Mendelian randomization applied to cancer: What it has and has not achieved

During the past decade, Mendelian randomisation (MR) has become one of the most frequently used tools for causal inference in modern epidemiology. By leveraging information on germline genetic variants as instruments for exposures, either using individual or summary-level data, MR applied to cancer outcomes aims to approximate features of a randomised controlled trial (RCT) whether to appraise evidence for established risk factors, examine estimands difficult to address in conventional designs (such as lifetime effects of an intervention), or identify novel causes of cancer. Interaction-based MR designs can in principle identify causality even when an exposure is ubiquitous within a population. Recent enthusiasm has been fuelled by the wide availability of results from genome-wide association studies (GWAS) from large biobank datasets and large cancer consortia, as well as summary-statistic MR workflows, which have driven the exponential growth in publications. Yet nearly two decades after the first MR analyses of cancer risk factors, important questions remain as to whether MR has uncovered any truly new causes of cancer or led to discoveries that contributed to cancer prevention policy or practice.