Prenatal exposure to PM_2.5 and childhood cognition: Accounting for between-site heterogeneity in a pooled analysis of ECHO cohorts in the Northeastern United States

Background: Emerging studies have investigated the adverse health effects of PM_2.5 using data from multiple cohorts, and results often are not generalizable across cohorts. We aimed to assess associations between prenatal PM_2.5 and childhood cognition in two U.S. cohorts while accounting for between-site heterogeneity. Methods: Analyses included 348 mother-child dyads enrolled in the dual site (New York City and Boston) PRogramming of Intergenerational Stress Mechanisms (PRISM) cohort and in the First Thousand Days of Life (FTDL) study (Northern Virginia) participating in the Environmental influences on Child Health Outcomes (ECHO) national consortium. Residential prenatal PM_2.5 exposure was estimated using a validated satellite-based model and childhood cognition was measured using the NIH Toolbox Cognition Battery at three to eight years of age. We used a log-linear model applied to contingency tables formed by cross-classifying covariates by site to examine between-site heterogeneity using 3^rd trimester PM_2.5 exposure, age-corrected cognition scores, and covariates potentially causing heterogeneities. Multivariable linear regression models informed by the combinability analysis were used to estimate the coefficients and 95% confidence intervals (CIs) for the association between 3^rd trimester PM_2.5 exposure and age-corrected cognition scores (mean = 100, SD = 15). Results: The log-linear model indicated that inter-study associations were similar between PRISM-NYC and FTDL, which were different from those in PRISM-Boston. Accordingly, we combined the data of PRISM-NYC and FTDL cohorts. We observed associations between 3^rd trimester PM_2.5 and cognition scores, findings were varying by site, childsex, and test. For example, a 1 μg/m³ increase of 3^rd trimester PM_2.5 was associated with −4.35 (95% CI = −8.73, −0.25) mean early childhood cognition scores in females in PRISM-Boston. In the pooled NYC + FTDL site, the association between PM_2.5 and childhood cognition may be modified by maternal education and urbanicity. Conclusions: We found associations between prenatal PM_2.5 and impaired childhood cognition. Since multi-site analyses are increasingly conducted, our findings suggest the needed awareness of between-site heterogeneity.