Maternal exposure to black carbon and nitrogen dioxide during pregnancy and birth weight: Using machine-learning methods to achieve balance in inverse-probability weights

Abstract

Background: Low birth weight is associated with increased risks of health problems in infancy and later life. Among the epidemiological analyses suggesting an association between air pollution and birth weight, few have estimated the effects of black carbon (BC) or together with nitrogen dioxide (NO₂), and even fewer studies have used causal modelling.

Methods: We examined 1,119,011 birth records between 2001/01/01 and 2015/12/31 from the Massachusetts Birth Registry to investigate causal associations between prenatal exposure to BC and NO₂ and birth weight. We calculated mean residential BC and NO₂ exposures 0-30, and 31-280 days prior to birth from validated spatial-temporal models. We fit generalized propensity score models with gradient boosting tuned by a new algorithm to achieve covariate balance, then fit marginal structural models with stabilized inverse-probability weights.

Results: Throughout pregnancy, the average birth weight would drop by 17.0 g (95% CI: 15.4, 18.6) for an IQR increase of 0.14 μg/m³ in BC and would independently drop by 19.9 g (95% CI: 18.6, 21.3) for an IQR increase of 9.8 ppb in NO₂. Most of the negative effects of BC on birth weight are from 0 to 30 days before the delivery date. The estimated odds ratio of low birth weight for every IQR increase during the entire pregnancy was 1.131 (95% CI: 1.106, 1.156) for BC and 1.082 (95% CI: 1.062, 1.103) for NO₂.

Conclusions: We found that prenatal exposures to both BC and NO₂ were associated with lower birth weight.

Keywords: Birth weight; Black carbon; Causal method; Machine learning; Nitrogen dioxide; Propensity score.