SI have selected environmental exposures relevant to their primary outcome to be utilized in tests for G × E interactions (Table I). Designing a sufficiently powered study and locating an appropriate external study for replication are just two examples of major barriers to uncovering true interactions. When applying the standard logistic regression test for interaction, most individual studies will be limited to detecting interactions of large effect sizes. Nevertheless, new methods for G × E interaction testing have been and will continue to be developed to boost statistical power for detection while maintaining low type 1 error [Chatterjee and Carroll, 2005; Kraft et al., 2007; Murcray et al., 2009; Weinberg, 2009]. Methods based on logistic regression continue to dominate the field and generally test for interactions specifically, or main genetic associations allowing for heterogeneity in genetic effect across environment strata. Model-free or machine learning approaches in the context of GWAS are relatively new and currently computationally expensive. The performance of each method will vary with the distributional assumptions underlying the phenotypic outcome, the environment and their suspected interaction. The Analysis
