stratify on the exposure, and stratification (which can be considered a form of statistical adjustment) can introduce other potential biases in certain circumstances (see Box 3) [10]. Nevertheless, if differences in the magnitude of association observed in exposed and unexposed groups are very large, this is convincing evidence of a causal pathway via the exposure. Therefore, a logical extension of Mendelian randomization is that any GWAS will, in principle, identify exposures that are causally associated with the outcome of interest in that GWAS.
