Chunk #56 — Reasons to be Concerned about the Published cGxE Literature — Problems with the Recipe: Statistical Concerns in cGxE Research — cGxE versus gene-environment correlation (rGE)
In summary, although the statistical approaches for modeling interactions are well established, having confidence in the statistical validity of an interaction requires due diligence on the part of the investigator. These include attention to scaling issues, characterizing the underlying linearity of the relationships under investigation and determining whether nonlinear models are necessary, controlling for relevant confounders including various forms of rG-E, and insuring plotted results are not unduly influenced by the constraints imposed for rendering an easy-to-interpret graph. Perhaps the greatest challenge is to minimize the likelihood that an observed interaction is not a Type 1 error given that various data sets have a large number of candidate Gs and candidate Es, there is considerable flexibility in approach to analysis, and under most plausible conditions, power to detect GxE is likely to be low. Many of the issues described above (as well as some others) are described by Roismann et al. (2012) who provide a list of thoughtful guidelines for addressing various issues such as characterizing whether an obtained interaction is of a cross-over type (and to the extent the magnitude of the cross-over is meaningful), the problem of nonlinearity, and Type 1 errors.
