For Mendelian randomization to be useful, genetic assignment to the variable of interest must exist, and the genes influencing this variable cannot directly impact the outcome through other pathways.

“Now there's a caveat here which is to for a mandelian randomization to be useful a couple things have to be true. what you said has to be true. There has to be genetic assignment to the variable of interest. So that's true for many things. It's true for height. It's true for body composition. Uh it's true for uh many psychiatric disorders. I mean there's lots of things for which genetics play a significant role. Uh and LDL cholesterol turns out to be one of them. But this is the other important thing that people often forget when talking about MR, which is that those same genes cannot have a direct impact on another variable that affects the outcome of interest because if they do, your randomization just got wonky for sure.”