-Speier et al. 2005; Calderón-Garcidueñas et al. 2002; Suglia et al. 2008). Studies conducted on animal models also suggest possible neuroinflammation as a result of long exposure to diesel PM (Gerlofs-Nijland et al. 2010; Levesque et al. 2011). Moreover, exposure to fine and ultra-fine particles in the prenatal period may be associated with systemic cytokine response in the mother, which could impair fetal development but not contribute to a shortened gestation (Block and Calderón-Garcidueñas 2009; Kim et al. 2014; Peters et al. 2006).

Our study had several strengths. This research leveraged an existing administrative data linkage to efficiently investigate the role of preterm birth in the relationship between air pollutants and subsequent academic achievement. This administrative linkage provides population-based data on a large, diverse group of young children. The mediation of the association between exposure to diesel and perchloroethylene and test z-scores by preterm birth has not been assessed in any other cohort of this size, adding to the novelty of the present study. The children’s academic achievement was assessed using routinely-given standardized tests, allowing for aggregation across multiple years. The use of DAGs to identify the minimally sufficient adjustment sets coupled with the use of causal mediation methods facilitates our ability to interpret our results within a causal framework.

The study had some limitations as well. The study population is from an older cohort and exposure levels have changed. Although the concentrations of diesel and perchloroethylene have changed over time (Kheirbek et al. 2013), overall patterns within NYC have stayed generally consistent. Globally the exposure to ambient air pollutants remains high (World Health Organization 2016), supporting the need for work in this area and relevance for looking at time periods in the US with higher exposure levels. Although we adjusted our models for potential confounding by including covariates identified as part of our DAG’s minimally sufficient adjustment set, we could not rule out the possibility of residual confounding due to unmeasured variables. For example, parenting behaviors, particularly in early childhood, could vary by neighborhood and could also affect children’s educational attainment(Ronfani et al. 2015). The impact of other spatially-varying factors was not considered. For example, neighborhood noise has been found to covary with ambient air pollution(Kheirbek et al. 2014)and to impact cognitive test scores in animal models(Barzegar et al. 2015). Adjusting for the NDI may account for some of these spatially-varying factors, as they could lie on the same backdoor path. Finally, the impact of other TRAP or non-TRAP co-pollutants that covary with diesel PM concentrations could also impact both mediator and outcome(Jedrychowski et al. 2015; Stingone et al. 2016; Suglia et al. 2008).

This analysis only included the children who had available measures of preterm birth and standardized test scores. We acknowledge that missing information on the mediator, and particularly on ELA tests, could have affected our findings. Our sensitivity analysis illustrated that children of foreign-born Hispanic mothers had both higher exposure to air pollutants and were more likely to be missing data on ELA tests. This could lead to selection-bias in our results, and potentially contribute to an underestimation of the associations as individuals with higher levels of exposure were excluded from final models. It also highlights another potential limitation of our study: the use of standardized tests as a marker for neurodevelopment and academic success. Achievement on these tests is affected by other factors beyond a child’s individual aptitude, including their learning environments both in and out of school and impacts of biases within the test itself. For example, previous research in NYC has shown that children from linguistically-isolated households often have lower scores on language tests (Steifel 2003). We do not have information on many of these factors that could influence test performance and may be more common among individuals exposed to higher levels if air pollution, so residual confounding remains a possible explanation for our findings.

Another possibility for selection bias could be due to the exclusion of children who did not attend public schools in NYC in the analytic sample. The children in public and private schools could differ in various sociodemographic characteristics including race and household income (Domanico 2020). Children who moved out of the city prior to school enrollment may be sociodemographically different in ways that could be associated with both exposure and outcome. Additionally, the study population included children who were born in 1994–1998. We did not observe considerable sociodemographic differences across individuals stratified by birth year (Supplementary table 4). This supports our assumptions of similar confounding structure across the population, despite different birth years.

Exposures to air pollutants were assessed by the estimated concentrations of these pollutants for the census tract in which the mothers were reported to live at the time of delivery (EPA 2016). Research is mixed about the critical window of exposure in relation to preterm birth, but it is possible that examining exposure early in pregnancy may lead to different findings of mediation by preterm birth. Most studies of residential mobility during pregnancy suggest that impacts on exposure are limited, as people tend to move short distances and/or to areas with similar exposure levels.(Bell and Belanger 2012; Chen et al. 2010) The use of area-based measures can lead to misclassification, as other locations visited by pregnant women and time spent indoors are not accounted for within the exposure assignment. In contrast to personalized estimates of pollution exposure, the use of ambient estimates as a proxy for exposure is not affected by personalized behaviors and thus not affected by confounding (Weisskopf and Webster 2017).

Conclusion

In this study, we did not find evidence for a mediating effect of preterm birth on the association between diesel PM/perchloroethylene and performance on children’s standardized math and ELA tests in third grade. This suggests other pathways between early exposure to air pollution and neurodevelopment should be investigated with mediation. The observed association between diesel PM/perchloroethylene and children’s math and ELA test z-scores, albeit small, could be meaningful at the population level as it could result in larger number of children at the lower tail of the population distribution of math/ELA standardized tests. Given the racial and socioeconomic disparities in exposure to air pollutants, it is of great importance to improve our understanding of the pathways connecting early-life exposure to adverse school outcomes, in order to identify early points of intervention.

Supplementary Material

Supplementary Material

Acknowledgements:

This study was supported in part from a grant from the NIH/NIEHS ES027022.

References

  1. Baron IS, Litman FR, Ahronovich MD, Baker R. 2012. Late preterm birth: A review of medical and neuropsychological childhood outcomes. Neuropsychology Review 22:438–450. [DOI] [PubMed] [Google Scholar]
  2. Barzegar M, Sajjadi FS, Talaei SA, Hamidi G, Salami M. 2015. Prenatal exposure to noise stress: Anxiety, impaired spatial memory, and deteriorated hippocampal plasticity in postnatal life. Hippocampus 25:187–196. [DOI] [PubMed] [Google Scholar]
  3. Becerra TA, Wilhelm M, Olsen J, Cockburn M, Ritz B. 2013. Ambient air pollution and autism in los angeles county, california. Environ Health Perspect 121:380–386. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Beck-Speier I, Dayal N, Karg E, Maier KL, Schumann G, Schulz H, et al. 2005. Oxidative stress and lipid mediators induced in alveolar macrophages by ultrafine particles. Free Radical Biology and Medicine 38:1080–1092. [DOI] [PubMed] [Google Scholar]
  5. Bell ML, Belanger K. 2012. Review of research on residential mobility during pregnancy: Consequences for assessment of prenatal environmental exposures. J Expo Sci Environ Epidemiol 22:429–438. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Block ML, Calderón-Garcidueñas L. 2009. Air pollution: Mechanisms of neuroinflammation and cns disease. Trends in Neurosciences 32:506–516. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Calderón-Garcidueñas L, Azzarelli B, Acuna H, Garcia R, Gambling TM, Osnaya N, et al. 2002. Air pollution and brain damage. Toxicologic Pathology 30:373–389. [DOI] [PubMed] [Google Scholar]
  8. Chan E, Leong P, Malouf R, Quigley MA. 2016. Long-term cognitive and school outcomes of late-preterm and early-term births: A systematic review. Child: Care, Health and Development 42:297–312. [DOI] [PubMed] [Google Scholar]
  9. Chen L, Bell EM, Caton AR, Druschel CM, Lin S. 2010. Residential mobility during pregnancy and the potential for ambient air pollution exposure misclassification. Environ Res 110:162–168. [DOI] [PubMed] [Google Scholar]
  10. Cowell WJ, Bellinger DC, Coull BA, Gennings C, Wright RO, Wright RJ. 2015. Associations between prenatal exposure to black carbon and memory domains in urban children: Modification by sex and prenatal stress. PloS one 10:e0142492. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Darrow LA, Klein M, Flanders WD, Waller LA, Correa A, Marcus M, et al. 2009. Ambient air pollution and preterm birth: A time-series analysis. Epidemiology (Cambridge, Mass) 20:689–698. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Domanico R 2020. A statistical profile of new york’s k-12 educational sector: Race, income, and religion. Manhattan Institute for Policy Research. [Google Scholar]
  13. EPA. 2010. Review of the environmental protection agency’s draft iris assessment of tetrachloroethylene:National Academies Press. [PubMed] [Google Scholar]
  14. EPA. 2016. Technology transfer network 1996 national-scale air toxics assessment. Available: https://archive.epa.gov/airtoxics/nata/web/html/natsaov.html [accessed 5/30/2019 2019].
  15. Gerlofs-Nijland ME, van Berlo D, Cassee FR, Schins RPF, Wang K, Campbell A. 2010. Effect of prolonged exposure to diesel engine exhaust on proinflammatory markers in different regions of the rat brain. Particle and Fibre Toxicology 7:12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Greenland S, Daniel R, Pearce N. 2016. Outcome modelling strategies in epidemiology: Traditional methods and basic alternatives. International journal of epidemiology 45:565–575. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Guxens M, Garcia-Esteban R, Giorgis-Allemand L, Forns J, Badaloni C, Ballester F, et al. 2014. Air pollution during pregnancy and childhood cognitive and psychomotor development: Six european birth cohorts. Epidemiology (Cambridge, Mass) 25:636–647. [DOI] [PubMed] [Google Scholar]
  18. Ha S, Yeung E, Bell E, Insaf T, Ghassabian A, Bell G, et al. 2019. Prenatal and early life exposures to ambient air pollution and development. Environmental research 174:170–175. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Harris Maria H, Gold Diane R, Rifas-Shiman Sheryl L, Melly Steven J, Zanobetti A, Coull Brent A, et al. 2015. Prenatal and childhood traffic-related pollution exposure and childhood cognition in the project viva cohort (massachusetts, USA). Environmental Health Perspectives 123:1072–1078. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Jedrychowski WA, Perera FP, Camann D, Spengler J, Butscher M, Mroz E, et al. 2015. Prenatal exposure to polycyclic aromatic hydrocarbons and cognitive dysfunction in children. Environmental Science and Pollution Research 22:3631–3639. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Karner AA, Eisinger DS, Niemeier DA. 2010. Near-roadway air quality: Synthesizing the findings from real-world data. Environmental Science & Technology 44:5334–5344. [DOI] [PubMed] [Google Scholar]
  22. Kheirbek I, Johnson S, Ito K, Matte T, Kass D, Steven A. Caputo J, et al. 2013. New york city trends in air pollution and its health consequences. [Google Scholar]
  23. Kheirbek I, Ito K, Neitzel R, Kim J, Johnson S, Ross Z, et al. 2014. Spatial variation in environmental noise and air pollution in new york city. Journal of Urban Health 91:415–431. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Kim E, Park H, Hong Y-C, Ha M, Kim Y, Kim B-N, et al. 2014. Prenatal exposure to pm10 and no2 and children’s neurodevelopment from birth to 24months of age: Mothers and children’s environmental health (moceh) study. Science of The Total Environment 481:439–445. [DOI] [PubMed] [Google Scholar]
  25. Levesque S, Taetzsch T, Lull ME, Kodavanti U, Stadler K, Wagner A, et al. 2011. Diesel exhaust activates and primes microglia: Air pollution, neuroinflammation, and regulation of dopaminergic neurotoxicity. Environmental health perspectives 119:1149–1155. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Li X, Huang S, Jiao A, Yang X, Yun J, Wang Y, et al. 2017. Association between ambient fine particulate matter and preterm birth or term low birth weight: An updated systematic review and meta-analysis. Environmental Pollution 227:596–605. [DOI] [PubMed] [Google Scholar]
  27. Lipkind HS, Slopen ME, Pfeiffer MR, McVeigh KH. 2012. School-age outcomes of late preterm infants in new york city. American Journal of Obstetrics and Gynecology 206:222.e221–222.e226. [DOI] [PubMed] [Google Scholar]
  28. Liu Y, Xu J, Chen D, Sun P, Ma X. 2019. The association between air pollution and preterm birth and low birth weight in guangdong, china. BMC Public Health 19:3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Messer LC, Laraia BA, Kaufman JS, Eyster J, Holzman C, Culhane J, et al. 2006. The development of a standardized neighborhood deprivation index. Journal of urban health : bulletin of the New York Academy of Medicine 83:1041–1062. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Morse SB, Zheng H, Tang Y, Roth J. 2009. Early school-age outcomes of late preterm infants. Pediatrics 123:e622–e629. [DOI] [PubMed] [Google Scholar]
  31. Oberdorster G, Sharp Z, Atudorei V, Elder A, Gelein R, Kreyling W, et al. 2004. Translocation of inhaled ultrafine particles to the brain. Inhal Toxicol 16:437–445. [DOI] [PubMed] [Google Scholar]
  32. Perera Rauh, Whyatt Tsai, Tang Diaz, et al. 2006. Effect of prenatal exposure to airborne polycyclic aromatic hydrocarbons on neurodevelopment in the first 3 years of life among inner-city children. Environmental health perspectives 114:1287–1292. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Peters A, Veronesi B, Calderón-Garcidueñas L, Gehr P, Chen LC, Geiser M, et al. 2006. Translocation and potential neurological effects of fine and ultrafine particles a critical update. Particle and fibre toxicology 3:13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Pfeiffer MR, Slopen ME, Curry A, McVeigh KH. 2012. Creation of a linked inter-agency data warehouse: The longitudinal study of early development. In: A research report from the new york city department of health and mental hygiene. [Google Scholar]
  35. Quigley MA, Poulsen G, Boyle E, Wolke D, Field D, Alfirevic Z, et al. 2012. Early term and late preterm birth are associated with poorer school performance at age 5 years: A cohort study. Archives of Disease in Childhood - Fetal and Neonatal Edition 97:F167. [DOI] [PubMed] [Google Scholar]
  36. Ren Y, Yao X, Liu Y, Liu S, Li X, Huang Q, et al. 2019. Outdoor air pollution pregnancy exposures are associated with behavioral problems in china’s preschoolers. Environmental Science and Pollution Research 26:2397–2408. [DOI] [PubMed] [Google Scholar]
  37. Rich DQ, Demissie K, Lu S-E, Kamat L, Wartenberg D, Rhoads GG. 2009. Ambient air pollutant concentrations during pregnancy and the risk of fetal growth restriction. Journal of epidemiology and community health 63:488–496. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Ricklan SJ, Cuervo I, Rebarber A, Fox NS, Shirazian T. 2021. Two decades of interventions in new york state to reduce maternal mortality: A systematic review. The Journal of Maternal-Fetal & Neonatal Medicine 34:3514–3523. [DOI] [PubMed] [Google Scholar]
  39. Ronfani L, Brumatti LV, Mariuz M, Tognin V, Bin M, Ferluga V, et al. 2015. The complex interaction between home environment, socioeconomic status, maternal iq and early child neurocognitive development: A multivariate analysis of data collected in a newborn cohort study. PloS one 10:e0127052. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Rose G 2001. Sick individuals and sick populations. International journal of epidemiology 30:427–432. [DOI] [PubMed] [Google Scholar]
  41. Rudolph KE, Goin DE, Paksarian D, Crowder R, Merikangas KR, Stuart EA. 2019. Causal mediation analysis with observational data: Considerations and illustration examining mechanisms linking neighborhood poverty to adolescent substance use. American Journal of Epidemiology 188:598–608. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Savitz DA, Bobb JF, Carr JL, Clougherty JE, Dominici F, Elston B, et al. 2013. Ambient fine particulate matter, nitrogen dioxide, and term birth weight in new york, new york. American journal of epidemiology 179:457–466. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Sentís A, Sunyer J, Dalmau-Bueno A, Andiarena A, Ballester F, Cirach M, et al. 2017. Prenatal and postnatal exposure to no2 and child attentional function at 4–5years of age. Environment international 106:170–177. [DOI] [PubMed] [Google Scholar]
  44. Stingone JA, McVeigh KH, Claudio L. 2016. Association between prenatal exposure to ambient diesel particulate matter and perchloroethylene with children’s 3rd grade standardized test scores. Environmental research 148:144–153. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Suglia SF, Gryparis A, Wright RO, Schwartz J, Wright RJ. 2008. Association of black carbon with cognition among children in a prospective birth cohort study. American journal of epidemiology 167:280–286. [DOI] [PubMed] [Google Scholar]
  46. Tan Y, Yang R, Zhao J, Cao Z, Chen Y, Zhang B. 2017. The associations between air pollution and adverse pregnancy outcomes in china. In: Ambient air pollution and health impact in china, (Dong G-H, ed). Singapore:Springer; Singapore, 181–214. [DOI] [PubMed] [Google Scholar]
  47. Tingley D, Yamamoto T, Hirose K, Keele L, Imai K. 2014. Mediation: R package for causal mediation analysis. [Google Scholar]
  48. VanderWeele TJ, Vansteelandt S. 2009. Conceptual issues concerning mediation, interventions and composition. Statistics and its Interface 2:457–468. [Google Scholar]
  49. VanderWeele TJ. 2016. Mediation analysis: A practitioner’s guide. Annu Rev Public Health 37:17–32. [DOI] [PubMed] [Google Scholar]
  50. Weisskopf MG, Webster TF. 2017. Trade-offs of personal versus more proxy exposure measures in environmental epidemiology. Epidemiology (Cambridge, Mass) 28:635–643. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Wichmann HE. 2007. Diesel exhaust particles. Inhal Toxicol 19 Suppl 1:241–244. [DOI] [PubMed] [Google Scholar]
  52. Williams BL, Dunlop AL, Kramer M, Dever BV, Hogue C, Jain L. 2013. Perinatal origins of first-grade academic failure: Role of prematurity and maternal factors. Pediatrics 131:693–700. [DOI] [PubMed] [Google Scholar]
  53. World Health Organization. 2016. Air pollution levels rising in many of the world’s poorest cities. Available: https://www.who.int/news/item/12-05-2016-air-pollution-levels-rising-in-many-of-the-world-s-poorest-cities [accessed 9/14/2021 2021].
  54. Xu X, Ha SU, Basnet R. 2016. A review of epidemiological research on adverse neurological effects of exposure to ambient air pollution. Frontiers in Public Health 4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Yuan Y 2011. Multiple imputation using sas software. Journal of Statistical Software 45:1–25. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material
NIHMS1840372-supplement-Supplementary_Material.docx (352.5KB, docx)

RESOURCES