Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Nov 19.
Published in final edited form as: Environ Res. 2022 Aug 27;214(Pt 4):114163. doi: 10.1016/j.envres.2022.114163

Prenatal exposure to PM2.5 and childhood cognition: Accounting for between-site heterogeneity in a pooled analysis of ECHO cohorts in the Northeastern United States

Xueying Zhang a,*, Shelley H Liu b, Mariel Geron c, Yueh-Hsiu Mathilda Chiu a,c,f, Richard Gershon d, Emily Ho d, Kathi Huddleston e, Allan C Just a,f, Itai Kloog a,f,g, Brent A Coull h,i, Michelle Bosquet Enlow j,k, Robert O Wright a,f, Rosalind J Wright a,c,f
PMCID: PMC9675417  NIHMSID: NIHMS1850097  PMID: 36030921

Abstract

Background:

Emerging studies have investigated the adverse health effects of PM2.5 using data from multiple cohorts, and results often are not generalizable across cohorts. We aimed to assess associations between prenatal PM2.5 and childhood cognition in two U.S. cohorts while accounting for between-site heterogeneity.

Methods:

Analyses included 348 mother-child dyads enrolled in the dual site (New York City and Boston) PRogramming of Intergenerational Stress Mechanisms (PRISM) cohort and in the First Thousand Days of Life (FTDL) study (Northern Virginia) participating in the Environmental influences on Child Health Outcomes (ECHO) national consortium. Residential prenatal PM2.5 exposure was estimated using a validated satellite-based model and childhood cognition was measured using the NIH Toolbox Cognition Battery at three to eight years of age. We used a log-linear model applied to contingency tables formed by cross-classifying covariates by site to examine between-site heterogeneity using 3rd trimester PM2.5 exposure, age-corrected cognition scores, and covariates potentially causing heterogeneities. Multivariable linear regression models informed by the combinability analysis were used to estimate the coefficients and 95% confidence intervals (CIs) for the association between 3rd trimester PM2.5 exposure and age-corrected cognition scores (mean = 100, SD = 15).

Results:

The log-linear model indicated that inter-study associations were similar between PRISM-NYC and FTDL, which were different from those in PRISM-Boston. Accordingly, we combined the data of PRISM-NYC and FTDL cohorts. We observed associations between 3rd trimester PM2.5 and cognition scores, findings were varying by site, childsex, and test. For example, a 1 μg/m3 increase of 3rd trimester PM2.5 was associated with −4.35 (95% CI = −8.73, −0.25) mean early childhood cognition scores in females in PRISM-Boston. In the pooled NYC + FTDL site, the association between PM2.5 and childhood cognition may be modified by maternal education and urbanicity.

Conclusions:

We found associations between prenatal PM2.5 and impaired childhood cognition. Since multi-site analyses are increasingly conducted, our findings suggest the needed awareness of between-site heterogeneity.

Keywords: Air pollution, Childhood cognition, Log-linear model, Heterogeneity test

1. Introduction

Exposure to ambient fine particulate matter (aerodynamic diameter ≤2.5 μm; PM2.5) adversely affects neurodevelopmental outcomes in children even with exposures as early as pregnancy. While mechanisms are not fully elucidated, enhanced oxidative stress (Saenen et al., 2019) Saenen Nelly et al., (2015) and inflammation (Guerra et al., 2013) are thought to play central roles. Associations between air pollution exposure and a range of early childhood neurodevelopmental outcomes have been described (Lopuszanska and Samardakiewicz, 2020; Volk et al., 2021), including impaired cognition (Bansal et al., 2021; Guxens et al., 2014; Porta et al., 2016; Peterson et al., 2022). Experimental and human MRI studies show that PM2.5 can impact multiple brain regions, including the hippocampus (Chao et al., 2017; Di Domenico et al., 2020), visual cortex (Guo et al., 2021), frontal and temporal lobes (Chen et al., 2015), and corpus callosum (Chen et al., 2015), and that these effects begin prenatally (Peterson et al., 2022). Given that each region contributes to specific cognitive features, PM2.5 exposure may be associated with disrupted programming of multiple cognitive functions. For example, studies link PM2.5 exposure with measures of global cognition (e.g., intelligence) and specific cognitive functions, such as working memory (Calderón-Garcidueñas et al., 2016), executive function (Gui et al., 2020), attention (Donzelli et al., 2019), and language (Mullen et al., 2020). In addition, recent epidemiological studies from our group and others demonstrate sex differences in the effects of prenatal PM2.5 on specific cognitive functions (Chiu et al., 2016; Lertxundi et al., 2019; Rahman et al., 2022; Sunyer et al., 2015).

To broaden generalizability of findings and increase statistical power through larger samples, which facilitates consideration of effect modification (e.g., by child sex), studies of air pollution and health outcomes often pool data across different sites. However, pooled analyses considering ambient air pollution effects on health face a number of challenges that need to be carefully considered. Adverse health effects of air pollution are often inconsistent across study regions and populations. Although potential reasons have not been fully explicated, the different sources of air pollution and varying socioeconomic characteristics across study populations can contribute to the variation because those factors could modify or confound the association between air pollution and adverse health outcomes (Samet and White, 2004). Also, ambient pollutant components can be elevated in different patterns in time and across geographic areas and impact study participants differently due to dissimilarity of a number of characteristics related to where they live, such as degree of urbanicity, proximity to major roadways and other combustible sources, and weather or climate-related factors (e.g., temperature). Moreover, many prior studies have examined associations between prenatal PM exposure and childhood cognition in the setting of relatively high exposure levels, which may not be generalizable to lower-level exposure that is more typical across many parts of the United States (U.S.) (i.e., below the U.S. Environmental Protection Agency annual standard of 12 μg/m3). Thus, examining associations between prenatal PM2.5 exposure and child health outcomes across wider U.S. geographic locations is essential to more fully understand the impact of ambient fine particulate matter exposure on neurodevelopment in children.

Researchers traditionally have investigated how different study settings and populations contribute to variations in the findings on the adverse health effects of air pollution by using meta-analysis or pooled datasets that adjust for site (Dadvand et al., 2013; Guxens et al., 2014; Parker and Woodruff, 2008; Pedersen et al., 2013; Rosa et al., 2017). These analyses, which use pooled data with a larger sample size than individual studies, have not always found stronger effects than analyses conducted for each site considered independently (Dadvand et al., 2013; Guxens et al., 2014; Rosa et al., 2017). In fact, when data are simply pooled together without addressing these between-site heterogeneities, the associations between air pollution and adverse health outcomes are often attenuated by sites that had null or opposites associations (Dadvand et al., 2013; Guxens et al., 2014; Rosa et al., 2017). We have previously demonstrated a novel approach for assessing the combinability of heterogeneous populations prior to combining their data that allows for a better understanding of underlying cohort differences and provides increased power to detect associations that would be undetected by more traditional methods for combining cohorts (Rosa et al., 2017).

The current study demonstrates the use of combinability analyses to examine the association between prenatal exposure to ambient PM2.5 and cognition in children assessed in early to middle childhood using National Institutes of Health Toolbox Cognition Battery (NIHTB-CB) (Gershon et al., 2013) in two socioeconomically and ethnically diverse longitudinal pregnancy cohorts. Further, because of the common, computerized phenotyping instrument (NIHTB-CB), we reduced heterogeneity due to variability inherent in measuring neurobehavioral domains using different instruments. One key advantage of the NIHTB-CB is that it assesses and provides scores for global cognition and for specific domains (e.g., working memory). Specifically, we conducted log-linear model analyses of multidimensional contingency tables to examine between-site heterogeneity in the associations among PM2.5 exposure, cognition summary scores, and key sociodemographic covariates. Contrasts between cognition domains indicated which domain was more affected by prenatal PM2.5 exposure.

2. Methods

2.1. Study participants

Data used in the present analyses were from two longitudinal pregnancy cohort studies participating in the national Environmental influences on Child Health Outcomes (ECHO) study (Gillman and Blaisdell, 2018) which integrated existing cohorts for harmonized data collection in a national consortium. These included the dual-site PRogramming of Intergenerational Stress Mechanisms (PRISM) study and the First Thousand Days of Life Study (FTDL) describe in greater detail below. Local Institutional Review Boards (IRBs) and the central ECHO IRB (Western IRB) reviewed all research methods and procedures.

2.1.1. PRISM

The PRISM study comprises a prospective pregnancy cohort designed to examine the effects of prenatal and early life psychosocial and environmental exposures on child developmental outcomes. Beginning in 2011, pregnant women receiving prenatal care from the Beth Israel Deaconess Medical Center and East Boston Neighborhood Health Center in Boston, Massachusetts as well as Mount Sinai Hospital in New York City, New York were recruited at mid-to late-pregnancy. Women were considered eligible if they were English or Spanish speaking, 18 years or older, and pregnant with a singleton. Exclusion criteria included maternal intake of ≥7 alcoholic drinks per week prior to pregnancy, any alcohol intake after pregnancy recognition, HIV+ status, or delivery of an infant born with congenital abnormalities that would impede ongoing participation. Written informed consent was obtained from women prior to study participation in their preferred language. At the time of these analyses, 1109 mother-child dyads, including 399 from Boston (enrolled March 2011 to October 2013) and 710 from NYC (enrolled April 2013 to February 2020) were enrolled prenatally in PRISM, with continued active follow-up. In 2016, funding was obtained to support ongoing prenatal recruitment and follow-up as part of the ECHO program (Blaisdell et al., 2021; Gillman and Blaisdell, 2018).

2.1.2. FTDL

FTDL is a prospective pregnancy cohort of family trios (mother, father, and child) originally designed to study genomic influences on child health outcomes. This study was initiated in 2012 in a large suburban hospital with a catchment area that included Washington DC and Northern Virginia. Between April 2012 to October 2019, more than 5000 children and their biological parents were enrolled. Study eligibility required both biological parents to be over 18 years of age and either Spanish or English speaking. Families were recruited in community obstetrical practices in the second trimester of pregnancy. Women completed prenatal and delivery surveys to gather demographics and assess stress, medical history, lifestyle choices, and family health history. The medical record was abstracted and pertinent diagnoses for mother and child were noted. An electronic survey system was delivered to the family every 6 months to inquire on the child’s growth and development and general health issues to include sleep and stress. Starting in 2016, participants from the ongoing FTDL study were invited to participate in the national ECHO program, with 1400 mothers consenting for themselves and their children to participate between 2016 and 2021.

2.2. PM2.5 exposure

Geocoding of residential address history was conducted by a Geographic Information System (GIS) specialist using ArcGIS as previously detailed (Brunst et al., 2018). Individual-level prenatal daily exposure levels of ambient PM2.5 were estimated using a hybrid model that combines satellite-retrieved aerosol optical depth (AOD) products from the two Moderate-Resolution Imaging Spectroradiometer (MODIS) instruments on the NASA Terra and Aqua satellites, combined with PM2.5 monitoring data and a series of spatiotemporal predictors (local meteorological factors including air temperature and relative humidity, percentage of developed area, population density, elevation, traffic density, land use type, local emissions and others), and then predicted using an extreme gradient boosting (XGBoost) modeling approach at 1 km × 1 km spatial resolution (Just et al., 2020). The modeled daily PM2.5 levels were validated through regressing daily PM2.5 measured using filter-based monitors (obtained by U.S. EPA Air Quality System and Interagency Monitoring of Protected Visual Environments Network), resulting in a robust out of sample 10-fold cross-validation (R2 = 0.87). The model demonstrates excellent predictions of withheld observations (overall RMSE of 2.10 μg/m3 and RMSE of 3.11 μg/m3 in our spatial cross-validation). As previous work by our group has observed varied effects of PM2.5 exposure during different gestational stages of pregnancy (Chiu et al., 2016; Wilson et al., 2017), we calculated trimester-average exposure by taking the arithmetic mean of daily PM2.5 concentrations for each trimester. Given prior epidemiological research in the Northeastern U.S. regions using ambient pollution exposures estimated via similar ambient modeling as well as personal monitoring suggest that traffic-related pollution exposure later in pregnancy (i.e., 3rd trimester) particularly influences cognitive outcomes (Harris et al., 2016; Perera et al., 2003), these analyses focused on PM2.5 averaged over the third trimester to demonstrate our approach to assess combinability.

2.3. NIH Toolbox Cognition Battery

Children’s cognition was assessed in early to middle childhood using the NIHTB-CB (Gershon et al., 2013) in two socioeconomically and ethnically diverse longitudinal pregnancy cohorts across three geographic regions in the Northeastern US. The NIHTB-CB was added to data collection through the ECHO program in 2018 to assesse a range of cognitive abilities considered to be important for adaptive functioning across the lifespan. The computer-based program has demonstrated good validity and test-retest reliability in both English and Spanish from ages 3 to 85 years and can be completed within 30 min (Gershon et al., 2013; Weintraub et al., 2014). During an in-person laboratory visit, children completed the NIHTB-CB on a tablet device. Staff across study sites who guided administration were trained centrally by developers of the NIHTB. Seven NIHTB-CB component tests were administered: Dimensional Change Card Sort Test (DCCS) (Zelazo et al., 2013), List Sorting Working Memory Test (List sorting) (Tulsky et al., 2014), Flanker Inhibitory Control and Attention Test (Flanker) (Zelazo et al., 2013), Oral Reading Recognition Test (Reading) (Gershon et al., 2014), Pattern Comparison Processing Speed Test (Pattern Comparison) (Carlozzi et al., 2015), Picture Sequence Memory Test (PSMT) (Dikmen et al., 2014), and Picture Vocabulary Test (PVT) (Gershon et al., 2014). After children completed the protocol, the software estimated four summary scores – cognition total composite (CTC) score (assessed by all tests listed above), early childhood composite (ECC) score (assessed by PVT, Flanker, DCCS, and PSMT), and composite scores reflecting two cognition subdomains: “crystallized” cognition composite (CCC) (knowledge-dependent based on past learning experiences; assessed by PVT and Reading tests) and “fluid” cognition composite (FCC) (capacity for new learning and information processing in novel situations, assessed by Flanker, DCCS, PSMT, List Sorting, and Pattern Comparison tests) based on the performance of the seven component tests (Akshoomoff et al., 2013). We used the age-corrected NIHTB-CB test scores, which were converted from the raw scores for each test and standardized (M = 100; SD = 15), as the primary study outcomes in our analyses. The standardization was conducted using the data collected from a nationally representative sample of test-takers, where the average scores of each age were curved to 100. A score of 115 or 85, for example, would indicate that the participant’s performance is 1 SD above or below the national average, respectively, when compared with like-aged participants. Higher scores indicate better performance. Age at which children completed the NIHTB-CB are summarized in the supplementary document, Table S1. All tests were administered to children between 3 to 8 years of age with the exception of the Reading test, which was administered to children aged 7 years and older.

2.4. Covariates and effect modifiers

We considered the following covariates based on their hypothesized association with air pollution and cognitive outcomes in childhood: gestational age at birth, maternal race and ethnicity (Non-Hispanic White, Black/Hispanic-Black, Non-Black Hispanic, and other race), maternal educational attainment, maternal age at delivery and parity defined as number of births including the index pregnancy obtained through maternal report at enrollment. We further adjusted models for child’s age at time of cognitive assessment in order to minimize the potential for residual confounding by child age. Notably, participants across study sites resided in both urban and suburban areas, which potentially impacts air pollution exposure and effects. We thus additionally included urbanicity conceptualized as the degree to which the geographical area in which the woman resided during pregnancy was an urban vs. non-urban area. The regions of urbanicity were obtained from the U.S. Census Bureau Urban Area Shapefile (Ratcliffe et al., 2016). The Census Bureau defined urban areas as cities with a population of more than 50,000 residents and adjacent satellite cities/towns with more than 2500 residents. We mapped the census tract in which women resided during pregnancy to the Urban Area Shapefile (Zhang et al., 2021). The “urban” residence was defined for those living in a census tract fully covered by the Urban Area Shapefile; otherwise, residences were classified as non-urban.

Child sex was considered as an effect modifier based on prior work demonstrating sexually dimorphic effects of prenatal exposure to ambient pollutants on child neurodevelopmental outcomes (Chiu et al., 2016; Rahman et al., 2022). The other effect modifiers were selected based on the results of heterogeneity testing described in the 2.5.2 section.

2.5. Statistical analysis

2.5.1. Pooled effects vs. site-specific effects prior to combinability analyses

We first used multivariable linear regression models to compare the associations estimated using pooled data and site-specific data. Study site (PRISM-Boston, PRISM-NYC, and FTDL) was adjusted as a confounder in the pooled model and adjusted with 2-way interaction terms between 3rd trimester PM2.5 and site when estimating the site-specific effects. We then compared the estimated pooled associations and site-specific associations for detecting the presence and magnitude of between-site heterogeneity.

2.5.2. Heterogeneity test

Following a similar method developed for birth weight (Rosa et al., 2017), we applied log-linear models to multidimensional contingency tables (Agresti and Coull, 2002; Agresti, 2003) to examine the between-site heterogeneity in the associations among PM2.5 exposure, cognition, and select covariates (maternal education, maternal age at delivery, gestational age, urbanicity), which were disproportionally distributed among sites. For tests of heterogeneity, the 3rd trimester PM2.5 and age-corrected ECC scores were selected based on the prior results comparing pooled and site-specific associations (Fig. 1). The 3rd trimester PM2.5, site (PRISM-Boston, PRISM-NYC, and FTDL), maternal education (some college or less, college graduate, or graduate degree), maternal age at delivery, gestational age, and urbanicity (urban vs non-urban) were included in the log-linear model to examine if the association of interest varied by site. Given that the log-linear model is only applicable to categorical variables, continuous (3rd trimester PM2.5, age-corrected ECC scores, maternal age at delivery, and gestational age) variables were dichotomized by a median split.

Fig. 1.

Fig. 1.

Open in a new tab

The association between 3rd trimester PM2.5 and NIH Toolbox Cognition Battery age-corrected scores in the pooled sample and by site. Coefficients and 95% confidence intervals were estimated using multivariable linear regression models, adjusting for child sex, gestational age at birth, child age at test, maternal race/ethnicity, maternal education, maternal age at delivery, study site (only for the pooled model), parity, and urbanicity. The site-specific associations were estimated using an interaction term between 3rd trimester PM2.5 and site.

Detailed steps of this log-linear modeling framework used to assess study heterogeneity have been previously reported (Rosa et al., 2017). In brief, between-site heterogeneities were coded as interaction terms between site and other study variables in the log-linear models. We started with a “saturated” model with all possible interaction terms, and then we applied log-likelihood ratio tests (LRT) to examine whether site-interaction terms significantly contributed to the model’s goodness of fit. A LRT compares the goodness of fit between one model (full model) and another model with dropped terms (reduced model). Next, we conducted LRT in a stepwise approach. That is, we gradually dropped site-interactions from high-rank to low-rank and conducted LRT after each drop. A significant LRT result indicated that the dropped site-interaction terms significantly contributed to the full model’s fit. If a significant LRT was found, we went back to the full model to identify site-interaction terms with a significant p-value (<0.1). A significant p-value indicated that the associations between variables included in the interaction term was significantly different by site. We then considered whether the between-site heterogeneity was evident. All significant site interactions were further explored with odds ratios (OR) to examine the direction of interactions by site. In the final step, we reviewed the results of the log-linear models and site-specific ORs to determine whether to pool the data of any study sites that fit the homogeneity assumption. When results of the heterogeneity test suggested sites fulfilled the homogeneity assumption, we combined the data of those sites and conducted the subsequent multivariable linear regression.

2.5.3. Multivariable linear regression informed by combinability analyses

The associations between the 3rd trimester PM2.5 and age-corrected NIHTB-CB scores were estimated using multivariable linear regression. Study site and child sex were considered in the models as effect modifiers. Additional effect modifiers were explored based on the results of heterogeneity testing.

3. Results

3.1. Descriptive analysis results

Table 1 summarizes characteristics and 3rd trimester PM2.5 exposures of the 348 mother-child dyads included in these analyses. The average age of children at the time of cognitive assessment was 7.14 (1.06) (mean ± SD) in PRISM-Boston, which was higher than in PRISM-NYC (4.22 ± 0.54) and in the FTDL study (4.29 ± 0.64). Mothers in PRISM-NYC were, on average, younger at birth and included more mothers identifying as being of an underrepresented ethnic or racial group than women from the other two sites. Among the three sites, mothers in the FTDL study had higher educational status, with more than half (54%) reporting graduate degrees. Prenatal exposure level of PM2.5 was higher in PRISM-NYC than in the other two sites. Those differences between study sites posed potential heterogeneities in the association between PM2.5 exposures and NIHTB-CB scores.

Table 1.

Mother and child participant characteristics and PM2.5 exposures, by study site.

Variable Pooled PRISM Boston PRISM NYC FTDL
N 348 63 90 195
Children’s characteristics
Age at test, mean (SD) 4.79 (1.31) 7.14 (1.06) 4.22 (0.54) 4.29 (0.64)
Male, N (%) 184 (52.87%) 38 (60.32%) 57 (63.33%) 89 (45.64%)
Gestational age, mean (SD) 38.64 (2.03) 38.99 (1.89) 38.39 (2.37) 38.63 (1.89)
Born <37 gestational week, N (%) 39 (11.21%) 8 (12.70%) 11 (12.22%) 20 (10.26%)
Mother’s characteristics
White, N (%) 193 (55.46%) 32 (50.79%) 5 (5.56%) 156 (80.00%)
Black, N (%) 72 (20.69%) 15 (23.81%) 51 (56.67%) 6 (3.08%)
Hispanics, N (%) 86 (24.71%) 23 (36.51%) 48 (53.93%) 15 (7.81%)
Maternal education, N (%)
< 12th grade 27 (7.76%) 4 (6.35%) 23 (25.56%) 0 (0.00%)
High school grad 30 (8.62%) 1 (1.59%) 25 (27.78%) 4 (2.05%)
Some college 62 (17.82%) 15 (23.81%) 32 (35.56%) 15 (7.69%)
College completed 98 (28.16%) 16 (25.40%) 8 (8.89%) 74 (37.95%)
Graduate degree 131 (37.64%) 27 (42.86%) 2 (2.22%) 102 (52.31%)
First time delivery, N (%) 131 (37.64%) 16 (25.40%) 37 (41.11%) 78 (40.00%)
Age at birth, mean (SD) 31.61 (5.26) 32.28 (5.34) 27.44 (5.60) 33.32 (3.84)
Geospatial characteristics
3rd trimester PM2.5, μg/m3 8.62 (1.51) 8.26 (1.47) 8.80 (2.09) 8.65 (1.15)
Living in urban area, N (%) 312 (89.66%) 52 (82.54%) 89 (98.89%) 171 (87.69%)
Open in a new tab

The mean and standard deviation of age-corrected NIHTB-CB scores across sites are shown in Table 2. We created a correlation graph for NIHTB-CB instrumental and composite scores (supplementary document Figure S1) to show their underlying relationships. Children in the FTDL study had higher scores for ECC than children in PRISM-NYC (p-value<0.0001) and higher List Sorting scores than children in PRISM-Boston (p-value = 0.03). Children in PRISM-Boston had higher Pattern Comparison scores than those in the FTDL study (p-value = 0.01). Children in PRISM-NYC had lower PVT scores than children in PRISM-Boston (p-value<0.0001) and in the FTDL sample (p-value<0.0001). The Reading test was not administered in PRISM-NYC and FTDL, as the children in these sites were not old enough. For the same reason, List Sorting and Pattern Comparison had a limited sample size in PRISM-NYC (N = 2 and 2, respectively). Hence, the three composite scores (CTC, CCC, and FCC) calculated using those tests have limited sample size. Due to limited sample sizes for Reading, Pattern Comparison, CTC, CCC, and FCC, we excluded those tests and composite scores from subsequent analyses.

Table 2.

NIH Toolbox age-corrected scores by study site.

Overall PRISM-Boston (N = 63) PRISM-NYC (N = 90) FTDL (N = 195) Comparison by Site
NIH Toolbox scores, by composite score typeb Boston vs. NYC Boston vs. FTDL NYC vs FTDL
N Mean (SD) N Mean (SD) N Mean (SD) N Mean (SD) p-value a p-value a p-value a
Cognition Total Composite (CTC) 36 98.61 (14.55) 36 98.61 (14.55) 0 NA 0 NA NA NA NA
Crystallized Cognition Composite (CCC) 38 103.32 (15.16) 38 103.32 (15.16) 0 NA 0 NA NA NA NA
Fluid Cognition Composite (FCC) 80 97.70 (15.53) 49 95.92 (15.92) 1 122.00 (NA) 30 99.80 (14.41) NA 0.32 NA
Early Childhood Composite (ECC) 229 104.60 (15.85) 59 102.17 (15.75) 35 97.74 (17.59) 135 107.44 (14.78) 0.41 0.06 <0.0001
NIH Toolbox scores, by test Boston vs. NYC Boston vs. FTDL NYC vs FTDL
N Mean (SD) N Mean (SD) N Mean (SD) N Mean (SD) p-value p-value p-value
Dimensional Change Card Sorting (DCCS) 145 97.34 (14.10) 61 97.44 (14.98) 45 90.11 (12.22) 39 95.15 (14.75) 0.65 0.26 0.47
Flanker Inhibitory Control and Attention (Flanker) 159 101.29 (14.35) 61 102.44 (14.77) 60 100.23 (15.60) 38 101.11 (11.54) 0.64 0.47 0.22
List Sorting Working Memory (List sorting) 86 104.65 (28.77) 52 97.75 (15.33) 2 133.50 (106.77) 32 114.06 (35.92) 0.75 0.03 0.85
Oral Reading Recognition (Reading) 38 102.03 (15.01) 38 102.03 (15.01) 0 NA 0 NA NA NA NA
Pattern Comparison Processing Speed (Pattern Comparison) 100 82.58 (17.77) 62 86.34 (18.87) 2 57.00 (4.24) 36 77.53 (13.52) 0.06 0.04 0.47
Picture Sequence Memory Test (PSMT) 247 105.50 (20.25) 61 102.20 (16.60) 6 114.50 (21.21) 180 106.32 (21.25) 0.37 0.14 0.47
Picture Vocabulary Test (PVT) 347 99.22 (15.91) 62 103.69 (16.55) 90 90.93 (15.98) 196 101.62 (14.32) <0.0001 0.64 <0.0001
Open in a new tab

Note:

a

p-values were collected for multiple comparisons using Benjamini-Hochberg.

b

Cognition total composite score is assessed by all seven tests; crystallized cognition composite is assessed by PVT and Reading tests; fluid cognition composite was assessed by Flanker, DCCS, PSMT, List Sorting, and Pattern Comparison tests; early childhood composite score is assessed by PVT, Flanker, DCCS, and PSMT.

3.2. Multivariable linear regression analysis prior to combinability

Fig. 1 shows the effect estimates for 3rd trimester PM2.5 and age-corrected NIHTB-CB scores as an example. The contrasts between cognition domains indicate which domains were more affected by the prenatal PM2.5 exposure. There was a significant association found for the 3rd trimester PM2.5 and age-corrected ECC in PRISM-Boston (ECC: coefficient = −2.99, 95% CI = [−5.73, −0.16]). The associations from the pooled analyses and the other two cohorts were null. We then conducted subsequent analyses to examine and identify the sources of between-site heterogeneity.

3.3. Results of log-linear model: combinability analyses

Table 3 demonstrates the between-site heterogeneity using the log-linear modeling framework (Rosa et al., 2017). According to the LRT p-values, 3-way site interaction terms significantly contributed to the goodness-of-fit of the models. We further assessed the p-values of all 3-way site-interaction terms and found that the associations of two pairs of study variables (3rd trimester PM2.5 vs. maternal education, maternal age at delivery vs. urbanicity) were significantly (p-value <0.1) modified by site, specifically between PRISM-Boston and the FTDL study. The results of log-linear models and LRT further indicated that the homogeneity assumption between sites was not satisfied, which was due to the varied relationships between these same two pairs of variables (3rd trimester PM2.5 vs. maternal education, maternal age at delivery vs. urbanicity) in different sites.

Table 3.

Results of heterogeneity tests.

Log-linear model description Site interactions testedb Deviance DF LRTa p-value site-interaction term with p-value < 0.10
Saturated Ref 0 0 – –
Excluded 7-way site-interaction terms S*CPEAGU 0.22 4 1.00 –
Excluded 6-way site-interaction terms S*any 5 of CPEAGU 0.38 28 1.00 –
Excluded 6- and 5-way site-interaction terms S*any 4 of CPEAGU 2.44 76 1.00 –
Excluded 6-, 5- and 4-way site-interaction terms S*any 3 of CPEAGU 49.33 136 0.89 –
Excluded 6-, 5-, 4-, and 3-way site-interaction terms S*any 2 of CPEAGU 102.58 176 0.08 Boston vs. FTDL: 3rd trimester PM2.5b and maternal education; age at delivery and urbanicity
Open in a new tab

Note:

a

LRT was compared between models dropped n-way and (n-1)-way interaction terms.

b

C: NIH Toolbox Early Childhood Composite scores; P: 3rd trimester PM2.5; E: maternal education; A: maternal age at delivery; G: gestational age; U: urban; S: study site.

Fig. 2 illustrates the site-modified associations to identify the direction of effect modification. Fig. 2A–D shows the two pairs of variables (3rd trimester PM2.5 vs. maternal education, and maternal age at delivery vs. urbanicity) where associations varied by site. In Fig. 2A, there was a decreasing trend (red dashed line) of PM2.5 as maternal education increased in PRISM-Boston. As depicted in Fig. 2B, the OR of maternal education was 0.24 (0.07, 0.82), indicating that women with graduate degrees were exposed to lower 3rd trimester PM2.5 levels than women with college and less degrees in this site. Such an association was not found in PRISM-NYC or in the FTDL sample. Fig. 2C–D shows the relationship between maternal age at delivery and urbanicity by study site (data for PRISM-NYC are not shown in Fig. 2D because only one mother was living in a non-urban area). We observed an opposite relationship between age at delivery and urbanicity in PRISM-Boston versus the FTDL study.

Fig. 2.

Fig. 2.

Open in a new tab

A) Distribution of 3rd trimester PM2.5 by maternal education across the three study sites. B) Odds ratio (OR) between 3rd trimester PM2.5 and maternal education. C) Maternal age at delivery by urbanicity across the three study sites. D) OR between age at delivery and urbanicity. Maternal education was categorized into two levels (college or less vs. > college). Continuous variables (PM2.5 and age at delivery) were categorized by a median split.

Based on the series of heterogeneity tests described above, we concluded that the inter-study associations were distinct for PRISM-Boston compared to PRISM-NYC and the FTDL study. Accordingly, we combined the PRISM-NYC and FTDL data, collectively called “NYC + FTDL” in the manuscript).

3.4. Multivariable linear regression analysis informed by combinability analyses

We estimated the association between the 3rd trimester PM2.5 and NIHTB-CB age-corrected scores using multivariable linear regression models. Site (PRISM-Boston and NYC + FTDL) and child sex were considered as effect modifiers. Fig. 3 shows the adjusted coefficients and 95% CIs stratified by site and child sex. We observed significant associations for ECC and PSMT in female children in PRISM-Boston. Specifically, a 1 unit (μg/m3) increase in 3rd trimester PM2.5 was associated with an average decrease of −4.35 (95% CI = −8.73, −0.25) and −6.96 (−12.58, −1.17) points in age-corrected scores of ECC and PSMT, respectively.

Fig. 3.

Fig. 3.

Open in a new tab

The associations between 3rd trimester PM2.5 and NIH Toolbox Cognition Battery age-corrected scores. Adjusted coefficients and 95% CIs were estimated using multivariable linear regression models. Models were adjusted for gestational age at birth, child age at test, maternal race/ethnicity, maternal education, maternal age at delivery, parity, and urbanicity. “NYC + FTDL” indicates combined data from the PRISM-NYC and FTDL sites.

In NYC + FTDL, we observed positive associations between PM2.5 and age-corrected PVT scores [1.77, (0.19, 3.31)] and marginally significant associations for PSMT [2.17, (−0.77, 5.25)] (Fig. 3) (i.e., suggesting positive effects of PM2.5 exposure on these cognitive outcomes rather than the anticipated negative effect of PM2.5). We hypothesized that the positive associations could be biased by more complex relationships between maternal education and urbanicity in these two sites as the heterogeneity test suggested that the association between maternal education and PM2.5 were dependent upon site. We explored this further by adding maternal education and urbanicity as additional effect modifiers to the NYC + FTDL site.

Fig. 4 shows the sex-specific associations between 3rd trimester PM2.5 and age-corrected scores of PSMT and PVT in NYC + FTDL sample, stratified by maternal education and urbanicity. The full results including the PRISM-Boston site are available in the supplementary material Figure S2. In Fig. 4, the association for PVT [1.79, (0.12, 3.41)] remained significantly positive in females in urban areas. We observed a significantly negative association for PVT [−6.90, (−13.5, −0.3)] in males living in non-urban areas. This pattern suggests that associations between PM2.5 and childhood cognitive outcomes may be further modified by complex relationships among maternal education and urbanicity in the NYC + FTDL pooled site.

Fig. 4.

Fig. 4.

Open in a new tab

The sex-specific associations between 3rd trimester PM2.5 and two age-corrected NIH Toolbox Cognition Battery scores (Picture Sequence Memory Test and Picture Vocabulary Test), stratified by A) maternal education and B) urbanicity. Adjusted coefficients and 95% CIs were estimated using multivariable linear regression models. “NYC + FTDL” indicates combined data from the PRISM-NYC and FTDL cohorts. Urban group was defined as women whose residential census tract were 100% covered by Urban Area Shapefile (Ratcliffe er al. 2016).

4. Discussion

While emerging studies link ambient fine particulate matter exposure over childhood with neurocognitive outcomes assessed using the NIHTB-CB in middle childhood (Cserbik et al., 2020), this is the first study to examine associations between prenatal PM2.5 exposure and performance-based measures using the NIH Toolbox in preschool and early school-aged children. Study participants were enrolled from two cohorts participating in the national ECHO study (Gillman and Blaisdell, 2018) in order to exemplify challenges that need to be consdered when pooling cohort data. A common practice when analyzing multi-site epidemiological data is to include a term for ‘site’ to account for unmeasured effects at each respective location. These analyses demonstrate that this practice should be carefully considered when site can have complex relationships with important exposure differences as well as relevant sociodemographic variables. We leveraged data from three sites in the Northeastern U.S., spanning urban and non-urban areas, to exemplify an approach to assess study heterogeneity and combinability of studies for pooled analyses to better inform how to consider site in these analyses. Prior to assessing combinability, we found significant associations between increased exposure to PM2.5 in 3rd trimester and impaired cognition in the Boston PRISM cohort site, specifically lower scores on the ECC and PSMT components assessed using the NIHTB-CB, while associations from the pre-combinability pooled analysis and in the other two cohorts considered independently were null. The combinability approach using log linear models to evaluate a high-dimensional associations for key covariates identified heterogeneous inter-study association patterns indicating similar underlying associations between the NYC-PRISM site and the FTDL northern Virginia site. Linear models combining studies according to the pooling indicated by the log linear models (i.e., combinability analysis) revealed a significant negative association between higher 3rd trimester PM2.5 exposure and ECC scores in female children born to women in the PRISM-Boston sample, but not in the combined data of the PRISM-NYC and FTDL sites. Indeed, the associations for PSMT and PVT scores were positive among children from the combined NYC + FTDL site. When this was explored further, analyses suggested that the association between prenatal PM2.5 exposure and cognitive functioning was modified by maternal education and urbanicity across the NYC and northern Virginia sites.

The heterogeneity test was applied in this study to minimize site-specific effects. For example, the estimated pooled association could be counteracted when associations between exposure and outcome are in the opposite direction in sites. Pooling data is important for studies such as the ECHO program which can take advantage of harmonized cohort data across the contiguous U.S. in future larger scale analyses. As the national ECHO cohort spans urban, suburban and rural areas, consideration of combinability will become even more important. For example, when data are available to pool multiple cohorts with fine particulate matter exposure data as well at outcome data from the NIHTB-CB, we will be able to further disentangle the suggested influence of educational status and urbanicity explored herein. After combining sites that fit the original homogeneity assumption, we further examined modifying effects of urbanicity on the relationship between PM2.5 and cognitive outcomes in the NYC and northern Virginia cohorts. These secondary analyses suggested that “Berkson’s bias” may exist when examining the association between PM2.5 and childhood cognition in sites such as NYC, i.e., while the more highly educated mothers may be more likely to live in places with worse air pollution problems, their children performed well on cognition tests accounting for the observed positive association between PM2.5 exposure and children’s cognition. This should be further explored leveraging a larger sample size.

Our finding of sex-specific associations is in agreement with Calderón-Garcidueñas et al. (2016) who demonstrated a stronger association of air pollution exposure and IQ observed in females compared to males in a sample of Mexico City school children. Rahman et al. (2022) and Chiu et al. (2016) have also reported sex-specific associations between prenatal PM2.5 exposure and neurodevelopmental outcomes. In Rahman et al. (2022), the effects of prenatal PM2.5 were significant for both sexes but slightly stronger for females than males in the temperamental fear domain in 6 month olds. In Chiu et al. (2016), the effect for prenatal PM2.5 and memory function was stronger in females than in males in early school-aged children. Our sex effect finding was in contrast to two Mexico City studies, which did not observe sex differences in the effect of prenatal PM2.5 exposure on childhood cognitive (Bansal et al., 2021) or neurobehavioral problems (McGuinn et al., 2020). In addition, a Spanish study (Lertxundi et al., 2019) reported greater vulnerability in males when examining associations between prenatal PM2.5 and infant cognitive functioning assessed using the McCarthy Scales of Children’s Abilities. The discrepancy between our and other studies’ sex-specific findings could be partially explained by the outcomes investigated. Some cognitive functions might be subject to greater sex differences. For example, sex differences in verbal ability is reasonably well established (Mann et al., 1990). A study of 3000 child twin pairs reported that environmental factors had sex-specific effects on early verbal cognition but not on non-verbal development (Galsworthy et al., 2000). It is consistent with our finding that the verbal-related functions were found to have greater sex differences than other examined cognitive functions. The literature examining sex-specific effects of early life exposure to particulate air pollution and children’s neurodevelopmental outcomes demonstrates mixed findings. Moreover, whether girls or boys are more vulnerable to the effects varies across neurocognitive domains being considered. An earlier review (Clougherty, 2010) summarized that the health effects of air pollution were stronger among boys in early life, including during prenatal development, and among girls in later childhood. More recent studies demonstrate mixed results on sex differences (Calderón-Garcidueñas et al., 2016; Chiu et al., 2016; McGuinn et al., 2020; Bansal et al., 2021; Rahman et al., 2022). Moreover, Chiu and colleagues (Chiu et al., 2016) found associations between higher PM2.5 exposure levels during gestation and lower IQ and attention problems among boys, while increased prenatal exposure to PM2.5 was associated with poorer memory outcomes in girls.

Notable strengths of these analyses include the socioeconomically and racially/ethnically diverse sample residing in both urban and non-urban environments, the air pollution exposure estimates were obtained through the same validated satellite-based model across the three study sites, and childhood cognition was assessed using a validated, standardized computer-based protocol with centralized training for staff across sites led by developers of the NIHTB-CB. There are also some limitations. As the COVID-19 pandemic posed difficulties for continuing in-person examinations across the consortium, these analyses demonstrate the combinability methods in a sample limited to the Northeaster U.S. Future data will be available across the contiguous U.S. in the ongoing national ECHO program which will enroll up to 50,000 care-givers and their children to allow implementation in a larger and even more diverse population. The log-linear model implemented herein applies to categorical variables. The continuous variables were dichotomized by a median split, which might not be the best cutoff for comparing inter-study associations, as the associations between certain study variables could be linear. Other factors not examined due to the lack of available data may have contributed to the observed between-site heterogeneity (e.g., modifying genetic loci, maternal alcohol intake in pregnancy). These factors should be addressed in future work. Exposure assessment was limited to PM2.5 during pregnancy given currently available data for these ECHO cohorts. Future work should consider concurrent exposure to other air pollutants (e.g., NO2) as well as joint or cumulative associations between pre- and postnatal air pollution exposure on child cognitive outcomes. For the former, advanced methods will need to be developed to consider air pollution mixtures and cohort heterogeneity. Lastly, the urbanicity index available for these analyses indicated most study participants were living in urban areas. Future studies could select urbanicity indicators with higher variation to balance the numbers of urban and non-urban groups, which will make the stratified results still more comparable.

5. Conclusion

These analyses underscore the need to carefully consider variable distribution of exposure and related sociodemographic factors, and account for these differences in pooled analyses in epidemiology. We demonstrate an approach that assesses the combinability of heterogeneous populations prior to combining their data which can provide a better understanding of underlying cohort differences and provide increased power to detect associations that would be undetected by more customary methods for combining cohorts. Given that pooled analyses and collaborative larger-scale multi-site studies are increasingly being conducted in the field of environmental epidemiology (Dadvand et al., 2013; Gillman and Blaisdell, 2018; Pedersen et al., 2013), such approaches will have broad applicability. Future work should focus on improving the methodology in pooling data in multi-site studies and continuing to advance statistical approaches to accounting for between-site heterogeneity.

Supplementary Material

Supplementary

Funding sources

The PRogramming of Intergenerational Stress Mechanisms (PRISM) cohort has been supported under US National Institutes of Health (NIH) grants R01HL095606, R01HL114396, R21ES021318, and UG3/H3OD023337; statistical analyses and phenotyping support was provided by U54TR004213, and P30ES023515. The authors declare they have no competing financial interests. During preparation of this manuscript, XZ was supported by NIH grants P30ES023515 and UG3/H3OD023337. BC was supported by NIH grants UG3OD023337, R01ES028811 and P30ES000002. SHL was supported by NIH grants R03ES033374 and K25HD104918.

Footnotes

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.envres.2022.114163.

Data availability

The authors do not have permission to share data.

References

  1. Agresti A, Coull BA, 2002. The analysis of contingency tables under inequality constraints. J. Stat. Plann. Inference 107, 45–73. 10.1016/S0378-3758(02)00243-4. [DOI] [Google Scholar]
  2. Agresti A, 2003. Categorical Data Analysis. John Wiley & Sons. [Google Scholar]
  3. Akshoomoff N, Beaumont JL, Bauer PJ, Dikmen SS, Gershon RC, Mungas D, et al. , 2013. Viii. Nih toolbox cognition battery (cb): composite scores of crystallized, fluid, and overall cognition. Monogr. Soc. Res. Child Dev 78, 119–132. 10.1111/mono.12038. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bansal E, Hsu H-H, de Water E, Martínez-Medina S, Schnaas L, Just AC, et al. , 2021. Prenatal pm2.5 exposure in the second and third trimesters predicts neurocognitive performance at age 9–10 years: a cohort study of Mexico city children. Environ. Res 202, 111651 10.1016/j.envres.2021.111651. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Blaisdell CJ, Park C, Hanspal M, et al. , 2021. The NIH ECHO Program: investigating how early environmental influences affect child health. Pediatr. Res 10.1038/s41390-021-01574-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Brunst KJ, Sanchez-Guerra M, Chiu Y-HM, Wilson A, Coull BA, Kloog I, et al. , 2018. Prenatal particulate matter exposure and mitochondrial dysfunction at the maternal-fetal interface: effect modification by maternal lifetime trauma and child sex. Environ. Int 112, 49–58. 10.1016/j.envint.2017.12.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Calderón-Garcidueñas L, Jewells V, Galaz-Montoya C, van Zundert B, Pérez-Calatayud A, Ascencio-Ferrel E, et al. , 2016. Interactive and additive influences of gender, bmi and apolipoprotein 4 on cognition in children chronically exposed to high concentrations of PM2.5 and ozone. Apoe 4 females are at highest risk in Mexico city. Environ. Res 150, 411–422. 10.1016/j.envres.2016.06.026. [DOI] [PubMed] [Google Scholar]
  8. Carlozzi NE, Beaumont JL, Tulsky DS, Gershon RC, 2015. The nih toolbox pattern comparison processing speed test: normative data. Arch. Clin. Neuropsychol 30, 359–368. 10.1093/arclin/acv031. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Chao M-W, Yang C-H, Lin P-T, Yang Y-H, Chuang Y-C, Chung M-C, et al. , 2017. Exposure to pm2.5 causes genetic changes in fetal rat cerebral cortex and hippocampus. Environ. Toxicol 32, 1412–1425 10.1002/tox.22335. [DOI] [PubMed] [Google Scholar]
  10. Chen J-C, Wang X, Wellenius GA, Serre ML, Driscoll I, Casanova R, et al. , 2015. Ambient air pollution and neurotoxicity on brain structure: evidence from women’s health initiative memory study. Ann. Neurol 78, 466–476 10.1002/ana.24460. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Chiu Y-HM, Hsu H-HL, Coull BA, Bellinger DC, Kloog I, Schwartz J, et al. , 2016. Prenatal particulate air pollution and neurodevelopment in urban children: examining sensitive windows and sex-specific associations. Environ. Int 87, 56–65 10.1016/j.envint.2015.11.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Clougherty JE, 2010. A growing role for gender analysis in air pollution epidemiology. Environ. Health Perspect 118 (2), 167–176. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Cserbik D, Chen JC, McConnell R, Berhane K, Sowell ER, Schwartz J, Hackman DA, Kan E, Fan CC, Herting MM, 2020. Fine particulate matter exposure during childhood relates to hemispheric-specific differences in brain structure. Environ. Int 143. https://doi: 10.1016/j.envint.2020.105933. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Dadvand P, Parker J, Bell Michelle L, Bonzini M, Brauer M, Darrow Lyndsey A, et al. , 2013. Maternal exposure to particulate air pollution and term birth weight: a multi-country evaluation of effect and heterogeneity. Environ. Health Perspect 121, 267–373 10.1289/ehp.1205575. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Di Domenico M, Benevenuto SGdM., Tomasini PP, Yariwake VY, de Oliveira Alves N, Rahmeier FL, et al. , 2020. Concentrated ambient fine particulate matter (pm2.5) exposure induce brain damage in pre and postnatal exposed mice. Neurotoxicology 79, 127–141. 10.1016/j.neuro.2020.05.004. [DOI] [PubMed] [Google Scholar]
  16. Dikmen SS, Bauer PJ, Weintraub S, Mungas D, Slotkin J, Beaumont JL, et al. , 2014. Measuring episodic memory across the lifespan: nih toolbox picture sequence memory test. J. Int. Neuropsychol. Soc 20, 611–619. 10.1017/S1355617714000460. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Donzelli G, Llopis-Gonzalez A, Llopis-Morales A, Cioni L, Morales-Suárez-Varela M, 2019. Particulate matter exposure and attention-deficit/hyperactivity disorder in children: a systematic review of epidemiological studies. Int. J. Environ. Res. Publ. Health 17, 67. 10.3390/ijerph17010067. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Galsworthy MJ, Dionne G, Dale PS, Plomin R, 2000. Sex differences in early verbal and non-verbal cognitive development. Dev. Sci 3, 206–215. 10.1111/1467-7687.00114. [DOI] [Google Scholar]
  19. Gershon RC, Cook KF, Mungas D, Manly JJ, Slotkin J, Beaumont JL, et al. , 2014. Language measures of the nih toolbox cognition battery. J. Int. Neuropsychol. Soc 20, 642–651. 10.1017/S1355617714000411. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Gershon RC, Wagster MV, Hendrie HC, Fox NA, Cook KF, Nowinski CJ, 2013. NIH toolbox for assessment of neurological and behavioral function. Neurology 80, S2–S6. 10.1212/WNL.0b013e3182872e5f. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Gillman MW, Blaisdell CJ, 2018. Environmental influences on child health outcomes, a research program of the national institutes of health. Curr. Opin. Pediatr 30, 260–262. 10.1097/MOP.0000000000000600. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Guerra R, Vera-Aguilar E, Uribe-Ramirez M, Gookin G, Camacho J, Osornio-Vargas AR, et al. , 2013. Exposure to inhaled particulate matter activates early markers of oxidative stress, inflammation and unfolded protein response in rat striatum. Toxicol. Lett 222, 146–154. 10.1016/j.toxlet.2013.07.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Gui Z, Cai L, Zhang J, Zeng X, Lai L, Lv Y, et al. , 2020. Exposure to ambient air pollution and executive function among Chinese primary schoolchildren. Int. J. Hyg Environ. Health 229, 113583. 10.1016/j.ijheh.2020.113583. [DOI] [PubMed] [Google Scholar]
  24. Guo L, Wang H, Zhou J, Tang W, Wang R, Xiao Z, et al. , 2021. Magnetic resonance imaging investigations reveal that pm2.5 exposure triggers visual dysfunction in mice. Ecotoxicol. Environ. Saf 227, 112866 10.1016/j.ecoenv.2021.112866. [DOI] [PubMed] [Google Scholar]
  25. Guxens M, xf, nica, Garcia-Esteban R, Giorgis-Allemand L, Forns J, et al. , 2014. Air pollution during pregnancy and childhood cognitive and psychomotor development: six european birth cohorts. Epidemiology 25, 636–647. 10.1097/EDE.0000000000000133. [DOI] [PubMed] [Google Scholar]
  26. Harris MH, Gold DR, Rifas-Shiman SL, Melly SJ, Zanobetti A, Coull BA, Schwartz JD, Gryparis A, Kloog I, Koutrakis P, Bellinger DC, 2016. Sep 1. Prenatal and childhood traffic-related air pollution exposure and childhood executive function and behavior. Neurotoxicol. Teratol 57, 60–70. 10.1016/j.ntt.2016.06.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Just AC, Arfer KB, Rush J, Dorman M, Shtein A, Lyapustin A, et al. , 2020. Advancing methodologies for applying machine learning and evaluating spatiotemporal models of fine particulate matter (PM2.5) using satellite data over large regions. Atmos. Environ 239, 117649 10.1016/j.atmosenv.2020.117649. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Lertxundi A, Andiarena A, Martínez MD, Ayerdi M, Murcia M, Estarlich M, et al. , 2019. Prenatal exposure to PM2.5 and NO2 and sex-dependent infant cognitive and motor development. Environ. Res 174, 114–121. 10.1016/j.envres.2019.04.001. [DOI] [PubMed] [Google Scholar]
  29. Lopuszanska U, Samardakiewicz M, 2020. The relationship between air pollution and cognitive functions in children and adolescents: a systematic review. Cognit. Behav. Neurol 33, 157–178. 10.1097/WNN.0000000000000235. [DOI] [PubMed] [Google Scholar]
  30. Mann VA, Sasanuma S, Sakuma N, Masaki S, 1990. Sex differences in cognitive abilities: a cross-cultural perspective. Neuropsychologia 28, 1063–1077. 10.1016/0028-3932(90)90141-A. [DOI] [PubMed] [Google Scholar]
  31. McGuinn LA, Bellinger DC, Colicino E, Coull BA, Just AC, Kloog I, et al. , 2020. Prenatal PM2.5 exposure and behavioral development in children from Mexico city. Neurotoxicology 81, 109–115. 10.1016/j.neuro.2020.09.036. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Mullen C, Grineski SE, Collins TW, Mendoza DL, 2020. Effects of PM2.5 on third grade students’ proficiency in math and English language arts. Int. J. Environ. Res. Publ. Health 17, 6931. 10.3390/ijerph17186931. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Parker JD, Woodruff TJ, 2008. Influences of study design and location on the relationship between particulate matter air pollution and birthweight. Paediatr. Perinat. Epidemiol 22, 214–227. 10.1111/j.1365-3016.2008.00931.x. [DOI] [PubMed] [Google Scholar]
  34. Pedersen M, Giorgis-Allemand L, Bernard C, Aguilera I, Andersen A-MN, Ballester F, et al. , 2013. Ambient air pollution and low birthweight: a european cohort study (escape). Lancet Respir. Med 1, 695–704. 10.1016/S2213-2600(13)70192-9. [DOI] [PubMed] [Google Scholar]
  35. Perera FP, Rauh V, Tsai WY, Kinney P, Camann D, Barr D, Bernert T, Garfinkel R, Tu YH, Diaz D, Dietrich J, 2003. Feb. Effects of transplacental exposure to environmental pollutants on birth outcomes in a multiethnic population. Environ. Health Perspect 111 (2), 201–205. 10.1289/ehp.5742. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Peterson BS, Bansal R, Sawardekar S, Nati C, Elgabalawy ER, Hoepner LA,et al. , 2022. Prenatal exposure to air pollution is associated with altered brain structure, function, and metabolism in childhood. JCPP (J. Child Psychol. Psychiatry) 10.1111/jcpp.13578. [DOI] [PubMed] [Google Scholar]
  37. Porta D, Narduzzi S, Badaloni C, Bucci S, Cesaroni G, Colelli V, et al. , 2016. Air pollution and cognitive development at age 7 in a prospective Italian birth cohort. Epidemiology 27, 228–236. 10.1097/EDE.0000000000000405. [DOI] [PubMed] [Google Scholar]
  38. Rahman F, Coull BA, Carroll KN, Wilson A, Just AC, Kloog I, et al. , 2022. Prenatal PM2.5 exposure and infant temperament at age 6 months: sensitive windows and sex-specific associations. Environ. Res 206, 112583 10.1016/j.envres.2021.112583. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Ratcliffe M, Burd C, Holder K, Fields A, 2016. Defining Rural at the U.S https://www.researchgate.net/profile/Michael-Ratcliffe-2/publication/311533270_Defining_Rural_at_the_US_Census_Bureau/links/584aad3708aeb19dcb758910/Defining-Rural-at-the-US-Census-Bureau.pdf. (Accessed 18 March 2022). Last time accessed at.
  40. Rosa MJ, Pajak A, Just AC, Sheffield PE, Kloog I, Schwartz J, et al. , 2017. Prenatal exposure to PM2.5 and birth weight: a pooled analysis from three north american longitudinal pregnancy cohort studies. Environ. Int 107, 173–180. 10.1016/j.envint.2017.07.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Saenen ND, Martens DS, Neven KY, Alfano R, Bové H, Janssen BG, et al. , 2019. Air pollution-induced placental alterations: an interplay of oxidative stress, epigenetics, and the aging phenotype? Clin. Epigenet 11, 124. 10.1186/s13148-019-0688-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Saenen Nelly D, Plusquin M, Bijnens E, Janssen Bram G, Gyselaers W, Cox B,et al. , 2015. In utero fine particle air pollution and placental expression of genes in the brain-derived neurotrophic factor signaling pathway: an environage birth cohort study. Environ. Health Perspect 123, 834–840. 10.1289/ehp.1408549. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Samet JM, White RH, 2004. Urban air pollution, health, and equity. J. Epidemiol. Community 58, 3–5. 10.1136/jech.58.1.3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Sunyer J, Esnaola M, Alvarez-Pedrerol M, Forns J, Rivas I, López-Vicente M,et al. , 2015. Association between traffic-related air pollution in schools and cognitive development in primary school children: a prospective cohort study. PLoS Med 12, e1001792 10.1371/journal.pmed.1001792. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Tulsky DS, Carlozzi N, Chiaravalloti ND, Beaumont JL, Kisala PA, Mungas D, et al. , 2014. NIH toolbox cognition battery (NIHTB-CB): List sorting test to measure working memory. J. Int. Neuropsychol. Soc 20, 599–610. 10.1017/S135561771400040X. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Volk HE, Perera F, Braun JM, Kingsley SL, Gray K, Buckley J, et al. , 2021. Prenatal air pollution exposure and neurodevelopment: a review and blueprint for a harmonized approach within ECHO. Environ. Res 196, 110320 10.1016/j.envres.2020.110320. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Weintraub S, Dikmen SS, Heaton RK, Tulsky DS, Zelazo PD, Slotkin J, et al. , 2014. The cognition battery of the nih toolbox for assessment of neurological and behavioral function: validation in an adult sample. J. Int. Neuropsychol. Soc 20, 567–578. 10.1017/S1355617714000320. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Wilson A, Chiu YM, Hsu HL, Wright RO, Wright RJ, Coull BA, 2017. Potential for bias when estimating critical windows for air pollution in children’s health. Am. J. Epidemiol 186, 1281–1289. 10.1093/aje/kwx184. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Zelazo PD, Anderson JE, Richler J, Wallner-Allen K, Beaumont JL, Weintraub S, 2013. Ii. NIH toolbox cognition battery (CB): measuring executive function and attention. Monogr. Soc. Res. Child Dev 78, 16–33. 10.1111/mono.12032. [DOI] [PubMed] [Google Scholar]
  50. Zhang X, Smith N, Spear E, Stroustrup A, 2021. Neighborhood characteristics associated with covid-19 burden—the modifying effect of age. J. Expo. Sci. Environ. Epidemiol 31, 525–537. 10.1038/s41370-021-00329-1. [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
NIHMS1850097-supplement-Supplementary.docx (1,017.6KB, docx)

Data Availability Statement

The authors do not have permission to share data.

RESOURCES