Therefore, care should be taken to identify and appropriately control for genetic ancestry. Confounding may also arise if the variant has pleiotropic effects which influence the outcome other than through Dasatinib cost the exposure of interest, or if the variant is in linkage disequilibrium with another genetic variant which also influences the outcome [20••]. In such cases, one cannot

be confident that any ‘causal’ effect observed operates through the exposure of interest. In some MR studies of lifestyle behaviours, it may be possible to perform a test of pleiotropy by investigating associations of the genetic variant with the outcome in individuals not exposed to the behaviour. This has been demonstrated in MR studies of alcohol use in East Asians, which have stratified analyses by sex. The alcohol-related variant influences blood pressure in males (who consume alcohol) but not in females (who tend not to consume alcohol in many East Asian cultures for social and historical reasons), indicating that the likely mechanism of the genetic effect on blood pressure is through alcohol consumption [34•]. However, whilst stratifying on an exogenous variable such as sex, as described above, can be a useful tool in some MR studies, care must be taken not to reintroduce

confounding through collider bias 35• and 36]. This can occur when MR analyses are stratified on the measured exposure of interest Amobarbital and can amplify or mask associations between the genetic variant and outcome within the exposure strata [37]. A further potential concern is the possibility of canalization, which is the process of developmental compensation selleck screening library to buffer against the effects of disruptive genetic or environmental influences during development [9••]. If exposure to elevated

levels of a risk factor during foetal development or post-natal growth results in tissue changes which compensate for this, the genetic variant will still associate with the risk factor of interest, but any potential effects on a disease outcome may be reduced. However, canalization is less problematic for exposures which tend to occur later in development, such as smoking and alcohol consumption [7]. There are a number of other statistical issues in relation to MR, particularly surrounding the use of two-stage instrumental variable analysis (e.g., weak instrument bias). These are beyond the scope of this review, but are discussed in detail elsewhere 38, 39 and 40]. Inferring causation from observational data is notoriously problematic. Although MR relies on certain assumptions that may not always apply, it nevertheless has the potential to dramatically advance our understanding of the causal role of modifiable environmental exposures on a variety of outcomes. As GWAS continue to reveal variants associated with a range of behavioural phenotypes, the applications of MR will grow.