We used an ANOVA to test whether the interaction model (model 4) explained significantly more of the variation in the phenotype than the additive model (model 3). A summary of the cross-cohort GWAS and MWAS predictions is presented in Table 1.