Should we adjust for delivery hospital in studies of air pollution and pregnancy outcomes?

BACKGROUND: In conducting a study of ambient air pollution and pregnancy outcome in New York City, we identified delivery hospital as a potential confounder, given its association with both maternal residence and therefore air pollution exposure, and with clinical practices and as a potential marker of outcome misclassification in the coding of pregnancy complications. Motivated by evidence that adjustment for delivery hospital affected associations between air pollution and pregnancy outcome, we undertook a detailed empirical examination of the role of delivery hospital that warrants consideration by others addressing this topic. METHODS: In a study of air pollution and pregnancy outcome, we identified births from 2008 to 2010 to residents of New York City and, after restrictions, included 238,960 in the analysis. Air pollution exposure estimates for ambient fine particles (PM(2.5)) and nitrogen dioxide (NO(2)) were derived from a community-wide exposure study and assigned based on geocoded maternal residence. We examined the impact of adjusting for delivery hospital and explored the relationship between delivery hospital and both exposure and pregnancy outcomes. RESULTS: Statistical adjustment for delivery hospital markedly attenuated the relationship of air pollution with birth weight and gestational hypertension, with smaller effects on preterm birth and preeclampsia. Delivery hospital was associated with estimated maternal air pollution levels after adjusting for individual-level patient characteristics, more strongly for PM(2.5) than for NO(2). Delivery hospital predicted pregnancy outcome after adjustment for individual attributes, with larger hospitals and those that managed a greater volume of complicated cases having lower birth weight, more medically indicated preterm births, and more diagnosed gestational hypertension. Evaluation through the use of directed acyclic graphs illustrates the potential for adjustment for hospital to reduce residual spatial confounding, but also indicates the possibility of introducing bias through adjustment of a mediator. CONCLUSIONS: Based on these results, delivery hospital warrants closer consideration in studies of air pollution and other spatial factors in relation to pregnancy outcomes. The possibility of confounding by delivery hospital needs to be balanced with the risk of adjusting for a mediator of the air pollution—pregnancy outcome association in studies of this type.