Handling of multicohort data in pooled analysis of environmental exposures: confounding by cohort of the relationship between perfluoroalkyl substances and birthweight in the ECHO Program.

The relationship between exposure to perfluoroalkyl substances (PFAS) and birthweight remains unclear. Pooling data across multiple cohorts can increase power, leading to more representative populations and exposure distributions, but confounding by cohort can be a major source of bias. To understand this potential bias, we assessed the relationship between exposure to 5 PFAS and birthweight utilizing data from 5480 mother-infant dyads across 17 Environmental influences on Child Health Outcomes (ECHO) Cohort sites. The relationship was assessed in several ways: covariate-adjusted models without cohort adjustment as well as adjustment via fixed and random effects. Findings from analysis with cohort adjustment resulted in significant inverse relationships for 4 of the 5 PFAS considered. Adjustment via fixed and random effects produced similar findings. Failure to adjust for cohort resulted in biases of varying directions and magnitudes depending on the PFAS considered. Results were supported in simulated data. In this study, we saw evidence of confounding bias by cohort even after adjustment for covariates, while adjustment by both fixed and random effects for cohort resulted in comparable results.