Two-Sample Mendelian Randomization: Association between Instrumental Variable and Outcome

Question

[Nikki]

Can someone tell me when selecting the instrumental variables for Two-sample MR analysis, why we don't care the association between Instrumental Variable and Outcome variable? If G is a valid instrumental variable, doesn't it apply that the effect of G on the outcome Y is fully mediated by the exposure variable X?

My understanding is that when we perform the MR, we don't know that whether X is causally associated with Y, or their association is just because the confounding variable. And if we select the instrumental variable not only strongly associated with exposure but also associated with the outcome, it is actually leading the selection bias, am I right?

Answer 1

[Zeng Jingyu]

Hi Nikki,

I have to say that I didn't fully understand your question. You said "don't care the association between Instrumental Variable and Outcome variable" but in my opinion, we care about it. Normally we say the instrumental variable should not be associated with the outcome, so when choosing the G (instrumental variable), we simply exclude the SNPs that are significantly associated with Y (outcome).

And your understanding is right. I guess that's why there's not many of us answer this question.

To know more about the practice, I recommend an article for you:

Causal effects of blood lipids on amyotrophic lateral sclerosis: a Mendelian randomization study by Ping et al.

That's the first article I read when learning MR.