Preprint Replicated blood-based biomarkers for Myalgic Encephalomyelitis not explicable by inactivity, 2024, Beentjes, Ponting et al

[bobbler]

Not on form today so I am only adding perspective from a very layperson trying to access it when not feeling well point of view. Trying to start from the very basics so I don't say something really stupid because I got the wrong end of the stick.

My first question: Decode ME looked at genes rather than actual blood 'findings'. Is this where it is talking about things like cholesterol etc talking about the same thing as you'd get in your standard blood test(s)?

So when it mentions biomarker then it is more likely a combination of findings/pattern that was/is distinct? (say compared to genes where it might be 'finding on dodgy gene')

And when it is looking at for example a number of things flagging up as being significantly different between cases and controls in a large number of these (but not high enough to be x other disease on their own - eg 'elevated' but not necessarily 'out of reference range') they provide clues as to what might be affected in the bodies of pwme?

And these are all just snapshot measures, so not necessarily in or out of PEM or anything specific etc. Yet consistent enough across cases to say in pwme this 'level' of this particular thing 'gets affected'?

And if there are obvious combinations here of things that tend to operate in a system then that might indicate them all going in the same direction could be connected to each other?


And somehow they've been able to make sure that where they've found these differences, they've also employed some method to answer the question that always begs of 'is it all just due to being sedentary'. How has that been done? Is it because there is already similar past research showing the similar 'patterns/biomarkers' for 'sedentary' and they've just flagged out or excluded these, or because they don't exist? Or is there another method used to rule that out?

 

[Simon M]

I would just say here 'More females than males are diagnosed with ME/CFS'. I think the quoted ratio is too high. Quality studies tend to find that the sex ratio isn't so marked, maybe 70% female to 30% male, sometimes even 60:40. I also don't think the idea that females report more severe symptoms is well evidenced..

The severity data comes from DecodeME.

(Sex ratio has been discussed elsewhere. I would be very interested to see any good papers showing a significantly lower sex ratio - the best would be from community studies, and they are based on tiny numbers of cases and the error margins on M and F prevalence are wide. But if there is new evidence, I'm keen to see it.)

I don't think the study can discount a 'psychosomatic' account because psychosomatic can imply that thinking modifies peripheral body functions in such a way as to cause symptoms and that may be biochemically mediated.

Agreed.

I also don't think depression is a psychological illness (or psychosomatic). It's a biological disease that principally affects mood. But it also affects the cardiovascular and immune system

I think that depends. Depression can be a human response to terrible life circumstances (such as chronic illness, though in that case there would be confounding of biological chances.) and I'm not sure there is evidence that kind of depression is caused by biological factors (sorry if the evidenceis in your reference list - I don't have the energy to check).

 

[chillier]

This looks like a really exciting paper I'm looking forward to properly reading through it.

For blood traits, these are the ones that had a Cohen's D > 0.2 for both males and females:

View attachment 22908

Yes you're right these aren't that small. I had misread and assumed that most statistically 'significant' findings would have very small effect sizes < 0.2 SD which doesn't seem the case. Apologies for the confusion (will EDIT my first post above).

Here is the same table you showed but for the NMR results rather than blood traits (only showing some of the top ones, there's many more with Cohen's D > 0.2). What's really nice is that is corroborates the lipids/fats results from the blood biochemistry (as measured enzymatically) by measuring it in an orthogonal way (NMR). This to me helps allay concerns about the results being something technical or not clinically relevant.




What's also cool is that findings relating to high levels of triglycerides, unsaturated fatty acids and low levels of phosphatidylcholines are what are reported lipkin and fiehn's metabolomics study - and in general these findings are probably the most well replicated of the metabolomic data done at baseline (also seen in Naviaux and I think possibly Tronstadt and Hanson's work).

 

[chillier]

I'm not totally sure how you're supposed to interpret the results from these direct effect plots:



These are reported as 'effect sizes' but I don't think they are comparable to an effect size like Cohen's D which indicates the magnitude of the difference after taking into account the variance in the data - otherwise we've somehow gone from an effect size of ~0.4 in the supplementary data to an colossal effect size of ~7 in their model which can't be right.

I think these are probably unscaled values indicating that gamma glutamyl transferase levels are 7 units higher in the patients than controls on average (After taking account of the exercise mediator). Correct me if I'm wrong.

 

[Hutan]

It is a female-dominant disease, with females outnum- bering males by up to five-to-one; females also report more severe symptoms [13, 14].

(Sex ratio has been discussed elsewhere. I would be very interested to see any good papers showing a significantly lower sex ratio - the best would be from community studies, and they are based on tiny numbers of cases and the error margins on M and F prevalence are wide. But if there is new evidence, I'm keen to see it.)

Five to one is a very high ratio. Yes, it says 'up to five-to-one', which covers a lot of ground, but still. I don't think it gives an accurate representation of what we know.

You yourself (Simon) said this on the epidemiology thread, suggesting a ratio of 3 to 1 or 4 to 1.

The Nacul 2011 Prevalence study found around 80% female cases for ME/CFS (Table 4 https://bmcmedicine.biomedcentral.com/articles/10.1186/1741-7015-9-91). The huge Norwegian study gives an incidence ratio of 3.2 female cases for every male case (76% female). And just about every mecfs study ever done, covering many thousands of cases, have 75%-80% female cases (I always check).

The first meta analysis I found when searching the forum was this one of 8 studies in Korea and Japan, admittedly Fukuda criteria. It found a ratio of 2.2 to 1 in Korea and 1.2 to 1 in Japan.

Results: Of the eight studies (five in Korea, three in Japan) included, the total prevalence rate of Korean studies was 0.77% (95% CI 0.34–1.76), and 0.76% (95% CI 0.46–1.25) for the Japanese studies. The prevalence rate in females was approximately two-fold higher than males in Korean studies (1.31% female vs. 0.60% male), while the gender difference was less obvious in Japanese studies (0.76% female vs. 0.65% male).

My sense is that it would be more representative of the data to note that epidemiology is poor but suggests a ratio of three to one (75%:25%) - the Norwegian study would do for reference. And note that that is for diagnoses. It is often said that most people with ME/CFS aren't diagnosed, and so there's a lot we don't know. (There's also the issue of some diseases that seem to trigger ME/CFS tending to infect more men e.g. Q fever, Ross river fever. Indications are the gender ratios of the resulting ME/CFS are more even for those diseases.)

 

[Andy]

My first question: Decode ME looked at genes rather than actual blood 'findings'. Is this where it is talking about things like cholesterol etc talking about the same thing as you'd get in your standard blood test(s)?

So when it mentions biomarker then it is more likely a combination of findings/pattern that was/is distinct? (say compared to genes where it might be 'finding on dodgy gene')

This is not a DecodeME paper. This team have looked at the blood test results from participants in the UK Biobank.

 

[Chris Ponting]

Five to one is a very high ratio. Yes, it says 'up to five-to-one', which covers a lot of ground, but still. I don't think it gives an accurate representation of what we know.

You yourself (Simon) said this on the epidemiology thread, suggesting a ratio of 3 to 1 or 4 to 1.


The first meta analysis I found when searching the forum was this one of 8 studies in Korea and Japan, admittedly Fukuda criteria. It found a ratio of 2.2 to 1 in Korea and 1.2 to 1 in Japan.


My sense is that it would be more representative of the data to note that epidemiology is poor but suggest a ratio of three to one (75%:25%) - the Norwegian study would do for reference. And note that that is for diagnoses. It is often said that most people with ME/CFS aren't diagnosed, and so there's a lot we don't know. (There's also the issue of some diseases that seem to trigger ME/CFS tending to infect more men e.g. Q fever, Ross river fever. Indications are the gender ratios of the resulting ME/CFS are more even for those diseases.)

In DecodeME we see a female/male ratio of 5-to-1 (https://openresearch.nihr.ac.uk/articles/3-20), and across all ICD10 G93.3 codes in England we see a ratio of 4-to-1 (https://www.medrxiv.org/content/10.1101/2024.01.31.24302070v1). As these are for UK populations and count >17,000 cases, I do think personally that these are the most relevant and reliable estimates.

 

[chillier]

@Chris Ponting

Thank you for your answer!



In this extreme example for Chylomicron concentration from fig4A, is it therefore correct that the ME group has a Z score ~ 5e-5 higher than the control group? How does this result end up as significant which such an extremely small z score difference?

Apologies I realize I might be missing something obvious.

 

[Hutan]

In DecodeME we see a female/male ratio of 5-to-1 (https://openresearch.nihr.ac.uk/articles/3-20),

But we can't assume that the voluntary response to a mail-in study is the same as population prevalences. See this study for example. Men might be less likely to be diagnosed with ME/CFS, less likely to network with others with the disease and so hear about the study, and less likely to participate in a study.
DecodeMe was a sample of 20,000 or so people out of a population of maybe 250,000? There's plenty of room for bias in the DecodeME cohort.

Here's another paper, a large study (50 million), that still doesn't account for a likely bias in diagnosis, that found a similar gender ratio to that of the Korean and Japanese meta-analysis.
Estimating Prevalence, Demographics, and Costs of ME/CFS Using Large Scale Medical Claims Data and Machine Learning, 2019
Discussed on the forum here

Techniques of data mining and machine learning were applied to a large database of medical and facility claims from commercially insured patients to determine the prevalence, gender demographics, and costs for individuals with provider-assigned diagnosis codes for myalgic encephalomyelitis (ME) or chronic fatigue syndrome (CFS). The frequency of diagnosis was 519–1,038/100,000 with the relative risk of females being diagnosed with ME or CFS compared to males 1.238 and 1.178, respectively. While the percentage of women diagnosed with ME/CFS is higher than the percentage of men, ME/CFS is not a “women's disease.” Thirty-five to forty percent of diagnosed patients are men.

 

[chillier]

I agree with a number of comments about the treatment of causation.
More generally, I think the isolation of inactivity is an interesting approach but rather arbitrary, as others have said. Diet, drugs, social contact with people with viruses and all sorts may come in.

I'm not sure I agree with that. I would expect that inactivity and ME/CFS status are particularly confounded more so than the other factors you mention. Also given Wuest's recent study where he argued an insulin resistance phenotype in healthy individuals assigned to bed rest, I think it's important to make sure any insulin resistance phenotype seen here is not the result of a similar process.

Deconditioning is also an argument frequently made to try and explain away symptoms, so I think from a social perspective it's good to specifically address what is and isn't activity related if it's possible.

I agree with a number of comments about the treatment of causation.
Perhaps the thing that worries me most is that the analysis seems to imply that there is some homogeneous 'ME' process throughout the cohort. There may be but there may not. The 'ME' cohort may include people with two completely different processes generating similar reports of symptoms. Being devil's advocate, one of those processes might be entirely 'psychosomatic'.

Interesting. I don't know if this is addressed in the paper - but are these biomarkers all seen together in the same individual, or do the traits present together in different patient clusters? Eg one cluster with high triglyceride and and a separate cluster with elevated liver markers, and maybe a big cluster with no trait differences at all?

 

[Simon M]

You yourself (Simon) said this on the epidemiology thread, suggesting a ratio of 3 to 1 or 4 to 1.

Bakken: this is the lowest of studies I trust. DecodeME is likely better as bigger and broader recruitment (doesn't need GP referral) at 83% female (Mail in rate sample return was VERY high, reducing bias risk.)

Japan/S Korea prevalence. As before, prevalence studies tend to have small samples, wide confidence intervals, so are unreliable for sex ratios. And most of the studies I've seen from these countries (not necessarily these ones) are of questionable quality.
Nacul: not ideal either: no doctor ME diagnosis for many cases (questionnaire only of cases selected by medical codes with a GP reviewing records only).

ere's another paper, a large study (50 million), that still doesn't account for a likely bias in diagnosis, that found a similar gender ratio to that of the Korean and Japanese meta-analysis.
Estimating Prevalence, Demographics, and Costs of ME/CFS Using Large Scale Medical Claims Data and Machine Learning, 2019

Insurance claim databases have a more than iffy record on diagnosis (including having the highest rates of incidence in those aged 65+, suggesting it is can be used as a bucket code).

It is possible the true ratio is as 3:1, still pretty high, but I think it could well be 5:1 (I think DecodeMe is 6:1). However, we are revisiting old ground and I will leave it there.

 

[Chris Ponting]

@Chris Ponting

Thank you for your answer!

View attachment 22915

In this extreme example for Chylomicron concentration from fig4A, is it therefore correct that the ME group has a Z score ~ 5e-5 higher than the control group? How does this result end up as significant which such an extremely small z score difference?

Apologies I realize I might be missing something obvious.

Not at all, we were not clear. For each trait (e.g. Chylomicron concentration) we have estimated the Natural Direct Effect (NDE) and it is this that we are plotting here, not the Z-score. And because this trait (Concentration_of_Chylomicrons_and_XL_VLDL) is measured in mmol/litre (https://biobank.ndph.ox.ac.uk/showcase/field.cgi?id=23481) we are showing the x-axis on this scale. We are not plotting the Z-score (here, the NDE divided by its estimation error). Hope this helps.

 

[Chris Ponting]

I'm not totally sure how you're supposed to interpret the results from these direct effect plots:

View attachment 22914

These are reported as 'effect sizes' but I don't think they are comparable to an effect size like Cohen's D which indicates the magnitude of the difference after taking into account the variance in the data - otherwise we've somehow gone from an effect size of ~0.4 in the supplementary data to an colossal effect size of ~7 in their model which can't be right.

I think these are probably unscaled values indicating that gamma glutamyl transferase levels are 7 units higher in the patients than controls on average (After taking account of the exercise mediator). Correct me if I'm wrong.

Yes, these are NDE values in the units measured for gamma glutamyltransferase (U/l) and are not Z-scores. See above. Thank you.

 

[chillier]

Not at all, we were not clear. For each trait (e.g. Chylomicron concentration) we have estimated the Natural Direct Effect (NDE) and it is this that we are plotting here, not the Z-score. And because this trait (Concentration_of_Chylomicrons_and_XL_VLDL) is measured in mmol/litre (https://biobank.ndph.ox.ac.uk/showcase/field.cgi?id=23481) we are showing the x-axis on this scale. We are not plotting the Z-score (here, the NDE divided by its estimation error). Hope this helps.

Fantastic, thank you, that makes sense. : )

In that case I wonder as a suggestion if it might be good to report the units alongside each blood trait/molecule, or to do away with it completely and use z scores as the numbers currently don't mean anything to the reader without units. If not maybe there's some way to make it a bit clearer in the legend.

 

[tuppence]

I agree with a number of comments about the treatment of causation.

I don't think the study can discount a 'psychosomatic' account because psychosomatic can imply that thinking modifies peripheral body functions in such a way as to cause symptoms and that may be biochemically mediated. In fact, tha 'biosychosocial' view simply requires that psychosocial factors modulate the process, which might be by change in diet and so on.

The severity data comes from DecodeME.


I think that depends. Depression can be a human response to terrible life circumstances (such as chronic illness, though in that case there would be confounding of biological chances.) and I'm not sure there is evidence that kind of depression is caused by biological factors (sorry if the evidenceis in your reference list - I don't have the energy to check).

Depression can be identified as a physical/biological response to ‘terrible life circumstances’, made better, not by CBT or other psychological therapy to change your thinking or mental response, but by removing yourself physically from those ‘terrible life circumstances’.

This was my experience and the explanation provided to me by the ‘Expert Psychiatrist’ witness who helped me obtain significant compensation from my Employer who he considered were responsible for causing and maintaining my ‘depression’ damaging my ‘Pyche’ (which he described as a biological physical organ like any other) such that it would not allow me to function at previous high levels of responsibility or work for my employer ever again. He considered, I had justifiably lost all trust in my employer and their medical adviser to safeguard my health, both psychological and physical.

The BPS view of ME was the medical view that my employer was working with in responding to the health difficulties I was experiencing at work and they accepted the ‘Expert Report’ which placed the cause of my depression with the ‘terrible life circumstances’ it gave rise to at work for me with that unsubstantiated BPS medical view of ME directing and perpetuating it all.

Is not the discovery of yet more evidence of different biology in the blood of ME patients and the complete absence of any evidence demonstrating efficacy or applicability in the BPS approach to ME strengthening of the primary importance of taking a biological response to the scientific investigation and treatment of ME ? If so, then the paper should make/explain this in some way.

 

[JemPD]

'Full recovery from ME/CFS after the initial two years is rare'. I'm not sure what the best reference there is for that - the Dubbo study would probably work. The problem with saying that full recovery is rare is that it means that the many people who recover in the first year will think that they are special and have found the miracle cure, when in fact recovery at that time is common.

This is so important

Thank you very much to you and your team @Chris Ponting it means a lot to know people are working so hard and passionately on our behalf!

yes huge gratitude to all of you

 

[Hutan]

So as to not take the thread off topic, I'll address the arguments about the sex ratio used in this paper on the epidemiology discussion thread.
ME/CFS Epidemiology - sex ratios, female predominance

 

[Kiristar]

The data is from the UK Biobank, not the UK ME/CFS biobank.

I hope the samples from the UK ME/CFS biobank will be tested to see if the finding it replicated, as that's a much more carefully diagnosed cohort.

Yes I think that would be a very interesting extension. You would expect the ME specific indicators in that Cure ME cohort, which includes severe patients, who'd definitely be too unwell to attend the UK Biobank process, to be significantly more marked.

 

[EndME]

Yes I think that would be a very interesting extension. You would expect the ME specific indicators in that Cure ME cohort, which includes severe patients, who'd definitely be too unwell to attend the UK Biobank process, to be significantly more marked.

Similarly importantly you'd expect that in severe patients that the things that were used to model inactivity should be more abundant. In which case you'd have established that you did indeed have a good model of inactivity in the first place and should it not be the case it would put the model of inactivity into question. 2 flies with one stone.

 

[DMissa]

There may be but there may not. The 'ME' cohort may include people with two completely different processes generating similar reports of symptoms.

I reckon we are lucky if it's only two.