Effect modification in epidemiologic studies of low-level neurotoxicant exposures and health outcomes.

Little attention has been invested in exploring the possibility that the nature or magnitude of a neurotoxicant's health impact on children depends on host characteristics (e.g., sex, age) or contextual factors (e.g., socioeconomic status, other chemical exposures). Such effect modification is a property of a true association, and should be distinguished from confounding. In epidemiologic studies of children, most efforts to identify effect modification have been unsystematic, pursued as part of data analysis rather than of study design. As a result, most samples have insufficient statistical power to characterize effect modification with adequate precision. This may contribute to an inconsistency in results across studies. Failure to assess effect modification adequately may also lead to invalid inferences. If the magnitude of an association between a neurotoxicant exposure and a particular end point varies across strata of a third factor, an estimate that summarizes the association across strata of this factor will be inappropriate, overestimating the association in a stratum in which the association is absent, and underestimating it in a stratum in which it is present. Until such dependencies are identified, our understanding of the mechanism(s) of a compound's neurotoxicity will remain incomplete, as will the knowledge base required to formulate public policy that adequately protects the most sensitive subgroups of the population.