Performing Mendelian Randomization Using Structural Equation Models

Mendelian randomization (MR) is a method of estimating the causal effect of modifiable environmental exposures on medically relevant outcomes, identifying molecular biomarkers that are likely to be causal for disease, and... [ view full abstract ]

Mendelian randomization (MR) is a method of estimating the causal effect of modifiable environmental exposures on medically relevant outcomes, identifying molecular biomarkers that are likely to be causal for disease, and determining the suitability of drug targets for pharmacological intervention. However, MR studies are currently performed using very simple statistical methods based on e.g. two stage least squares and Wald ratios. These approaches lack flexibility to model more complicated causal networks involving many different variables, bidirectional relationships, and horizontal pleiotropy, which in some cases may invalidate analyses and bias estimates of causal effects. Structural Equation Modelling (SEM) is a very flexible statistical tool that allows the modelling of complex linear dependencies between variables and the estimation of causal effects. Despite the potential advantages, SEM has yet to have been employed in MR studies except in the simplest of situations. In this presentation I show how SEM can be combined with Mendelian randomization principles to estimate complicated causal effects that may not be easy to estimate using simple ratio or two stage least squares models. I illustrate the advantages afforded by SEM using large scale data from the UK Biobank Study.