The study population was identified via North Carolina birth certificate data, from 2003-2015, which were linked with the North Carolina Birth Defects Monitoring Program (BDMP) registry. The North Carolina BDMP is housed in the North Carolina Division of Public Health’s State Center for Health Statistics and is a statewide, population-based surveillance program. BDMP field staff conduct active case ascertainment of birth defects diagnosed within the first year of life. The BDMP staff systematically review medical records throughout the state and incorporate data from administrative health databases (i.e., vital records, Medicaid claims, hospital discharge data, and others) to supplement case ascertainment. We focused our analysis on CHDs that were shown in previous literature to have a higher prevalence among babies born of non-Hispanic Black women compared with non-Hispanic White women: pulmonary valve atresia/stenosis, TOF, and AVSD^2,21. Infants with the aforementioned CHDs were compared with all other infants in the study population, including those who may have other types of birth defects.

The ambient air pollution exposure data on particulate matter of 2.5 micrometers or less in diameter (PM_2.5) and ozone come from the Downscaler Model of the EPA CMAQ model. The CMAQ model incorporates meteorological, emission, and air pollution data together to give comprehensive estimates of ozone and particulates, as the model combines atmospheric science and air quality models²⁶. The Downscaler Model is a spatially-varying weighted model that takes data from CMAQ and smooths the output with weights that account for spatial and temporal differences. This allows for rigorous model predictions in geographical areas and on days in which data from a fixed-site pollution monitor may be missing or limited²⁷. Utilizing the Downscaler Model permits finer scale estimates of air pollution concentrations at the community level as the Downscaler Model creates averaged pollution estimates for each census tract²⁷. Daily pollution estimates of each census tract were approximated by assigning the model-estimated concentration at the centroid of the census tract to the entire census tract.

Air pollution exposure assignment for each pregnancy in the 2003-2015 North Carolina birth cohort was previously done by geocoding maternal addresses at delivery (found on the birth certificate) and assigning daily exposure values of PM_2.5 and ozone from the CMAQ Downscaler Model to each pregnancy^28,29. We were specifically interested in air pollution exposure during obstetric weeks 3-9 as this encompasses the critical window of susceptibility for CHDs. The exposure estimate for PM_2.5 is a 7-week average of the daily 24-hour average exposure of PM_2.5 (in micrograms per cubic meter (μg/m³)) during weeks 3-9 of pregnancy for the census tract containing the maternal address at date of birth. The exposure estimate for ozone is a 7-week average of the daily 8-hour maximum exposure of ozone (in parts per billion (ppb)) during weeks 3-9 of pregnancy for the census tract containing the maternal address at date of birth.

A total of 1,497,605 records of live-births were available for this research study. We applied several exclusion criteria (Supplementary Figure 1). We excluded 3,404 infants born with known chromosomal abnormalities to disentangle birth defects related to underlying chromosomal alterations from those that may be related to environmental factors. Additionally, we excluded pregnancies with critical windows occurring in 2002 because they were missing at least 1 week of pollution exposure data during weeks 3-9 of pregnancy (n=50,442). Our analytic dataset was also restricted to singleton births through excluding multiples (n=49,668). During the year of 2010, there were changes in reporting and data collection on the North Carolina birth certificate that resulted in the following missing covariates in our dataset for 2010 only: maternal education, maternal diabetes, and maternal smoking during pregnancy. The aforementioned missing covariates were identified as confounding variables in our research question^30–34. Due to the potential to introduce bias, we excluded births occurring in 2010 (n=109,116). Furthermore, we excluded women with diabetes (occurring prior to or during pregnancy) (n=56,410)^33–35. Lastly, we excluded births that were less than 500 grams in addition to births that were less than 20 weeks gestation or greater than 44 weeks gestation (n=3,280) due to the implausibility of these outliers in the data. After all exclusion criteria were applied, a total of 1,225,285 births remained for analyses.

Log-binomial models were created to examine the association between selected criteria air pollutants (PM_2.5 and ozone) and the CHDs of interest. PM_2.5 and ozone exposures were examined as continuous and categorical variables for comparability with other epidemiologic studies. Log-binomial regression models estimated continuous exposure as a unit change equal to the interquartile range (IQR), where the IQR for PM_2.5 and ozone are respectively 3.81 μg/m³ and 15.84 ppb. We examined various cut-points of PM_2.5 exposure at and below the current EPA annual average standard of 12 μg/m³ as well. We created three dichotomous variables at different PM_2.5 exposure levels to be examined in log-binomial models: 1) at or below 8 μg/m³ vs. greater than 8 μg/m³; 2) at or below 10 μg/m³ vs. greater than 10 μg/m³; 3) at or below 12 μg/m³ vs. greater than 12 μg/m³. We also created log-binomial models in which PM_2.5 exposure greater than 10 μg/m³ and, separately, greater than 12 μg/m³ was compared to PM_2.5 exposure that was at or below 8 μg/m³ in order to compare these cut-point groupings more directly.

Each CHD (coded dichotomously as yes/no) was examined with the aforementioned exposure groupings in unstratified and racially-stratified models. The stratified models were limited to non-Hispanic White and non-Hispanic Black individuals as our primary interest was to examine health disparities between these two race groups and because the CHDs examined in this study were shown in previous literature to be higher in non-Hispanic Black pregnancies. We created unadjusted single-pollutant models, adjusted single-pollutant models, and adjusted co-pollutant models. Confounding variables were identified a priori through a directed acyclic graph (DAG) (Supplementary Figure 2) and were included in adjusted models^{2,21,30–32,36}. Unstratified analyses included the following adjustment set: maternal smoking during pregnancy (yes/no), maternal age (<20yrs, 20-24yrs, 25-29yrs, 30-34yrs, ≥35yrs), maternal Medicaid status (yes/no), maternal education (less than high school, high school, more than high school), Social Vulnerability Index (SVI) (quartiles), and self-classified maternal race (non-Hispanic White, non-Hispanic Black, Hispanic, Asian/Pacific Islander, American Indian, other/unknown). The inclusion and adjustment of race in our statistical models of the overall population is as a proxy of residual and unmeasured factors related to structural racism. Models stratified by race included the same adjustment set mentioned previously, but omitted race. The adjustment for maternal Medicaid status, maternal education, and SVI quartile assignment serve as proxies for SES.

The SVI³⁷, created by the Centers for Disease Control & Prevention and the Agency for Toxic Substances & Disease Registry, utilizes census data to approximate relative social vulnerability of census tracts throughout the USA³⁷. We used data from the state-level SVI for North Carolina where census tracts in North Carolina are compared within the state. The SVI assigns all census tracts an index score of vulnerability on a scale between 0 and 1, where 0 indicates the least vulnerable and 1 indicates the highest vulnerability³⁸. Additional details on the SVI and index assignment can be found in Supplementary Figure 3.

As a sensitivity analysis, we computed all aforementioned analyses for both isolated and multiple birth defect cases due to potential differences in etiology of isolated and multiple cases of CHDs. In our study, isolated birth defect cases were defined as having only the defect of interest (i.e., pulmonary valve atresia/stenosis, or TOF, or AVSD). Infants considered to have multiple birth defects in our analyses had one of the CHDs of interest and at least one additional birth defect of any kind. Due to potential differences in the etiology and severity of pulmonary valve atresia and pulmonary valve stenosis, we carried out sensitivity analyses of the previously mentioned analyses for the two defects separately. Prevalence ratios (PRs) with 95% confidence intervals (95% CI) were computed in all models. All analyses were carried out in SAS version 9.4 (SAS Institute, Cary, NC, USA).

In our study, we observed that maternal exposure to PM_2.5 and ozone during the critical window of fetal heart development may be associated with an increase in prevalence of pulmonary valve atresia/stenosis, TOF, and AVSD. We observed a positive concentration-response gradient between PM_2.5 and TOF. We observed no clear concentration-response gradients between PM_2.5 and pulmonary valve atresia/stenosis or PM_2.5 and AVSDs. The strongest association within ozone models was observed between ozone and AVSDs. Racially-stratified results suggest that air pollution exposure may be impacting non-Hispanic Black and non-Hispanic White pregnancies differently. Specifically, we observed stronger effect estimates at lower air pollution exposures levels in non-Hispanic Black pregnancies for pulmonary valve atresia/stenosis and lower effect estimates for TOF in non-Hispanic Black pregnancies.

Our study is the first, to our knowledge, to investigate and present racial differences in associations between selected criteria air pollutants and CHDs. One previous study by Ritz et al., which examined 185 cases of pulmonary valve atresia/stenosis in relation to hourly ozone exposure²³, mentioned stratifying by race in their study, but did not present findings as authors noted that the number of cases in the strata were too small to be informative. In contrast, we examined 854 cases of pulmonary valve atresia/stenosis in relation to daily 8-hour maximum ozone concentrations from the EPA CMAQ Downscaler model. While no studies have presented racially-stratified effect estimates in the association between maternal exposure to selected criteria air pollutants (PM_2.5 and ozone) during pregnancy and selected CHDs (pulmonary valve atresia/stenosis, TOF, AVSDs), our overall findings within the unstratified analyses are also generally consistent with the previous epidemiologic literature.

While few studies examined the relationship between TOF and our pollutants of interest in stratified isolated vs. multiple birth defect analyses, our study examined ozone exposure during pregnancy in infants who have TOF in conjunction with other birth defects. The few previous epidemiologic studies that studied the relationship between maternal ozone exposure during pregnancy and AVSDs have generally found null or negative results^5,7,17. However, our study observed a positive association between maternal ozone exposure and AVSDs. The one study that found negative associations between maternal ozone exposure and AVSDs had a very small AVSD case count (n=14)⁵. In our overall analyses using quartiles models, we observed adjusted prevalence ratios ranging from 1.14 to 1.50, with the highest prevalence ratio of 1.50 in quartile 2.

Though we did not observe clear evidence of a stronger association between the air pollutants and all CHDs in our study for non-Hispanic Black individuals, we believe that some differences seen by race could result from effects of structural racism. Structural racism can generally be described as ideologies, processes, and systems embedded in society that reinforce racial and ethnic inequities^25,39. As a result of embedded racism within societal structures, minority populations, particularly non-Hispanic Black people in the United States, often live in areas with higher exposure to air pollution^18,25,40, have higher amounts of chronic stress, and poorer health outcomes²⁵. Effects of structural racism, such as a higher exposure to air pollutants and chronic levels of toxic stress, can contribute to effects such as increased oxidative stress, inflammation, and cell death within the body⁴¹; these physiological effects can in turn lead to adverse health outcomes.

In our study, we observed that non-Hispanic Black people generally had similar exposures to PM_2.5 and ozone during pregnancy compared to non-Hispanic White people. Results that we observed in our study are then likely not due to air pollution exposure differences but may rather be a result a different pathway, such as chronic stress exacerbating physiological effects of air pollution during pregnancy. Further investigation on what impact structural racism may have on birth outcomes in non-Hispanic Black individuals during pregnancy is warranted.

Among the strengths present in our study are comprehensive exposure assessment, recent data, large sample size for birth defect groups, and also a diverse study population that is well suited for studying racial disparities. Moreover, we are the first study that has presented racially-stratified estimates for the relationship between air pollution and selected heart defects. Many studies simply adjust for race in their models and may also be underpowered to examine differences in effect estimates by race. As disparities have been shown to exist in air pollution exposure and birth defect prevalence, examining race-specific effect estimates is important in understanding if racial differences are present in the effect of air pollution on birth defects.

The use of EPA’s CMAQ Downscaler Model to estimate exposure allows for estimation of air pollution in areas that do not have pollution monitors and on days in which data on air pollution levels is missing. In addition, the Downscaler Model has high spatial and temporal resolution and provides exposure data at the census tract level, allowing estimates within both urban and rural areas of North Carolina.

In addition to the high-quality exposure data used, our study represents the most recent birth cohort among the US-based studies that have examined pulmonary valve atresia/stenosis, TOF, and AVSD in relation to ozone and PM_2.5. While the most recent US-based study spans 2000-2009, our study includes births from 2003-2009 and 2011-2015, allowing for a longer study time period as well as 5 years of more recent data. Furthermore, our study has the largest number of pulmonary valve atresia/stenosis cases to be examined in relation to ozone and PM_2.5. While individual birth defects are rare, the number of pulmonary valve atresia/stenosis cases in our study is 854, whereas the largest case count from similar, previous literature⁹ is 436. With nearly double the cases of pulmonary valve atresia/stenosis, our study has higher statistical power to examine the relation between this CHD and pollutants of interest, especially in racially-stratified analyses. Additionally, the source of our birth defect data is robust as it comes from the North Carolina BDMP, which used active case ascertainment.

Our study also has a number of limitations. We estimated air pollution exposure during weeks 3-9 of pregnancy, the CHD critical window, based on the maternal address on the birth certificate. In cases where pregnant individuals move out of the census tract that they were living in between the first trimester of pregnancy and when they gave birth, there is potential for exposure misclassification due to residential address change. However, we believe the chance for misclassification is moderately low based on a previous study⁴² that observed that most pregnant individuals move within the same exposure region, thereby maintaining nearly the same level of air pollution exposure upon moving. The use of the birth certificate for covariate data on maternal smoking, education, and diabetes status is also a limitation. There is also potential for residual, unmeasured confounding in our study. While we utilized race as a proxy for structural racism, there may be additional unmeasured confounders related to structural racism. There may also be residual confounding due to unmeasured spatially-varying factors (e.g., other air pollutants, lack of green spaces, etc.) that could potentially be associated with CHDs.

In conclusion, results from our study suggest that PM_2.5 exposure during early pregnancy may be associated with an increased prevalence of pulmonary valve atresia/stenosis, TOF, and AVSD. While aPRs were greater than one, we found an increasing, concentration-response trend in models with TOF and PM_2.5 but no clear concentration-response trend in categorical models with PM_2.5 and the other two CHDs. Ozone exposure during early pregnancy was also shown to increase the prevalence of AVSDs in our study. We found that in models stratified by race, non-Hispanic Black and non-Hispanic White pregnancies may be impacted differently by air pollution when it comes to CHD prevalence.

Our findings are also consistent with previous literature that show air pollution to have differing impacts on various subtypes of CHDs^5,7–9,16. Additional studies are needed to confirm findings from our study. We recommend that future studies avoid grouping CHDs when evaluating the impact of air pollution as differing etiologies of CHDs may contribute to varying associations seen between CHDs and pollutants. We also recommend future epidemiologic studies estimate effects within race subgroups, when sample size permits, to improve understanding of observed racial disparities in these outcomes.