I wanted to ask you a question surely very basic but I hope you can help me to solve it.
My thought is the following:
I have a mendelian randomization study on e.g. LDL
My doubt arises about the external validity of the causal results obtained for the outcome X
And the question is the following;
Would the causal results be applicable to a population completely different or significantly different from the study population?
I believe that it would not be applicable since those different variables could function, let’s say, as “enzymes” that are the true culprits of the casual effect. Is this so, or am I wrong?
Could you help me to understand this? Thank you very much
Swanson, Sonja A.a,b; Tiemeier, Henninga,c,d; Ikram, M. Arfana; Hernán, Miguel A.b,e,f. Nature as a Trialist?: Deconstructing the Analogy Between Mendelian Randomization and Randomized Trials. Epidemiology 28(5) 653-659, September 2017. | DOI: 10.1097/EDE.0000000000000699
Thank you very much , I have read the document 3 times and I do not see in my head the answer to my question​:frowning:.
My doubt is : I am not sure if using the same exposures in the Mendelian study we would get the same results if the population under study is the population of a biobank X or another database.
That is, if the entire population 1 is subjected to contamination, I expect that the high exposure will be shown to be causal because the population will be endothelially impaired.
But if population 2 is not subjected to contamination in any case, I understand it possible that the high exposure will not show causality.
That is, I would conclude that the generalization is contingent just like an RCT about which populations are similar.
I understand your question better now. Yes, you are correct. Transportability of a study effect requires several assumptions, including the same distribution of effect modifiers between the populations. In your example, it sounds like there is an effect modifier present in one group that is not present in the other so the effect would not be transportable.
This paper has more about the assumptions for transportability: Degtiar, I., & Rose, S. (2023). A review of generalizability and transportability. Annual Review of Statistics and Its Application, 10(1), 501–524.
If we’re starting from a consensus that LDL is on the causal path toward atherosclerosis and if you’re asking about the impact of LDL in populations that lack other CV risk factors, then I have some articles you might be interested in and can dig them up for you if you’d like. On the other hand, if the purpose of this post is to cast doubt on the role of LDL in atherogenesis, I don’t see much point in providing more articles. A person who isn’t already convinced by the overwhelming existing evidence isn’t likely to be convinced by any additional articles I might provide. I’ll be guided by your response.
Speaking in general terms, regardless of how much evidence any model accumulates, I think there should be some amount of skepticism to minimize confirmation bias. If one only looks for confirmatory evidence then that is all one finds even if there is evidence to the contrary and looking for gaps in a model is what progresses science.
Many thanks to both of you, apologies because I thought I had answered and in fact I had not.
Regarding @ESMD 's reply, just clarifying that denying and clarifying and deepening are different things.
That is, you talk about statistical causality as higher probability of disease between groups , to physiological causality or sufficient cause at X level are substantially different things to me
I have been studying lipids for 12 years to talk about humility as Libby talks about, and how well we know things that:
Given ApoB , more LDLc is probably better actually.
LDLc adjusted by ApoB tells us little or nothing. Which makes us look at the proteome and not so much the lipidome as is classically done. Many lines of evidence in this regard.
LDL-TG is substantially more informative than LDL-c. Again it points to the proteome and its “quality.”
We know that even ApoB is no longer associated with risk when we look at ApoE/apocIII adjusted
-We have recently seen a Mendelian randomization that eliminates healthspan and lifespan relationships of LDLc.
You name it
My question addresses that point to try to read as well as possible the Mendelian randomization.
If we look at Mendelian randomization to two groups with bifurcation malformations as reported years ago by Velican, is LDL-c/ApoB, the clear example of matches/oxygen, with oxygen being equally distributed between the groups?.Honest question Here!
In fact, we know recently that LDLr problems in HeFH generate a special underlying problem in the endothelium that makes it subservient to the disease and interacts with LDL
The question is HUMBLE in seeking response from people who know the most about statistics and reading that specific line of evidence. I hope it is understood in this way.
653-659, September 2017. | DOI: 10.1097/EDE.0000000000000699
, I have read the document 3 times and I do not see in my head the answer to my question​:frowning:.