There's been a lot of discussion recently about the misuse of MR, except the proliferation of incorrectly drawing your assumptions doesn't seem to stem from the recent explosion in papers. Rather, many high impact papers seminal to the field feature an "incorrect" arrow a 🧵
In MR, IV2 is the assumption of no confounding of the relationship between the instruments and the outcome (independence assumption) Genetic variants are randomly allocated at conception, meaning they can't be confounded by lifestyle or biological factors
Rather, we want to ensure the instrument-outcome association is unconfounded by factors that can affect the likelihood of inheriting a genetic variant, such as population stratification or assortative mating
When we're illustrating our assumptions this means there should be an arrow *from* confounders *to* the instrument (C➡️G) All too often, we're seeing it the opposite way around, *from* instruments *to* confounders (G➡️C)
However, this simply illustrates an example of a horizontal pleiotropy pathway (i.e. an effect of the instrument on the outcome, through a pathway other than the exposure) This isn't wrong to include when illustrating your assumptions, but it is the wrong way to illustrate IV2
We know this can be a source of contentious debate amongst co-authors, @emmylooroll and I frequently have this debate with our colleagues Hopefully through our short letter you now understand why the correct illustration of IV2 should go confounder ➡️ instrument
There aren't many conversations I have on Whatsapp that I feel the need to turn into a published letter in @IntJEpidemiol but this was one of them! Read our letter here ⬇️ and we look forward to a proliferation in the correct illustrations of IV2 doi.org/10.1093/ije/dy…
@alicerosecarter @IntJEpidemiol Yes! The incorrect C->G arrow is so common & so not the right way to depict that assumption! It always bothers me. Thanks for writing this.
@alicerosecarter @IntJEpidemiol Agree your diagram is a better illustration of IV2, but IV2 could also be violated by a confounder of G and Y that is unrelated to X (such as ancestry). So I'd hesitate to say your diagram is fully correct - it illustrates an example of IV2 violation, but not all examples.