Is Exposure to Air Pollution During Pregnancy Linked to Lower Birth Weight?

Cowell et al. (2025) ask whether there are specific weeks during pregnancy when exposure to fine particulate air pollution (PM2.5) has the strongest association with birth weight. They analyze data from 16,868 full-term singleton births in the U.S. Environmental Influences on Child Health Outcomes (ECHO) cohort, using weekly, address-level PM2.5 exposure estimates derived from machine-learning models. They find that higher prenatal PM2.5 exposure is associated with lower birth weight for gestational age, with an overall effect size of −0.06 standard deviations per 1 μg/m³ increase in PM2.5. The most sensitive window occurs in early pregnancy (weeks 1–5), particularly for male infants, and effect sizes are larger in the South and Midwest.

Why This Article Was Selected for The Policy Scientist

This article addresses a policy-relevant question with broad implications for environmental regulation and population health: whether currently permitted levels of air pollution pose risks during specific stages of pregnancy. The topic is timely given recent revisions to U.S. PM2.5 standards and growing attention to early-life determinants of long-term health. Cowell and colleagues have published extensively on environmental exposures and child outcomes, and this study consolidates that line of work using the large, geographically diverse ECHO cohort. The dataset is high quality, with fine-grained, address-level exposure measures that enhance external validity across U.S. regions and plausibly to other high-income countries. The Bayesian distributed lag models are well suited for identifying sensitive exposure windows, though the study remains observational.

Full Citation and Link to Article

Cowell, W., Hsu, H. L., Just, A. C., Coull, B. A., Wilson, A., Hipwell, A., Karagas, M. R., Gilliland, F., Padula, A., Carroll, K., Kerver, J., Ghassabian, S., Camargo, C., Dabelea, D., et al. (2025). Air Pollution Exposure and Birth Weight in the ECHO Cohort. JAMA Network Open, 8(12), e2551459. https://doi.org/10.1001/jamanetworkopen.2025.51459 

Central Research Question

This study asks whether prenatal exposure to fine particulate matter (PM2.5) has time-specific effects on fetal growth and, if so, whether there are identifiable windows during pregnancy when exposure is most strongly associated with birth weight for gestational age. Rather than treating pregnancy as a single exposure period, the authors focus on week-by-week exposure timing to determine whether early, mid, or late gestation represents a period of heightened susceptibility. They also examine whether these exposure–outcome relationships differ by infant sex, maternal race and ethnicity, and U.S. geographic region. The central contribution lies in moving beyond average or trimester-level exposure measures to identify when exposure matters most, which has implications for both biological understanding and regulatory policy.

Previous Literature

A substantial epidemiologic literature has documented a negative association between prenatal PM2.5 exposure and birth weight, including several meta-analyses showing increased risks of low birth weight and small-for-gestational-age births. However, most prior studies relied on average exposure across pregnancy or broad trimester-level measures, which obscure potentially important variation in fetal vulnerability across developmental stages. Experimental and mechanistic studies suggest that PM2.5 can disrupt placental development through inflammation, altered gene expression, oxidative stress, and impaired trophoblast invasion, all of which are processes concentrated early in gestation. A smaller number of observational studies using distributed lag models have attempted to identify sensitive windows, with mixed findings regarding whether susceptibility is greatest in early, mid, or late pregnancy. Evidence on effect modification by infant sex or geography has been limited, and few studies have used exposure estimates with high spatial and temporal resolution. This study builds directly on that literature by combining a large, diverse cohort with weekly exposure estimates and flexible statistical models designed to identify windows of susceptibility.

Data

The analysis draws on data from the Environmental Influences on Child Health Outcomes (ECHO) Cohort, a large, multi-site U.S. consortium that harmonizes prospectively collected data from numerous pregnancy and birth cohorts. The final analytic sample includes 16,868 full-term singleton births occurring between 2003 and 2021 across 50 study sites in the contiguous United States. Birth weight and gestational age were obtained from medical records or validated reports and converted into sex-specific birth weight-for-gestational-age (BWGA) z scores using a national reference. Prenatal PM2.5 exposure was estimated at the residential address level using a high-resolution machine-learning model that generates daily exposure estimates across the United States; these were aggregated into weekly means across pregnancy. The dataset includes detailed information on maternal age, education, prepregnancy body mass index, parity, smoking during pregnancy, ambient temperature, infant sex, race and ethnicity, and geographic region. Overall, the data are notable for their size, geographic coverage, and fine-grained exposure measurement.

Methods

The authors use Bayesian distributed lag interaction models to estimate both cumulative and week-specific associations between PM2.5 exposure and BWGA z scores. Distributed lag models are well suited to this question because they allow exposure effects to vary flexibly over time rather than imposing arbitrary trimester boundaries. The Bayesian extension allows the shape of the exposure–response function and the magnitude of effects to differ across subgroups, such as infant sex or region. The primary estimand is the expected change in BWGA z score associated with a 1 μg/m³ increase in PM2.5 exposure at each gestational week, as well as the cumulative association across pregnancy. Models adjust for a standard set of maternal and environmental covariates, and missing covariate data are addressed using multiple imputation. Sensitivity analyses exclude sibling births to assess the influence of within-family clustering. While the methods are observational and do not attempt causal identification through quasi-experimental variation, they represent a strong descriptive approach for identifying temporal patterns in exposure effects.

Findings/Size Effects

In the full sample, higher prenatal PM2.5 exposure is associated with lower birth weight for gestational age, with a cumulative effect of approximately −0.06 standard deviations in BWGA z score per 1 μg/m³ increase in PM2.5 across pregnancy. The distributed lag models identify a critical window in very early pregnancy, concentrated in gestational weeks 1 through 5, during which exposure is most strongly associated with reduced fetal growth. When stratified by infant sex, the early pregnancy window persists primarily among male infants, with a similar cumulative effect size (−0.06 standard deviations), while associations among female infants are smaller and less precisely estimated. Analyses by race and ethnicity show little evidence of differential susceptibility after accounting for region. In contrast, analyses by geographic region reveal meaningful heterogeneity: negative cumulative associations are observed in the Northeast, Midwest, and South, with region-specific critical windows ranging from early to mid-pregnancy, while associations in the West are smaller and in some models slightly positive. Although the estimated effects are modest at the individual level, they are meaningful in a population context given the widespread nature of PM2.5 exposure.

Conclusion

This study makes an important contribution by demonstrating that the timing of prenatal air pollution exposure matters for fetal growth and that early pregnancy appears to be a particularly sensitive period. By leveraging a large, diverse cohort and high-resolution exposure data, the authors strengthen the external validity of prior findings and show that associations persist even at PM2.5 levels near or below current regulatory standards. The statistical methods are appropriate for identifying windows of susceptibility and represent an advance over simpler exposure averaging approaches. At the same time, the study remains observational and does not use causal inference designs or randomized interventions, limiting the strength of causal claims. Future research exploiting policy changes, natural experiments, or other quasi-experimental variation would strengthen inference about the causal impact of PM2.5 exposure timing. Overall, the findings reinforce the importance of considering developmental timing in environmental health research and suggest that existing regulatory thresholds may not fully account for risks during early pregnancy.