*Geek Box: Two-Sample Mendelian Randomisation
Mendelian Randomisation (MR) reflects the fact that the genes you inherit from your parents were “assigned” to you randomly. MR studies use enormous genetic databases to determine how robust a genetic variant is associated with a particular exposure [e.g., how strong is a gene associated with HbA1c or with blood pressure). These are known as “genome-wide association studies” [GWAS]. If the incidence of disease with a particular genetic variant is higher or lower than the population base rate of that disease, this indicates the gene is associated with that disease in that population.
When a gene has been identified from GWAS, researchers will look to investigate the genetic associations for that gene/exposure with a particular outcome from population cohorts with genetic data, e.g., the UK Biobank cohort. When you are reading an MR study, you will see references to the study being “one-sample” or “two-sample” MR. This refers to whether the exposure and outcome were assessed in the same population cohort [“sample”] or two different cohort “samples”.
This is only possible with genetic studies, and would not be possible in other research designs, where the exposure and outcome must be measured in the same population followed up over time. Thus, a “one-sample” MR is one where the genetic variant “exposure”, and the outcome are assessed in the same sample, e.g., the UK Biobank cohort.
On the other hand, a “two-sample MR” is a type of MR where the exposure is measured in one genetic sample and the outcome is measured from another genetic sample. For example, the exposure could be measured in the UK Biobank cohort, while the outcome could have been measured using data from 23andMe. This has the advantage of providing vast datasets, reducing the potential for “Winner’s Curse” bias – where newly discovered genetic effects are consistently overestimated – and providing more statistical power to detect true associations.