Fig. 3.

Causal pathways in an MR experiment. A causal effect of exposure (X) on outcome (Y) can be inferred if three core assumptions are met. These assumptions concern the genetic instrumental variable, Z, and state that: (i) Z has to be robustly associated with the exposure variable, (ii) Z cannot be related to any common causal factors of X and the outcome Y (these are labelled U), and (iii) Z may only be related to Y through X (Solovieff et al. 2013). The latter two assumptions can be summarised as the absence of pleiotropy for the instrumental variable.