Graboy has provided the citation I would have given, as well as an excellent explanation. I’m not sure what you mean by “heart disease is clearly not Mendelian”.
A clinical score changing with treatment is not unconfounded by mendelian randomization. When the genetics are clearly more complex than what you are mathematically randomizing for, the control doesn't solve the confounding. eg you haven't suddenly "proven" the effects are non-genetic. We already knew heart disease is non-mendelian. But showing something is non-mendelian doesn't mean you've shown it's not genetic. I hope that clarifies, because I'm not sure I can explain it to you in simpler terms.
This is quite the tone to take when the actual point being made has demonstrably sailed over your head, considering the reference provided to you explains it very clearly.
Which is more likely - nigh on every lipidologist, cardiologist and nutrition researcher is wrong, or you might have made a mistake yourself?
> Mendelian is characterized by effects having strong influence from a single gene. Heart disease is clearly more complex than gene -> heart disease
This seems like a misunderstanding. A single SNP clearly can affect CVD risk, that’s precisely what the paper shows. The assumptions required for an MR study to be valid do not include “the outcome must only be affected by a single gene and no other gene”. It’s required that there’s no pleiotropy present in the exposure (I.e. the SNPs). The exposure here isn’t heart disease, it’s SNPs that affect LDL-c levels, and the outcome being measured is CVD. So your point doesn’t pose an issue for the study and the inferences it makes.
But honestly - just read the paper. I think both that paper and the EAS consensus paper are very approachable.
The cited study addresses this, which is why I pointed to figure 3. They argue that if genes were causing heart disease not through LDL in any meaningful way, you wouldn't expect such a clean dose-response consistency across different genetic variants - it would be more jagged.
[1] https://en.wikipedia.org/wiki/Mendelian_randomization#Defini...
Here is a simple primer on mendelian randomization: https://www.psomagen.com/blog/what-is-mendelian-randomizatio...
Please review the key principles and assumptions section. Using MR to control for genetic confounding of heart disease fails all assumptions. Thats why it quite directly does not follow.
This is why the paper presented does not support the claim that LDL is the sole source of heart disease. I'd be interested to hear what the authors of that paper (which is legitimate) think about it being used to support the OP's claim because "mendelian randomization".
Is that what we were arguing about? I guess it was. At some point in thinking about this my frame must have shifted into agreement with you. Of course there are other causes of heart disease besides LDL, like blood pressure, duh. The smooth dose response is about the particular gene not being linked to heart disease through something other than LDL, roughly.
In general, Ference’s explanation just seems more parsimonious than Schooling’s. It’s possible that they’re right, but I think they would need to show that the specific genes in Ference’s study are affected by age in this way.
I wouldn’t claim to be a great expert in MR, though. I can just about keep up with surface level understanding of them but once you get into the nitty gritty stats, pleitropy testing etc it’s a bit over my head tbh.