Associations between weekly gestational exposure of fine particulate matter, ozone, and nitrogen dioxide and preterm birth in a North Carolina Birth Cohort, 2003–2015 (2024)

Abstract

Introduction

Preterm birth (PTB) is birth before completion of 37 weeks gestation. In the United States, the PTB rate was 10.5% in 2021, rising from 10.1% in 2020.¹ The PTB rate in the United States is close to the worldwide PTB rate (10.6%), and increases in the rate of PTB have been observed in high-income and highly industrialized countries.² Babies born prematurely are at risk for higher rates of infant mortality and morbidities, including developmental delays, visual and hearing impairments, and respiratory complications.¹

A growing body of studies reports associations between PTB and air pollution; generally, they focus on criteria air pollutants, such as fine particulate matter (PM_2.5), nitrogen dioxide (NO₂), and ozone (O₃).^3–9 While many studies have demonstrated associations between PTB and exposure to these criteria air pollutants, there is variability in the magnitude and direction of these associations. One source of variability is the exposure window being evaluated. Many studies average exposure over months, trimesters, or the entire pregnancy, but there are uncertainties regarding the critical window of exposure to these pollutants and PTB.^4–6,10 Qian et al.⁴ reported that the strongest effect for PTB was during the second trimester for PM_2.5, the first trimester and second month for NO₂, and the third trimester for O₃. Estarlich et al.⁵ reported that the second trimester appeared to be a more relevant period for NO₂ exposure and risk of being born prematurely. Ji et al.⁶ reported that third-trimester exposure to NO₂ may increase the risk of PTB.

Another potential source for heterogeneity in observed associations between PTB and criteria air pollutants is the estimation of single pollutant effects without control for potential copollutant confounding. Criteria air pollutants co-occur in ambient air as part of a mixture. The combination of pollutants and their relative contribution to the air pollution mixture can vary over space and time. Factors that contribute to this temporal and spatial variability include emission sources and pollutant behavior in the atmosphere; some pollutants are directly emitted (primary pollutants, e.g., NO, elemental carbon), others are created as the result of atmospheric chemistry/processes (secondary pollutants, e.g., O₃), and some are mixtures of both of these processes (e.g., PM_2.5, NO₂). Accounting for potential copollutant confounding using copollutant (or multipollutant) models may provide useful information regarding the variability in the magnitude and consistency of the associations between air pollution exposure and birth outcomes.

To address these uncertainties, we evaluated the associations between weekly gestational exposure to fine PM_2.5, NO₂, and O₃ with PTB in a North Carolina (NC) birth cohort from 2003 to 2015 (N = 1,367,517). We examine single pollutant models and copollutant and multipollutant models to evaluate potential confounding by co-occurring criteria pollutants. The inclusion of weekly exposure windows and copollutant and multipollutant models may help to reduce some uncertainties in the air pollution and birth outcome literature related to variability in reported associations.

Methods

Results

Of the 1,367,517 infant-gestational parent pairs included in this analysis, there were 114,706 (8.4%) PTB cases with residents in NC between 2003 and 2015 (Table ​(Table1).1). Gestational parent age at birth did not vary between preterm and term births, with a median age of 27 years. There was a higher proportion of Black, non-Hispanic gestational parents among the PTBs (32.3%) compared with the term births (22.3%) and a higher proportion of White, non-Hispanic gestational parents among term births (57.2%) compared with PTBs (50.6%). A higher proportion of PTB gestational parents were unmarried (47.2%) and Medicaid eligible at delivery (55.8%) than term birth gestational parents (39.1% and 49.7%, respectively).

The median (IQR) daily averaged concentration was 15.0 (10.4, 20.5) ppb for NO₂, 10.4 (9.1, 12.7) µg/m³ for PM_2.5, and 41.2 (38.9, 44.0) ppb for O₃ (Table ​(Table2).2). The three pollutants were not highly correlated with each other, with all Pearson rho values less than 0.5 (Supplemental Figure 1; http://links.lww.com/EE/A249; Supplemental Table 1; http://links.lww.com/EE/A249).

In single pollutant models, the associations between PM_2.5, NO₂, or O₃ and PTB were similar in magnitude across gestational weeks (Figure ​(Figure1A–C,1A–C, Supplemental Table 2; http://links.lww.com/EE/A249). There were positive associations between PM_2.5 and O₃ and PTB across gestational weeks, but generally null associations with NO₂ across gestational weeks. In the single pollutant models for PM_2.5, RDs per IQR (3.58 µg/m³) increase ranged from 15 (95% CI = 11, 18) to 32 (95% CI = 27, 37) per 10,000 births. In other words, there were an additional 32 PTB per 10,000 births for every 3.58 µg/m³ (IQR) increase in PM_2.5 in gestational week 21. In the single pollutant model for NO₂, the RDs per IQR (10.09 ppb) increase ranged from −7 (95% CI = −14, −1) to 0 (95% CI = −5, 4) per 10,000 births. In the single pollutant models for O₃, the RDs per IQR (5.07 ppb) increase ranged from 4 (95% CI = 1, 7) to 13 (95% CI = 10, 16) per 10,000 births.

In all combinations of the copollutant models (PM_2.5 and NO₂; PM_2.5 and O₃; NO₂ and O₃), the magnitude of the associations was similar across gestational weeks (Figure ​(Figure1A–C;1A–C; (Supplemental Table 3; http://links.lww.com/EE/A249). Generally, the associations between PM_2.5 and PTB were positive, whether adjusted for NO₂ and/or O₃, and were similar to the RDs observed in the single pollutant PM_2.5 models. In the copollutant models adjusted for O₃, the RDs for NO₂ were generally null and similar in magnitude to the single pollutant NO₂ models. However, in the copollutant model adjusted for PM_2.5, the RDs were decreased for NO₂, compared with the single pollutant NO₂ models, with RDs per IQR increase ranging from −10 to −29 per 10,000 births. In the copollutant models adjusted for NO₂, there were positive associations between O₃ and PTB, with RDs similar in magnitude to the single pollutant O₃ models. When adjusted for PM_2.5, the RDs for O₃ were generally negative or near null, ranging from −9 (95% CI = −12, −5) to 0 (95% CI = −4, 4), whereas in the single pollutant models, there were positive associations between O₃ and PTB.

In the multipollutant model (mutually adjusted for PM_2.5, NO₂, and O₃), the magnitude of the associations across gestational weeks were similar, with positive associations between PM_2.5 and PTB, negative associations with NO₂, and null associations with O₃ (Figure ​(Figure1A–C;1A–C; Supplemental Table 3; http://links.lww.com/EE/A249). While the RDs for PM_2.5 in the multipollutant models were similar to the RDs for PM_2.5 in the single pollutant PM_2.5 models, there was an increase in the magnitude of the RDs for NO₂ compared with the single pollutant models, and the RDs for O₃ were attenuated to the null in the multipollutant models compared with the single pollutant models.

The results of the sensitivity analyses are detailed in the supplemental materials; http://links.lww.com/EE/A249. Overall, we did not observe differences between the hybrid ensemble model and the fCMAQ model for PM_2.5. When comparing the RDs for PTB for weekly gestational PM_2.5 exposures less than 12, 10, and 8 µg/m³, there was more variability in the magnitude of the associations across the gestational weeks, with the largest variation in RDs for weekly gestational PM_2.5 exposures less than 8 µg/m³. There were varied associations among the thresholds for each pollutant (“very low,” below the 25th percentile; “low,” below the 50th percentile; “high,” above the 50th percentile; and “very high,” above the 75th percentile) per IQR increase in pollutant per 10,000 births. For exposure averaged over the entire pregnancy or by trimester, there were positive associations with PTB for PM_2.5 and O₃, but a null association for NO₂. Lastly, the results from the logistic regression models follow similar patterns of associations as the RD models.

Discussion

Overall, increased PM_2.5 exposure is associated with increased risk of PTB across gestational weeks, and in copollutant or multipollutant models with NO₂ and/or O₃, whereas associations from single pollutant models for NO₂ and O₃ were shifted to below the null value when PM_2.5 was included in copollutant or multipollutant models. When examining the RDs for PTB for weekly gestational PM_2.5 concentrations less than certain policy-relevant thresholds (i.e., 12, 10, and 8 µg/m³), there was more variability in the magnitude of the associations across the gestational weeks, with the largest variation in RDs for weekly gestational PM_2.5 concentrations less than 8 µg/m³.

Previous studies have reported stronger associations between PM_2.5, NO₂, and O₃ during middle and late pregnancy.^17,18 Our single pollutant results, showing increased risk of PTB associated with PM_2.5 and O₃ exposure across gestational weeks, and generally null effects for NO₂, are not entirely consistent with previous studies. Most similar to our study, Wang et al.,¹⁹ examined weekly exposures throughout gestation for PM_2.5, NO₂, and O₃ and PTB in Guangzhou, China. They reported increased hazard ratios (HRs) for exposure during the middle to late pregnancy (gestational weeks 20–28 for PM_2.5, 18–31 for NO₂, and 23–31 for O₃); the strongest associations were observed between weeks 25 and 28 for each pollutant (HRs ranged from 1.03 to 1.06). In an analysis of California birth data, Sheridan et al.,²⁰ examined the association between PTB and PM_2.5 averaged across weekly exposure windows and observed an increased risk in PTB for gestational weeks 17–24 and 36. Qian et al.,⁴ examined exposures to relatively higher air pollution concentrations in Wuhan, China and observed the strongest association between PM_2.5 and PTB for concentrations averaged over the second trimester. For NO₂, the strongest associations were observed for the first trimester and second month of gestation, while for O₃ the strongest association was observed for the third-trimester exposure window. In statistical evaluations of critical exposure windows for PM_2.5 and PTB²¹ and very PTB²² early and middle periods of pregnancy were reported to be most vulnerable. A recent systematic review and meta-analysis (8) reported elevated ORs for PTB and O₃ in both the first and second trimesters.

The RDs we observed for PM_2.5 across gestational weeks were relatively unchanged in copollutant models for either NO₂ or O₃. Adjustment for PM_2.5 resulted in attenuated RDs across gestational weeks for NO₂ and O₃. Similar results have not been consistently observed in other studies. In copollutant models, Wang et al.,¹⁹ noted that the HRs for O₃ were relatively unchanged, however, the effects of PM_2.5 were attenuated when adjusting for NO₂ and the effects of NO₂ were attenuated when adjusting for PM_2.5. When evaluating PM_2.5 and O₃, ORs were relatively unchanged when in copollutant models for either of the other two pollutants.⁴ Olsson et al.²³ reported increased odds of PTB associated with O₃ exposure (in copollutant models with NO₂) during the first and second trimesters, though ORs were attenuated and nearly null for O₃ concentrations averaged over the week preceding delivery. They also observed increased odds of PTB associated with NO₂ concentrations (in copollutant models with O₃) during the week preceding delivery, though they reported negative odds of PTB associated with NO₂ exposure during the first or second trimester. No results were reported for single pollutant models adjusted for covariates that we could compare with our single pollutant results.

Biologically plausible mechanisms by which air pollution exposure may cause PTB are mainly through translocation of particles from the respiratory tract to the cardiovascular system, placenta, and cord blood; and systemic inflammation and oxidative stress pathways.^17,18,24 These pathways may originate when sensory nerves in the respiratory tract are activated or by respiratory inflammation and oxidative stress, which can lead to reproductive organ-specific inflammation, including placental oxidative stress and intrauterine inflammation, altered umbilical cord lymphocyte distribution, and increased inflammation in cord blood.^17,18 Each of these pathways, alone or in combination, could provide a plausible biological mechanism by which gestational exposure to air pollution could contribute to the etiology of PTB.

Recent studies of air pollution and health effects use a variety of both fixed-site (i.e., monitors), models (e.g., CMAQ, dispersion models), and satellite-based (e.g., aerosol optical depth observations from satellites) methods, including hybrid methods that combine two or more fixed-site, model, and/or satellite-based techniques to measure, estimate, or predict air pollution concentrations used in epidemiologic studies. We used modeling data (i.e., fCMAQ, hybrid ensemble model) to estimate air pollution exposures across gestational weeks for a NC birth cohort. Many recent studies of air pollution and birth outcomes, including PTB, continue to rely on monitoring data to assign exposure.^4,19,20,23 The models used in these analyses have been well-validated and used in previous epidemiologic studies of birth outcomes.^25–27

We acknowledge certain limitations in our study that may affect inference or interpretability. We assigned exposure to three criteria air pollutants based on census tract of residence at birth, which may result in exposure misclassification if gestational parents spent a substantial amount of time outside of their home and outside of their census tract. The degree of misclassification could be different for gestational parents residing in urban and rural census tracts, since the geographic size of the census tract varies with population density (e.g., census tracts have a smaller area in urban centers and a larger area in rural communities). In addition, the degree of spatial variability in air pollutant concentrations differs by pollutant. We would anticipate less spatial variability for secondary or regional pollutants (i.e., PM_2.5, O₃) than we would from primary pollutants (i.e., NO₂). This could result in differences in the degree of exposure misclassification for different pollutants considered in our analyses. Finally, we use the gestational parent residential address at time of delivery to assign exposure. We do not have information on residential mobility during pregnancy. If gestational parents moved from one census tract to another during pregnancy, exposure misclassification may occur. The timing of a residential move during gestation might also affect our ability to identify critical windows of exposure.²⁸

Another limitation of our analyses is that we did not adjust our models for ambient temperature or other climate-related factors that could be related to both air pollution concentrations and be risk factors for PTB. We included month of conception as covariate in our models, which should help to control with seasonally varying climate-related factors in NC. Finally, by nature of our data source (i.e., vital records), we do not have information on indoor or occupational air quality and use ambient concentrations of a subset of criteria air pollutions in defining exposure.

There are notable strengths to our study design and analyses. We evaluated exposures across the entire pregnancy period, trimesters, and each week of gestation to identify critical windows of exposure and potentially explain heterogeneity in results provided for longer exposure windows. Our ability to compare results for exposures averaged over the entire pregnancy, trimesters, and each week of gestation to each other and to the results in other studies will contribute to the overall evidence base supporting the causal nature of the association between air pollution and PTB. In addition, we conducted analyses using both Poisson regression and logistic regression and present our results as RDs and ORs. Presenting RDs helps to increase interpretability of our results when communicated to policymakers and the public; presenting ORs facilitates comparison of our results with the result of previous epidemiologic studies of air pollution and PTB. Similarly, we compared our results for PM_2.5 using two available air quality models: the hybrid ensemble model and EPA’s fused CMAQ downscaler model. The pattern of results for PM_2.5 and PTB was similar for both models, demonstrating the robustness of our results to air quality model choice. Our results are based on a robust sample from a state-wide population-based cohort from a state with diverse demographic characteristics (e.g., race-ethnicity, education, income, urban/rural) and relatively low air pollution concentrations.

Conclusions

Overall, increased PM_2.5 exposure is associated with increased risk of PTB across gestational weeks and these associations are robust in copollutant or multipollutant models with NO₂ and/or O₃, whereas associations from single pollutant models for NO₂ and O₃ were attenuated to below the null value when PM_2.5 was included in copollutant or multipollutant models. In our population-based closed cohort, PM_2.5 concentrations drive associations observed between these criteria air pollutants and PTB. While no critical windows of exposure were evident in primary analyses, threshold analyses provide some evidence of mid to late weeks of pregnancy being a critical window for PM_2.5 exposure and PTB.

ACKNOWLEDGMENTS

The authors would like to thank Adrien Wilkie and Alexandra Larsen for comments on an early version of the manuscript.

Supplementary Material

Footnotes

References