Causal Relationship Between Diet, Lipids, Immune Cells, and Chronic Fatigue Syndrome: A Two-Mediation Mendelian Randomization Study, 2025, Li et al

[Murph]

I fear that maybe mendelian randomisation studies suck because it is cheap to do. You don't need any samples or pipettes, microscopes or a lab. You need nothing more than a database and a copy of some free stats software. Barriers to entry are low and so, even though the technique can be incredible when done right, in practice it attracts people with no real background in what they are looking at, nor plans to take their study further.

 

[Midnattsol]

hmmm. Affinity for breakfast is a novel concept in medical literature.

View attachment 26721

This is the sort of groundbreaking research we need. We should petition to get the name of this illness changed to breakfast affinity syndrome.

[SARCASM TAG INCLUDED HERE FOR FEAR OF MISUNDERSTANDING !!

The wording might be unique, but eating breakfast or not has been a hot topic in nutrition.

Which is also the case with cheese consumption, as saturated fat increase LDL cholesterol and cause cardiovascular events are standard guidance. However studies on dairy can be all over the place (dairy is a common source of saturated fat), and so new hypotheses are made up about dairy being «different». Pork is also seen as different and can be defined as a red meat in some cases and white in others (guidance is typically to avoid red meat) making it impossible to compare studies when only total intakes of specific food groups are reported.

I haven’t read the study, I just wanted to say it should be criticized for the right reasons

 

[Simon M]

Doesn't look like they've done Mendelian randomisation properly.

At the heart of Mendelian randomisation is the "instrumental variable", IV. This is a genetic factor that you know impacts the outcome of interest. This is one example I know:
– George Davey Smith, a pioneer of Mendelian randomisation, looked at the impact of vitamin D on the risk of multiple sclerosis. This was a hypothesis driven by data showing that the risk of MS increases significantly as you move further from the equator. We also know that vitamin D levels drop as you go further north, because with less daylight, the body makes less vitamin d.

George Davies Smith used the vitamin D receptor genes (maybe other genes too) ast the instrumental variable. Because they knew that variants of the vitamin D receptor were linked to effectively different levels of vitamin D (or different impacts of vitamin D on the body). It found that the risk of MS was indeed linked to the level of vitamin D (or impacts back on the body).

So this validated the hypothesis that vitamin D levels are a risk factor for MS.

I don't see anything like that in the study

 

[arnoble]

Data sources


7363 is a lot of samples. I'm also curious where they're from. I'm not sure what IEU is. Maybe this: https://gwas.mrcieu.ac.uk/

I asked Google's Gemini 2.5 and it told me the paper says: "2.1.2. GWAS of CFSThe summary data for CFS (ebi-a-GCST006764) were obtained from the IEU GWAS database."

... I cannot for the life of me see that in the paper! nor in the Supplementary Information here
... and if I enter that ebi-a-GCST006764 study into gwas.mrcieu.ac.uk/datasets here the site gives nothing

All very confusing.

I did find this though - although have no idea how to ascertain how the data was collected ... https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90038694/

 

[forestglip]

I did find this though - although have no idea how to ascertain how the data was collected ... https://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90038694/

Yeah, there's another CFS dataset on there as well:

ukb-b-8961: Non-cancer illness code, self-reported: chronic fatigue syndrome

They both have around 2000 controls and 500,000 controls, so I'm guessing they're both using UK BioBank data.

 

[forestglip]

Potentially some "tortured phrases" indicative of paper mills (found by AI):

To assess the reliability and stability of knowledge association, we analyzed the results for sensitivity.

I think that's meant to be "causal association"

Figure 2 caption:

Causal relationship between different types of lipids and CFS. SNPs: number of SNPs. CI, confidence interval; OR, ratio of ratios.

It's weird to refer to an odds ratio as "ratio of ratios".

 

[jnmaciuch]

I fear that maybe mendelian randomisation studies suck because it is cheap to do. You don't need any samples or pipettes, microscopes or a lab. You need nothing more than a database and a copy of some free stats software. Barriers to entry are low and so, even though the technique can be incredible when done right, in practice it attracts people with no real background in what they are looking at, nor plans to take their study further.

I think you’re right on the money here. I’ve seen an explosion of MR papers across a bunch of fields probably for this exact reason. It’s just another way to generate a bunch of publications in a short time frame.

MR is a very valuable technique but it needs to meet several criteria otherwise the findings are utterly useless.

And in this case, you by definition cannot use an MR study to do an association with behavioral traits. Unless there happens to be an allele that is so highly correlated with intense cheese craving that having it basically guarantees an abnormal amount of cheese consumption beyond any cultural/societal confounders. Which would be quite the interesting finding if they happened to find a gene like that, but somehow I doubt that they did.