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

[forestglip]

The ME/CFS group has the highest median BMI of all the groups

Small correction that one group, hypertension, had a higher median BMI.

 

[Hutan]

Thanks @forestglip. Time I had a cup of tea and a lie down.

 

[Trish]

No I know. Just saying you get a lot of conflicting information on this topic. Which comorbidities would you allow?

We selected most common comorbidities in the UK Biobank cohort of ME/CFS.

Gosh this thread has been busy. I haven't caught up with it all yet.

I don't know which comorbidities could be considered part of ME/CFS, but if you look at lists of common symptoms of ME/CFS, GI symptoms, ranging from IBS type to inability to eat are clearly there.

I think something can only be considered part of the ME/CFS syndrome if it started at the same time as the ME/CFS and the severity/occurrence fluctuates in severity in some way in parallel with the core ME/CFS symptoms. So for example if someone's PEM includes IBS flare ups or migraine headaches, that might indicate it's part of their ME/CFS, or at least closely linked with it. On the other hand if someone has had asthma since childhood, and their ME/CFS starts after an infection 20 years later, the asthma isn't part of their ME/CFS, it's a coincidental comorbidity.

I have 4 of your list of common comorbidities. They all started decades before my ME/CFS, and generally don't seem to be directly associated with my ME/CFS, not being worsened by exertion.

Maybe there's some genetic predisposition overlap for conditions related to sensitivities and allergies like hay fever, asthma, IBS and migraine that also predispose to ME/CFS. I don't think DecodeME indicated any overlap with such conditions though.

 

[Braganca]

@Hutan Could you write a letter or response for publication to the journal outlining your scientific critique? Maybe another forum member could look at the analysis and if they agree, co-author with you?

I appreciate that you have done so much work on this and do not like that you are being labeled as “attacking” — you have been polite as far as I can see, your points make sense as far as i can understand it. The history of labeling ME/CFS patients as aggressive is obviously a huge problem.

 

[Evergreen]

Finally got to read the paper. I cannot pretend to follow everything, but I found it fascinating and want to thank Huang, @MelbME and the other researchers for all their work.

For what it is worth, I very much appreciate the clarity with which @Hutan has explained her questions e.g. in this post. However, the more I read the paper and @MelbME's responses, the less I think @Hutan’s point is problematic. My grasp on all of this is so tentative, however, that I am unable to explain why beyond what @MelbME has already said. The sensitivity analysis and supplementary figure 7 are reassuring.

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:

Another recent study did analyse the circulating metabolome in relation to clinical outcomes and found that accumulated lipids, particularly the triglyceride-to-hosphatidylcholines ratio or total fatty acid were predictive of ME/CFS with biomarker potential (18). This agrees with the report of elevated circulating triglycerides (4) and indicates that lipid handling abnormalities may contribute to the clinical syndrome.

A very recent study by members of our team found that accumulated lipids, particularly the triglyceride-to-phosphatidylcholine ratio or total fatty acid were predictive of ME/CFS which lends confidence to our interpretation (18).

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

Edit: Colon followed by p copied over as little emojis with their tongues out! I have fixed that.

 

[Hutan]

Maybe there's some genetic predisposition overlap for conditions related to sensitivities and allergies like hay fever, asthma, IBS and migraine that also predispose to ME/CFS. I don't think DecodeME indicated any overlap with such conditions though.

There was a study that looked at correlations between ME/CFS and all sorts of health conditions in the UK Biobank data. There is a thread on it somewhere here. I remember poking around in the data, but I think I felt in the end that the ME/CFS labelling was probably too noisy to tell us much. I don't think many correlations that we might have expected, things like allergies and asthma, showed up,

 

[Concentric_Spiral]

Appreciate the intellectual energy and passion to truly understand ME/CFS from everyone here.

I think something can only be considered part of the ME/CFS syndrome if it started at the same time as the ME/CFS and the severity/occurrence fluctuates in severity in some way in parallel with the core ME/CFS symptoms.

That makes intuitive sense to me (layman). I don't know if my experience is typical or not, but I started acquiring a lot of food sensitivities as well as a lot of general gastritis around the same time as ME/CFS. So it seems like the gut issues were linked to the ME/CFS symptoms, since I had neither of them until I had both. But obviously it could be coincidental.

 

[Hutan]

I have just looked at the reviewers' comments.

Reviewer 1 gets the uncertainty with the diagnoses making the data unreliable but misses the issue with the inappropriateness of the comparisons

Reviewer 1 excerpt

Given this goal, the results as presented are only suggestive rather than compelling, and there are a number of challenges that the authors have not fully addressed in the design of the study:

1. The UK Biobank population is self-reported from patients' memory of a previous condition, often at a distance of many years. The ME/CFS patients are coming from a pain questionnaire and a verbal interview cohort, and are believed to have high rates of misdiagnosis. Given the challenge of accurate diagnosis, the risk of overfitting models to a single dataset, and lack of understanding of metabolomic variability over an extended period, replication in a totally disjoint ME/CFS cohort really should be performed.

Reviewer 2's comments aren't completely clear, but I think they are a true 'Reviewer 2'. I think they agree with my concerns, although I'm not totally sure.
Reviewer 2 - whole review

Are they novel and will they be of interest to others in the community and the wider field?

The use of biological databanks is an important resource to learn more about ME/CFS. Any study in this field is to be welcomed if it has been properly structured in the statistical field.

Is the work convincing, and if not, what further evidence would be required to strengthen the conclusions?

Considering the difficulties of the etiology of the disease, an appropriate initial analysis is complex. It would be appreciated to introduce the initial hypotheses and the process that has led to perform the statistics in this way. It remains to be concluded whether this is the best method. There are important gaps in the conclusions, actually ME/CFS patients have comorbid diseases and some results should be compared with patients with such diseases and not with healthy populations. There is an error in the discriminant analysis in this regard.

On a more subjective note, do you feel that the paper will influence thinking in the field?
It should be made clear that we have not compared the analyses, for example, of VLDL with respect to the population with hypercholesterolemia. There is no justification for this data to infer the disease.

Reviewer 2 doesn't seem very impressed. They say a study on ME/CFS would be welcome 'if it has been properly structured in the statistical field'. They note that there should have been comparisons with patients with comorbid diseases and not with healthy populations, and that there is 'an error in the discriminant analysis in this regard'. So, they seem to be saying that there was an error in the choice of comparators, which is what I am saying.

I'm not sure why the reviewer uses the pronoun 'we' there and I'm not sure what they are saying. But they seem to be raising the same issue that I have raised, that populations that could be expected to have hypercholesterolemia should not have been excluded from the comparison groups, and that the study is not justified in concluding that high VLDL is characteristic of ME/CFS.

The authors' reply to Reviewer 2 misses the point in the same way that my argument does not seem to have been understood

I think we did do what Reviewer #2 was asking so perhaps this clarification will help: ME/CFS patients have co-morbid diseases but not all the same comorbid disease, this means that any cohort of ME/CFS patients has high heterogeneity due to individuals in that cohort having various different co-morbid diseases. The co-morbid diseases may have conflicting impacts on biomarker data that cancel them out. We conducted our analysis in a way that compares ME/CFS to the common comorbidities to highlight what may stand out in ME/CFS alone. An alternate path was taken by NIH recently, they whittled down a group of ~500 ME/CFS to 17 patients that did not have any co-morbidity in an attempt to identify a signature of purely ME/CFS. We’ve taken the path of keeping the large patient volume and instead of just comparing to healthy, we have compared to co-morbid diseases and the general population

We don’t refer to hypercholesterolemia in the manuscript, we aren’t attempting to infer that people have disease from the biomarker data either. We are trying to infer what biomarkers may be representing ME/CFS as distinct from commonly experienced co-morbidities of ME/CFS.

Reviewer 3 is very complementary of nearly everything. They mention the issue with the uncertainty of diagnosis, which is something the authors acknowledge. They have completely missed the issue of the inadequate comparisons. Excerpt:

There is an obvious contribution to the field of knowledge about this serious disease, mainly with regard to the lipoprotein profile difference from controls and in my view a very important contribution about the influence from comorbidities which complicates diagnosis via specific biomarkers up till now. The final application of the algorithm and machine learning in achieving an improved rate of diagnosis emphasize the importance of this work.

 

[Hutan]

The sensitivity analysis and supplementary figure 7 are reassuring.

Why do you find Supplementary Figure 7 reassuring?

To thoroughly investigate the impact of comorbid conditions in ME/CFS requires stratifying the cohort into groups of isolated condition combinations, which can substantially reduce the sample size and the statistical power. For example, there were 211 ME/CFS individuals with a combination of depression and other comorbid conditions, and 24 individuals with depression only. We recognise that the other 265 comorbid conditions not analysed in this study may influence the biomarker associations. Therefore, we created another cohort with 354 ME/CFS individuals with or without hypertension, depression, asthma, IBS, hay fever, hypothyroidism, or migraine and performed association tests (Supplementary Fig. 7) and sensitivity analysis for this subset (Supplementary Data 9).

What does that even mean 'we created another cohort with 354 ME/CFS individuals with or without hypertension, depression, asthma, IBS, hay fever, hypothyroidism or migraine'? It's not clear how that cohort differs from the full cohort where all the individuals presumably also are 'with or without' the 7 conditions. Because 'with or without' is pretty all encompassing.

As far as I can see in Supplementary Figure 1, the cohort with 354 individuals looked a lot like the comparison group which I'm assuming is C2, with a lot of comparisons not being significant, which suggests plenty of variation but does not suggest that the reduced ME/CFS cohort has a major problem with blood lipids.

Like I said I'm not at all clear on what is being compared to what, but I don't think the analysis is proof that the study is not flawed, nor do I think it is evidence that people with ME/CFS have, on average, more problems with blood lipids than the average UK Biobank participant.

 

[Mij]

Maybe there's some genetic predisposition overlap for conditions related to sensitivities and allergies like hay fever, asthma, IBS and migraine that also predispose to ME/CFS. I don't think DecodeME indicated any overlap with such conditions though.

I've had migraines since age 11. I no longer have them since menopause (10 years).

 

[Peter T]

I think something can only be considered part of the ME/CFS syndrome if it started at the same time as the ME/CFS and the severity/occurrence fluctuates in severity in some way in parallel with the core ME/CFS symptoms. So for example if someone's PEM includes IBS flare ups or migraine headaches, that might indicate it's part of their ME/CFS, or at least closely linked with it. On the other hand if someone has had asthma since childhood, and their ME/CFS starts after an infection 20 years later, the asthma isn't part of their ME/CFS, it's a coincidental comorbidity.

If people with ME consistently develop a specific condition or range of conditions after the onset of their ME to a greater extent than the general population then it might be reasonable to expect there is some connection between their ME and that/those condition/s.

For example, though the studies involved are not fantastically reliable, there does seem a higher incidence of people developing food intolerances after the onset of their ME than we see in the general population. Assuming we would not see this in other health conditions, it is likely that a higher risk of developing food intolerances is a feature of ME.

We just need some better data.

 

[Utsikt]

The validation process is: we have clinicians diagnose these patients as mecfs and these others as not having mecfs, the algorithm correctly gets to the same answer with a certain accuracy repeatedly. People then trust the algorithm as a substitute for the clinician diagnosis.

So you’re essentially admitting that it can’t get better than a clinician diagnosis? Them why use it? What’s the added value?

That's outcome driven. It's the same way that we don't understand the biology of how some drugs work but they work in large trials and we use them.

You’ve already said that plenty of times. I’ve said that’s not the whole story for diagnosis. Why are you ignoring all of the things I’ve mentioned about traceability and explainability? You’re acting as if the current medicolegal systems in most western countries don’t exist.

You’re right that we don’t need to know why a medication works if we know that it works. We do need to know why a diagnosis was chosen over the alternatives because we need to be able to check if obvious mistakes were made in the decision process. That can’t be achieved with black box AI solutions.

Edit to clarify: under most current legal systems in the western world, a doctor would have to provide a reason for agreeing or disagreeing with the AI output. That necessitates that they have sufficient data to draw their own conclusion using traditional methods. Which means that you’d still have to do all the tests to rule out the alternatives.

For ME/CFS specifically, ruling out the alternatives is the labour intensive part. So this test won’t add any value by saving on work, tests, resources. I assume that interpreting the normal tests are straightforward, and that you don’t need AI to tell if a person has hyperthyroidism or not.

You also don’t need AI to check if the person fits the ME/CFS criteria, it’s just a list.

So I don’t understand the use case. It seems completely redundant in a clinical practice.

 

[bobbler]

Gosh this thread has been busy. I haven't caught up with it all yet.

I don't know which comorbidities could be considered part of ME/CFS, but if you look at lists of common symptoms of ME/CFS, GI symptoms, ranging from IBS type to inability to eat are clearly there.

I think something can only be considered part of the ME/CFS syndrome if it started at the same time as the ME/CFS and the severity/occurrence fluctuates in severity in some way in parallel with the core ME/CFS symptoms. So for example if someone's PEM includes IBS flare ups or migraine headaches, that might indicate it's part of their ME/CFS, or at least closely linked with it. On the other hand if someone has had asthma since childhood, and their ME/CFS starts after an infection 20 years later, the asthma isn't part of their ME/CFS, it's a coincidental comorbidity.

I have 4 of your list of common comorbidities. They all started decades before my ME/CFS, and generally don't seem to be directly associated with my ME/CFS, not being worsened by exertion.

Maybe there's some genetic predisposition overlap for conditions related to sensitivities and allergies like hay fever, asthma, IBS and migraine that also predispose to ME/CFS. I don't think DecodeME indicated any overlap with such conditions though.

If it obesity or something specific that’s the concern (or just example) regarding ‘possibly explained by’ then would it be interesting to see whether pwme+obesity vs pwme-obesity is a big difference here. My brain can’t work out what we’d be equivalently ‘measuring for’ but I hope the gist comes across. It’s like the mental work trying to envisage what the decodeME gene findings translate to picture-wise and meaning-wise vs blood-testing 100 people for x and saying 10 have low iron and are also tired, as its instead groups and probabilities I get and takes me being above par to hold the picture in my brain accurately and I don’t trust I’m there atm.

That doesn’t mean of course that the two [me/cfs, obesity] don’t interact or there some preceding thing behind both or type but if it doesn’t reduce the power so much it’s interesting maybe as we can assume in all sorts but the obesity+me/cfs could until checked be less than obesity-me/cfs on that even if we picked that as the main commodity at issue?

I don’t know what suggestions might end up being from this paper (including how other medics might interpret and run with talk about cholesterol) but an example anecdote is I’m not overweight and my cholesterol is good because my good is high even tho my bad isn’t low. I tend towards hypotension not hyper.

And I’ve no idea whether within certain parameters if there was a finding that it doesn’t necessarily mean ‘need to sort their cholesterol’ but a clue that could even point to a needed adaptation ?

Eg if everyone with me/cfs who’d been put on statins felt better me/cfs wise then that would say one thing to unpick, but given the prevalence of statins in the last decade I’d have thought we’d have had many anecdotes at least of that, of course if they don’t or feel worse there’s a multitude of can of worms of possibilities of what that could or couldn’t mean vs the chain of what they [statins] are doing isn’t a totally direct thing that only does intended to do for a non-me/cfs body when only looking at a functionally-fixed aim/endpoint like blood pressure or cholesterol numbers ie there’s knock-on effects all over the place possible.

 

[Utsikt]

Does an AI-diagnostic test add any value in our current healthcare system?

For breast cancer, you’re trying to improve the accuracy of the diagnosis beyond what an expert is able to do. We know we can do that in theory, because we’re able to check later on if the prediction was correct, and we know that the experts aren’t 100 % accurate on all aspects yet.

So the added value of AI for breast cancer is improved accuracy.

You might also save some resources long term if AI becomes good enough to e.g. allow for using only one assessor for «easy» cases, compared to the standard two.

Although the use of AI is expected to decrease the long term competence of the assessors, leading to more errors, so it’s not purely positive. This is a common challenge when humans are still in the loop (as they will be required to be for the foreseeable future for legal, ethical and trust reasons) - a driver that drives less will become worse at driving, and therefore worse at being in charge of driving in critical situations.

For ME/CFS, you’re not trying to improve the accuracy, because we can’t check afterwards if the diagnosis was right. Matching, not exceeding, the accuracy of a clinician’s assessment is the target.

One could argue that many clinicians don’t know how to diagnose ME/CFS so this might improve their diagnostic accuracy, but that wouldn’t really happen without education about what ME/CFS is, because they wouldn’t be able to control the AI output without sufficient knowledge. If they already have sufficient knowledge, the test doesn’t add anything in terms of accuracy.

So there is no added value in terms of improved accuracy beyond what a clinician can already achieve.

As explained in the previous comment, there’s also no added value in terms of saving resources, because the clinician would have to do the traditional tests regardless to check the AI’s assessment.

So AI arguably has a net negative value for the use case of ME/CFS: it adds more work (using it and checking it) and cost without adding accuracy.

 

[MelbME]

So you’re essentially admitting that it can’t get better than a clinician diagnosis? Them why use it? What’s the added value?

You’ve already said that plenty of times. I’ve said that’s not the whole story for diagnosis. Why are you ignoring all of the things I’ve mentioned about traceability and explainability? You’re acting as if the current medicolegal systems in most western countries don’t exist.

You’re right that we don’t need to know why a medication works if we know that it works. We do need to know why a diagnosis was chosen over the alternatives because we need to be able to check if obvious mistakes were made in the decision process. That can’t be achieved with black box AI solutions.

Edit to clarify: under most current legal systems in the western world, a doctor would have to provide a reason for agreeing or disagreeing with the AI output. That necessitates that they have sufficient data to draw their own conclusion using traditional methods. Which means that you’d still have to do all the tests to rule out the alternatives.

For ME/CFS specifically, ruling out the alternatives is the labour intensive part. So this test won’t add any value by saving on work, tests, resources. I assume that interpreting the normal tests are straightforward, and that you don’t need AI to tell if a person has hyperthyroidism or not.

You also don’t need AI to check if the person fits the ME/CFS criteria, it’s just a list.

So I don’t understand the use case. It seems completely redundant in a clinical practice.

This paper was a proof of concept, the diagnostic tool we are trying to create would be providing a clinician with a probability of ME against a background of conditions that are similar to ME. The purpose is to speed up the exclusion of disease portion of the diagnosis process and provide more confidence to a clinician providing a diagnosis. That is the time consuming part, we aren't aiming to exclude the CCC list from the process. The symptom matching as you point out is fast.

Does that clear up the questions? I am trying to answer your posts.

No it can't get better than a specialist diagnosis in accuracy, it can speed it up though and it can provide a tool to less ME-experienced GPs that makes their diagnosis more accurate and faster.

 

[EndME]

This paper was a proof of concept, the diagnostic tool we are trying to create would be providing a clinician with a probability of ME against a background of conditions that are similar to ME. The purpose is to speed up the exclusion of disease portion of the diagnosis process and provide more confidence to a clinician providing a diagnosis. That is the time consuming part, we aren't aiming to exclude the CCC list from the process. The symptom matching as you point out is fast.

Does that clear up the questions? I am trying to answer your posts.

No it can't get better than a specialist diagnosis in accuracy, it can speed it up though and it can provide a tool to less ME-experienced GPs that makes their diagnosis more accurate and faster.

I’ve skimmed over @Hutan’s writings and I can see how what she suggests is a problem, especially if the goal is a blackbox diagnostic panel, but admittedly I have only briefly skimmed things so I might have missed the relevant parts of the analysis.

The hardest and most useful part of an ME/CFS diagnosis is ruling out alternative diagnoses. If this study cannot provide information on that, because the “AI biomarker” may be a result of ME/CFS but it may also simply be driven by something like obesity, then it is particularly unhelpful because it does not rule out alternative explanations like obesity.

Nobody should be given an ME/CFS diagnosis on the basis of having hypertension or xyz. Even if it was supposed to be a holy blackbox, which just as others I see little value of, then you have to be certain that the holy blackbox is not measuring artefacts of something else.

I understand that there would have been problems in achieving statistical power if the cohort would have been too "clean" (although I do find the comments in relation to the intramural study to be rather unhelpful because what is considered to be a comorbidity in this study didn’t seem to be the largest problem in recruitment in the intramural study, and a larger problem would be in fact ensuring that symptoms are due to ME/CFS which I am not so certain of in the cohort of this study - let’s not forget that rigorous medical screening in the intramural study showed that a portion of participants had diseases that had gone undiscovered till then, but I can understand that multimorbidity and especially multiple diagnoses in an illness of large diagnostic delay can cause some significant problems for researchers in this field that has hard to circumvent especially when access to good cohorts/biobanks is extremely limited) but I don’t understand how that means one cannot analyse how much of the results are driven by things such as obesity by doing a relevant inter-group and outer-group comparison and showing the results of this, because at least I can’t see how say Figure 7 does this sufficiently, and the sensitivity analysis does not suggest that comorbidities aren't driving results, but I may be missing relevant analyses that has been done? @MelbME Could you perhaps point to these?

 

[forestglip]

The scientific arguments presented here by Hutan seem clear to me as well. The sensitivity analysis seems to support that these findings are being mostly driven by comorbidities. While there were still some significant findings, there were also hundreds of other comorbidities in the sensitivity analysis that were not excluded in the cases that could explain the remaining 31 differences.

If the idea is that ME/CFS is strongly associated with having many comorbidities, and so any findings relevant to the comorbidities may in fact be relevant to ME/CFS, why can't we say the same for the 7 other condition groups? Why not keep all the individuals that have hypertension and comorbidities? It seems highly plausible that hypertension goes hand in hand with many other conditions, so it would make sense to compare a heterogenous group with ME/CFS to a heterogenous group with hypertension.

I don't know exactly the reason it was necessary to make hypertension homogenous, but in any case, I think it's likely that the forest plot in figure 2 demonstrates nothing more than that comorbidities were included in one group and not in the others.

 

[MelbME]

Not every paper is wrong, not every paper is likely to materially misdirect research funding if it is wrong. Of course I'm not asking every paper to be retracted.

I want your team to be doing brilliant work that finds the answer to ME/CFS, so that I and others with ME/CFS and the people yet to get ME/CFS can be in a world where there is more understanding of our illness. So, I don't want you and others wasting time on research questions that are not well founded. To be honest, I'm well over this, I'm tired. This is not fun and I have many other things I'd rather do. But, if I'm right, then this paper is misleading in a substantial way and will misdirect research.

Chris, it might be most comfortable to assume that I am singling your team out for criticism on some basis other than the quality of the paper and its potential to impact on the success of future research. I assure you, I am not. I think it's a shame if the focus turns from the facts of the discussion to the nature of the person challenging the findings.



The table gives the median BMIs. You don't tell us the percentage of people who are obese in each cohort. The ME/CFS group has the highest median BMI of all the groups, with the median well into the overweight range. It would be useful to know the percentages of people with obesity in each of the groups in this study; I think this is a key piece of information that should have been presented given its relevance to blood lipids. Obesity is a health condition, and so I assume the homogeneous cohorts (i.e. cohorts where people could only have one health condition) excluded people who are obese.

It would be good to know either way, if people who are obese were allowed to be part of the disease and C2 cohorts. Chris?


The paper says that odds ratios were adjusted for cholesterol-lowering medication use. With 15.7% of the ME/CFS group using these medicines, but e.g. only 0.5% of the people in the IBS group using the medicines, only 0.6% of the people in the migraine group using the medicines and even in the hypertension group, only 9.7% using the medicines, it seems likely that the adjustment had a material effect on the odds ratios for the biomarkers for each group. Your answer doesn't tell me anything about the effect of the adjustment.

Those cholesterol-raising medication use rates in the disease groups are surely not indicative of the rates of use in the UK Biobank populations who have those diagnoses. The use of these medicines in the ME/CFS group is about the same as the rate of use by the whole Biobank population.


The key question surely is 'are the blood lipids of people with ME/CFS, on average, different to people of the same age and sex?'. I've gone back to read the abstract of the paper, and found it very interesting that it actually makes no claims about blood lipids being different. Perhaps the authors or the peer reviewers realised that the paper could not actually provide any evidence that blood lipids levels characterise ME/CFS with the comparison groups it used?

Nevertheless, the study did aim to find differences:

And the Discussion claims to have found differences:

Again, we can't know if the identified differences are due to ME/CFS or the fact that the ME/CFS group included comorbidities known to impact on blood lipid profiles while the comparison groups did not. I have not looked in detail, but I think its very possible that the reported differences could be explained by the presence of people with obesity and Type 2 diabetes and other conditions in the ME/CFS group, while those people were actively excluded from the comparison groups.

I think I've set out the problems I see as well as I can and am starting to repeat myself. From the information currently available, I don't think we can rely on the findings of this paper. I think it builds an edifice of clever, detailed computations on the foundation of a flawed study design.

You're misrepresenting or misunderstanding the papers intentions. You've tried to make the paper about lipids and then questioned why they aren't mentioned in abstract. I think that's the confusion, you think we set out to prove lipids are part of the mechanism of me/CFS. That's not right. Maybe this is the source of our disagreement here without us knowing?

Ponting an Co released a preprint before ours that compared ME to general population (our C1). This seems to be the data you are asking for, it was already published. They show the same signature we do with respect to lipids in the blood. https://www.medrxiv.org/content/10.1101/2024.08.26.24312606v1

The purpose of our paper was to explore the beginning of developing a differential diagnostic signature that could separate out ME/CFS from common comorbid conditions. Differential diagnostic signatures are used in cancer subtyping, this concept is novel for a disease like ME/CFS but we want to try see if it's possible. Diagnosis of ME/CFS take 4-5 years on average in Australia and the speed is largely dictated by clinician confidence in excluding other conditions and differentiating ME/CFS from other conditions.

So to do this we used healthy and 7 comorbid control populations to create a reference of what markers were relevant to those conditions. We couldn't do that with me/CFS because really no patients have no comorbidities conditions. We actually highlight in this manuscript or the next that more comorbidities actually predict ME/CFS and I personally think that either the mechanism of ME/CFS is producing comorbid conditions or the misunderstanding and breadth of symptoms of ME/CFS is leading to diagnosis of other conditions. Either way, more diagnosed comorbidities is a feature of ME/CFS against other conditions. So we allowed the full ME/CFS cohort in and did a sensitivity analysis where we stripped away the conditions and the numbers dramatically fell to 300s. The lower number of patients dropped the power but we still had significant lipid markers. The dropped power seemed to be responsible for the significance drop and not the comorbid conditions dropping from 3 to 0.6. the patterns remained the same but it dropped below our conservative threshold.

We were actually encouraged by reviewers of the paper to talk more about lipids, originally the focus was on the differential pipeline we put together. Maybe it's come across distorted because of this? The diagnostic is still a work in progress and we are developing it further to be more applicable to the translational aim.

Also your concern on multimorbidity is highlighted by us in the paper. We mention it limitations of the paper, we highlight that this portion of work is hypothesis generating. This is why we've felt targeted because everything is clear in the paper, nothing you've brought up is new. Your criticism isn't that we did this wrong, it's that you would have liked to see us do ME vs general population. Again, that's already been published.

We are working on a follow up to this paper though, perhaps tell me comparisons you would be interested to see and I can see if it fits.

 

[MelbME]

I don't know any of the history, but the scientific arguments presented here by Hutan seem clear to me as well. The sensitivity analysis seems to support that these findings are being mostly driven by comorbidities. While there were still some significant findings, there were also hundreds of other comorbidities in the sensitivity analysis that were not excluded in the cases that could explain the remaining 31 differences.

If the idea is that ME/CFS is strongly associated with having many comorbidities, and so any findings relevant to the comorbidities may in fact be relevant to ME/CFS, why can't we say the same for the 7 other condition groups? Why not keep all the individuals that have hypertension and comorbidities? It seems highly plausible that hypertension goes hand in hand with many other conditions, so it would make sense to compare a heterogenous group with ME/CFS to a heterogenous group with hypertension.

I don't know exactly the reason it was necessary to make hypertension homogenous, but in any case, I think it's likely that the forest plot in figure 2 demonstrates nothing more than that comorbidities were included in one group and not in the others.

The scientific arguments presented by Hutan are in the paper. We mention them. It's not the arguments it's the narrative or what they say out intention was with the paper and why we did this and that. It's opting out information to look like this is something that we didn't mention. Like I assume they've read the paper but others on here haven't and are just following on with how they represent it. That's the part I found troubling. Or maybe I'm being sensitive, there is a history but perhaps I can't shake that. I hyperfocus on fairness, it's why I care about ME/CFS to begin with, I don't have a personal ME/CFS connection.

We did remove hypertension from the ME/CFS cohort largely in sensitivity analysis. Hypertension was not increase din ME/CFS. This data was published by Ponting already. ME vs general population have same amount of hypertension. Still much more hypertension. We even compare ME/CFs directly to a cohort of 100% hypertension and ME/CFS had worse lipid profiles.

 

[MelbME]

I’ve skimmed over @Hutan’s writings and I can see how what she suggests is a problem, especially if the goal is a blackbox diagnostic panel, but admittedly I have only briefly skimmed things so I might have missed the relevant parts of the analysis.

The hardest and most useful part of an ME/CFS diagnosis is ruling out alternative diagnoses. If this study cannot provide information on that, because the “AI biomarker” may be a result of ME/CFS but it may also simply be driven by something like obesity, then it is particularly unhelpful because it does not rule out alternative explanations like obesity.

Nobody should be given an ME/CFS diagnosis on the basis of having hypertension or xyz. Even if it was supposed to be a holy blackbox, which just as others I see little value of, then you have to be certain that the holy blackbox is not measuring artefacts of something else.

I understand that there would have been problems in achieving statistical power if the cohort would have been too "clean" (although I do find the comments in relation to the intramural study to be rather unhelpful because what is considered to be a comorbidity in this study didn’t seem to be the largest problem in recruitment in the intramural study, and a larger problem would be in fact ensuring that symptoms are due to ME/CFS which I am not so certain of in the cohort of this study - let’s not forget that rigorous medical screening in the intramural study showed that a portion of participants had diseases that had gone undiscovered till then, but I can understand that multimorbidity and especially multiple diagnoses in an illness of large diagnostic delay can cause some significant problems for researchers in this field that has hard to circumvent especially when access to good cohorts/biobanks is extremely limited) but I don’t understand how that means one cannot analyse how much of the results are driven by things such as obesity by doing a relevant inter-group and outer-group comparison and showing the results of this, because at least I can’t see how say Figure 7 does this sufficiently, and the sensitivity analysis does not suggest that comorbidities aren't driving results, but I may be missing relevant analyses that has been done? @MelbME Could you perhaps point to these?

Real world patients have ME/CFS many have comorbidities predating ME/CFS. I believe a patient highlighted that many in our list they already had before ME.

The differential diagnostic needs to be able to identify people that have comorbidities and ME.

We aren't trying to translate this specific algorithm from the paper, we are building something distinct that has IP attached. To translate anything to practice you need IP, we can't protect what is published. It's more a proof of possibility that we present here.

The real tool is being built from pathology data because how early and often people get that information.