Mendelian Randomisation – better than epidemiology?

Executive summary

* This post introduces a technique called Mendelian Randomisation (MR).

* We are very familiar with the limitations of epidemiology (population studies). MR tries to overcome the ‘association not causation’ limitation by capitalising on the fact that gene versions are randomly inherited (which could result in randomly distributed confounders).

* The post uses a paper from October 2024, which used MR to try to claim that "reducing dietary saturated fat intake is beneficial for preventing and managing Alzheimer’s disease."

* This post uses that paper in two ways – i) to try to examine the research in my usual way and ii) to show how difficult it is to challenge studies that use MR compared to those that use epidemiology.

* Studies using the MR technique are like black holes. The information presented in population studies is frequently absent in MR studies. How were associated genes selected? How strong were associations between genes and nutrient intake? Can we really claim that genes determine dietary fat intake? And more.

* MR will increasingly be used in diet-disease claims. That's a key reason why I wanted to cover this technique/paper. As with the increasing number of modelling papers used to make diet-disease claims, it is going to be difficult for reviewers like me to dissect these studies and thus more difficult to interpret findings.

* I am most grateful to Dr Adrian Soto-Mota for his enormous help with this note. Adrian is an academic and medical doctor who also trained as a data scientist at Harvard and obtained a PhD in Physiology at Oxford. Adrian is presenting on the subject of Mendelian Randomisation at the Collaborative Science Conference in February 2025 (https://cosci.org/) – just the expert this note needed – thank you!