Mendelian randomisation as an instrumental variable approach to causal inference

In epidemiological research, the causal effect of a potentially modifiable phenotype or exposure on a particular outcome or disease is often of public health interest. Randomised controlled trials to investigate this effect are not always possible and inferences based on observational data can be distorted in the presence of confounders. However, if we know of a gene with an indirect effect on the disease via its effect on the phenotype, it can often be reasonably assumed that the gene is not itself associated with any confounding factors - a phenomenon called Mendelian randomisation. It is well known in the economics and causal literature that these properties define an instrumental variable and allow estimation of the causal effect, despite the confounding, under certain model restrictions. In this paper, we present a formal framework for causal inference based on Mendelian randomisation where the causal effect is defined as the effect of an intervention. Furthermore, we suggest a graphical representation of the data situation using directed acyclic graphs so that model assumptions can be checked by visual inspection. This framework allows us to address limitations of the Mendelian randomisation technique that have often been overlooked in the medical literature.