Implementing Mendelian randomization techniques can be challenging, principally because single genetic variants (and even polygenic risk scores) typically capture only a small proportion of variance in the exposure of interest. An ideal instrument would exactly mimic the exposure of interest without being associated with confounding variables, but this is, of course, impossible in practice. Genetic variants are, therefore, generally weak instruments, meaning that very large sample sizes are required to attain adequate statistical power. Historically, datasets were required with information on genotype, outcome, and exposure of interest in order to run such studies. Methodological developments now allow the application of Mendelian randomization across two different samples if no single sample is available that includes data on genotype, the exposure, and the outcome.
