An introduction to instrumental variables--part 2: Mendelian randomisation.

In the first part of this series, it was highlighted how even though randomised controlled trials can provide robust evidence for therapeutic interventions, for many types of exposure it may not be either practical or ethical to randomise patients to such studies (see part 1). Instrumental variables (IV) analyses have been increasingly employed in recent times in epidemiology to investigate the potential causal effects of an exposure. An IV is a variable that can realistically mimic the treatment allocation process in a randomised study and is assumed to be not directly related to outcome, except through the direct effect of treatment and not related to outcome through either measured or unmeasured confounders. As discussed in the first article, IV analyses can be useful in estimating direct treatment effects provided that the chosen instrument is strong. A particular type of IV analysis where a specific genetic variant has been used as the instrument known as 'Mendelian randomisation' has become increasingly common. The aim of the second part of this statistical primer is to outline the approach to Mendelian randomisation and some of the advantages and disadvantages of this approach.