Discriminating Myalgic Encephalomyelitis/Chronic Fatigue Syndrome and comorbid conditions using metabolomics in UK Biobank, 2024, Huang et al

[DMissa]

Can @melb and/or @DMissa comment a bit more on how the findings of Huang et al. 2024 link in with those of Missailidis, Armstrong et al. 2026 in simple terms please? (I'm aware team members overlap!)

I see these mentions of Huang 2024 in Missailidis 2026:


I'd like to hear more, in layman's words, of what you think of the Huang 2024 findings in light of Missailidis 2026?

The broad strokes that align are mentioned briefly in the paper but are left there. Didn't want to harp on about it too much because it would just be supposition. Can't make any confident or specific comment without more data. If we have primary cell data linked with biofluid data, collected together from the same people, it may be more possible to draw meaningful relationships between the studies. As it stands, what we have in one study is biofluid data, and in the other study transformed cell lines from a different cohort of people. Much too hard to dig into relationships between them in detail.

We have a couple of collaborative projects that may shed light here. Including a component of Chris's recent big grant

 

[Kitty]

We absolutely don't want fast track to false positives. But I'd probably prefer an err on side of the earlier recommendation of pacing management. Yes you may get some people being told to pace that don't need to.

Do you think that is a problematic point of view?

I don't find that problematic, I think it's eminently sensible when ME/CFS is suspected.

Thing is, people who're finding activity makes them feel ill will already (consciously or not) have started pacing to some extent. To have a physician endorse that as a reasonable course of action, at least whilst we see what's going on, would be really valuable.

What I'm not sure about is whether we need a test to endorse it as a reasonable course of action for physicians. If a patient's feeling ill and exhausted, advice to avoid overdoing it is just common sense.

Interesting debate, though!

 

[Evergreen]

Why do you find Supplementary Figure 7 reassuring?

What was dimly visible yesterday is obscured today - brain bunched. So I can't answer, sorry!

 

[Evergreen]

The broad strokes that align are mentioned briefly in the paper but are left there. Didn't want to harp on about it too much because it would just be supposition. Can't make any confident or specific comment without more data. If we have primary cell data linked with biofluid data, collected together from the same people, it may be more possible to draw meaningful relationships between the studies. As it stands, what we have in one study is biofluid data, and in the other study transformed cell lines from a different cohort of people. Much too hard to dig into relationships between them in detail.

We have a couple of collaborative projects that may shed light here. Including a component of Chris's recent big grant

Well that's good news!

Thank you for not speculating, and for explaining what data you would need to comment.

 

[Utsikt]

We absolutely don't want fast track to false positives.

But you’re literally saying you want to make the diagnosis faster and more accurate by using the test, and that doctors will learn to trust it because outcome is what matters.

Fatigue is one of the most common complaints at GPs. If the GPs with poor knowledge of ME/CFS (that you said could be helped by this) use your test for all those cases, you’d get an enormous amount of false positives.

Lets say Norwegian GPs perform 100,000 tests a year. That’s <0.6 % of the GP consultations (17.8M in 2024, with 1.2M excess compared to pre-covid expectations, so many will be LC related - study). For comparison, there are ~250,000 vitamin D tests a year for women aged 20-50 alone (source).

If just 1 % of the tests give a false positive for ME/CFS, you’d get 1000 false positives a year. That would make ~1/3 of all ME/CFS diagnoses a year definitive false positives (study on prevalence).

And if the doctors trust the test, the diagnosis might stick for a long time. Especially when most doctors don’t know what ME/CFS looks like, and they haven’t been taught because it isn’t needed because we have a diagnostic test for it..

But I'd probably prefer an err on side of the earlier recommendation of pacing management. Yes you may get some people being told to pace that don't need to.

Do you think that is a problematic point of view?

I don’t mind people being told to pace. But I do mind people being told they have ME/CFS when they don’t, and they might have something treatable instead that will be missed because your test that everyone trusts said so.

 

[MelbME]

But you’re literally saying you want to make the diagnosis faster and more accurate by using the test, and that doctors will learn to trust it because outcome is what matters.

Fatigue is one of the most common complaints at GPs. If the GPs with poor knowledge of ME/CFS (that you said could be helped by this) use your test for all those cases, you’d get an enormous amount of false positives.

Lets say Norwegian GPs perform 100,000 tests a year. That’s <0.6 % of the GP consultations (17.8M in 2024, with 1.2M excess compared to pre-covid expectations, so many will be LC related - study). For comparison, there are ~250,000 vitamin D tests a year for women aged 20-50 alone (source).

If just 1 % of the tests give a false positive for ME/CFS, you’d get 1000 false positives a year. That would make ~1/3 of all ME/CFS diagnoses a year definitive false positives (study on prevalence).

And if the doctors trust the test, the diagnosis might stick for a long time. Especially when most doctors don’t know what ME/CFS looks like, and they haven’t been taught because it isn’t needed because we have a diagnostic test for it..

I don’t mind people being told to pace. But I do mind people being told they have ME/CFS when they don’t, and they might have something treatable instead that will be missed because your test that everyone trusts said so.

Understand the concerns, important to consider.

False positives happen with just about every diagnostic tool. I believe there are certain standards to adhere to. Even the current diagnosis path has false positives. As an example, early MS can be misdiagnosed as ME/CFS, a good clinician will still look for follow up tests. The question for us is if the tool generates a net positive in patient care/outcome.

Typically specialists will still look for other explanations. Given that there is no treatment for ME/CFS beyond management, it would be far easier for a clinician to identify another reason for the patient experience.

I suspect a lot of patients, that would have been ME, now get diagnosed with Long COVID unless they are sure the trigger wasn't COVID. Do you see that as a problem?

 

[Utsikt]

Understand the concerns, important to consider.

What’s currently being done to address them?

False positives happen with just about every diagnostic tool. I believe there are certain standards to adhere to. Even the current diagnosis path has false positives. As an example, early MS can be misdiagnosed as ME/CFS, a good clinician will still look for follow up tests.

As in trusting the AI tests because it’s all about the outcome?

And you must be unaware of the realities of living with ME/CFS if you seriously believe that people with ME/CFS get appropriate checkups by their doctors. An ME/CFS diagnosis is a medical black hole. Everything will be attributed to it, or they’ll think you just believe you’re sick and stop listening.

The question for us is if the tool generates a net positive in patient care/outcome.

Exactly.

Typically specialists will still look for other explanations. Given that there is no treatment for ME/CFS beyond management, it would be far easier for a clinician to identify another reason for the patient experience.

What do you mean by «easier»? What’s easier?

I suspect a lot of patients, that would have been ME, now get diagnosed with Long COVID unless they are sure the trigger wasn't COVID. Do you see that as a problem?

LC isn’t really a diagnosis. I haven’t seen much to indicate that the people that understand that post-covid issues can be biological are unaware of ME/CFS or PEM.

 

[MelbME]

What’s currently being done to address them?

As in trusting the AI tests because it’s all about the outcome?

And you must be unaware of the realities of living with ME/CFS if you seriously believe that people with ME/CFS get appropriate checkups by their doctors. An ME/CFS diagnosis is a medical black hole. Everything will be attributed to it, or they’ll think you just believe you’re sick and stop listening.

Exactly.

What do you mean by «easier»? What’s easier?

LC isn’t really a diagnosis. I haven’t seen much to indicate that the people that understand that post-covid issues can be biological are unaware of ME/CFS or PEM.

Currently working on development. Will need to be validated to assess accuracy over multiple cohorts. The accuracy will dictate the value of this algorithm, it may not be useful clinically.

As in standards for accuracy of test.

What did I say that makes you think I believe all ME/CFS patients get appropriate care from clinicians? Sorry, that was not my intention at all. The poor care of ME/CFS is why we are attempting improve diagnosis and GP knowledge. Difficulty of diagnosis is a problem for clinicians but simplifying current criteria creates more false positives.

Easier for clinician to identify treatments for a patient if that patient has a diagnosis with known treatments.

What do you mean LC isn't really a diagnosis?

 

[Sean]

Given that there is no treatment for ME/CFS beyond management,

Barely even that. And we have to fight to get even that and dodge the tsunami of mismanagement relentlessly forcing itself upon our lives.

 

[Utsikt]

Currently working on development. Will need to be validated to assess accuracy over multiple cohorts. The accuracy will dictate the value of this algorithm, it may not be useful clinically.

As in standards for accuracy of test.

I’m not sure I understand what the output will be. The AI model will provide a percentage value (as all AI models do)- what does that represent the probability of?

What did I say that makes you think I believe all ME/CFS patients get appropriate care from clinicians? Sorry, that was not my intention at all.

That we have to expect certain standards. It seems like I misunderstood.

The poor care of ME/CFS is why we are attempting improve diagnosis and GP knowledge.

I still don’t understand why you need a test for that.

You can’t get more accurate than applying the criteria, so accuracy can’t be improved.

Looking for alternative explanations to avoid false positives of ME/CFS is a matter of the attitude and competence of the clinician, and can’t be made easier with a test, because you’d have to do the other tests to assess the result of the first one anyways. There is no time saved here.

If you think there is: can you please outline the diagnostic process with and without the test and explain which specific activities that are different?

Improving the knowledge of GPs about what ME/CFS is and how to deal with it can’t be done through developing a test. It can only be done by spreading the knowledge.

Difficulty of diagnosis is a problem for clinicians but simplifying current criteria creates more false positives.

I don’t think anyone have suggested to simplify the criteria.

Easier for clinician to identify treatments for a patient if that patient has a diagnosis with known treatments.

Why does that make it easier for them to identify another explanation for the patients’s experience?

What do you mean LC isn't really a diagnosis?

LC is usually defined as having symptoms ~12 weeks after a covid infection. That tells us nothing of value about what’s wrong with the patient or how to approach it. A patient with a reduced sense of smell would be lumped together with patients with post-ICU complications, very severe ME/CFS, or just slight brainfog. It’s about as useful as the diagnosis «not completely healthy».

 

[Hutan]

Problem with Obesity is that people don't often report they have it, some don't see it as a medical condition but as a temporary impact on their physical appearance. But BMI was used to determine it in UK Biobank, we controlled for BMI in the population.

Obesity is typically defined as a BMI over 30. Yes, if you have BMI, you can derive the percentage of people with obesity. There's no need to rely on biobank participants reporting having it.

Can you tell us what the percentages of obesity were in the C2 and ME/CFS cohorts, which was the comparison figure 1 illustrates? The reported BMI medians are not very useful.


The paper says in the Methods section

Logistics regression was used to estimate the odds ratio for biomarker associations with ME/CFS and comorbid cohorts against the C2 cohort. Odds ratios were adjusted for sex, age, cholesterol-lowering medication and fish oil supplements.

I couldn't see any report of controlling for BMI.

 

[MelbME]

Obesity is typically defined as a BMI over 30. Yes, if you have BMI, you can derive the percentage of people with obesity. There's no need to rely on biobank participants reporting having it.

Can you tell us what the percentages of obesity were in the C2 and ME/CFS cohorts, which was the comparison figure 1 illustrates? The reported BMI medians are not very useful.


The paper says in the Methods section

I couldn't see any report of controlling for BMI.

You're right, "controlled for" was the wrong word. BMI was looked at in the decomposition analysis. I think BMI difference of 0.8 between healthy and ME/CFS had a maximum of a ~10% boost in the odds ratio of lipids. Take that 10% away and they are still all significant. Represents a relatively small effect.

I'd say the deconditioning and multimorbidity are far bigger factors in the ME/CFS for the lipids than BMI.

We use BMI as a continuous variable. The obesity and overweight cutoffs in BMI are really arbitrary whole numbers. Treating BMI as a continuous variable is better.

We are conducting a follow up analysis of the UK Biobank with more data now provided. Any suggestions on comorbidities or considerations are welcome.

Let me know if you want to see any specific comparisons. Based on discussion here it seems like a multi-morbidity control would be good. What about exertion control (self report only)?