Mendelian randomisation and the goal of inferring causation from observational studies in the vision sciences

Randomised controlled trials (RCTs) allow reliable causal inferences to be drawn regarding the effectiveness of specific interventions. However, they are expensive to carry out, and not all exposure-outcome relationships can be tested in an RCT framework: for example, it would be unethical to deliberately expose participants to a putative risk factor, or the time-scale involved may be prohibitive. Mendelian randomisation (MR) has been proposed as an alternative approach for drawing causal inferences, with the major advantage that the method can often be applied to existing, cross-sectional study datasets. Therefore, results from an MR study can be obtained much more quickly and cheaply than through an RCT.

The validity of causal inferences from an MR study are dependent on two key assumptions, neither of which can be tested fully. Nevertheless, several approaches have been proposed in the last 3 years that either highlight questionable results, or provide valid causal inference if the necessary assumptions are met only in part. Compared to certain other areas of clinical practice, the ophthalmic research community has been slow to adopt MR.

An MR study cannot match an RCT in its strength of evidence for a claim of causality. However, MR still has much to offer. In some circumstances, an MR study can provide causal insight into research questions that cannot be addressed by an RCT, while more generally, an MR study can be used to evaluate the supporting evidence before deciding to embark on a lengthy and costly RCT.