What is the biological reality of gene-environment interaction estimates? An assessment of bias in developmental models.

backgroundStandard models used to test gene-environment interaction (G × E) hypotheses make the causal assumption that there are no unobserved variables that could be biasing the interaction estimate. Whether this assumption can be met in nonexperimental studies is unclear because the interactive biological pathways from genetic polymorphisms and environments to behavior, and the confounders that can be introduced along these pathways, are often not delineated. This is problematic in the context of studies focused on caregiver-child dyads, in which common genes and environments induce gene-environment correlation. To address the impact of sources of bias in G × E models specifically assessing the interaction between child genotype and caregiver behavior, we provide a causal framework that integrates biological and statistical concepts of G × E, and assess the magnitude of bias introduced by various confounding pathways in different causal circumstances.

methodsA simulation assessed the magnitude of bias introduced by four types of confounding pathways in different causal models. Unadjusted and adjusted statistical models were then applied to the simulated data to assess the efficacy of these procedures to capture unbiased G × E estimates. Finally, the simulation was run under null effects of the genotype to assess the impact of biasing sources on the false-positive rate.

resultsCommon environmental pathways between caregiver and child inflated G × E estimates and raised the false-positive rate. Evocative effects of the child also inflated G × E estimates.

conclusionsGene-environment interaction studies should be approached with consideration to the causal pathways at play and the confounding opportunities along these pathways to facilitate the inclusion of adequate statistical controls and correct inferences from study findings. Bridging biological and statistical concepts of G × E can significantly improve research design and the communication of how a G × E process fits into a broader developmental framework.