Blog: ME/CFS onset had two peaks, which may be a clue to causes

[Simon M]

ME/CFS onset had two peaks, which may be a clue to causes​

New blog. (Formatted paper. Thread)

A new study strengthens the findings that ME/CFS is a disease with a highly unusual feature. Analysis of survey data on patients across Europe found there are two peak ages for
getting ME/CFS, around 16 and the late 30s – a rare bimodal pattern. There are differences between people in the two peaks. Those in the early peak are more likely to report an infectious onset, be more severely ill, and have close relatives with ME/CFS. That combination of age peaks is unique, even amongst those diseases that do have two peaks, and this could be a clue to the biology behind ME/CFS

The biology of ME/CFS is hard to pin down, with few unique features. But a 2014 study of medical records in Norway suggested a striking one: two different ages when people are most likely to get ill. Two onset age-peaks is unusual in disease biology, but it is seen in a few cancers and some autoimmune diseases.

It started with a tweet

This unusual pattern seemed to me back then as though it might help unravel the biology behind ME/CFS, but, over the years, the clue wasn’t pursued. Perhaps no one thought the finding was real: the study was limited to one country, and data were for age at diagnosis – which can be very different from age at onset.

Last year, I decided to see if we could get stronger evidence by focusing on age at onset rather than diagnosis, and by looking in more countries. An initial trawl of data sources found nothing, so I tweeted to ask if anyone had something like this. And struck gold.

Trude Schei, Head of the ME Norwegian ME Association, replied to say that she had led the 2021 European ME Association (EMEA) survey, and had a database of over 10,000 responses with onset age, year, suspected trigger and other invaluable data from countries across Europe. She and her co-lead, Professor Arild Angelsen, were willing to share the data. Suddenly, a study was on.


We formed a team consisting mostly of people who live with ME/CFS: having it, or being close to someone who does. Dr Audrey Ryback at the University of Edinburgh led the study, joined by Charlie Hillier, a scientist with severe ME/CFS, along with Dr Joshua Dibble, also based at the University of Edinburgh, Schei, Angelsen and myself.

Two onset age peaks​

We looked at ten European countries with the largest number of responses to provide us with enough data to get a reliable picture from each country. One third of the 9,380 responses came from Norway, which had the largest number of responses by far, and for Norway two age peaks were visible with the naked eye:

Graphs showing the age people reported their ME/CFS started for Norway (left) and all other countries. (Figure 1 B and C in paper).

For all other countries – both combined (shown above) and individually – the two peaks were less obvious and we turned to statistical techniques to robustly identify whether there were peaks in the data.

First we used a ‘dip’ test, which tells us whether we have evidence for more than one peak in our data. This produced results for Norway and all other countries combined that were highly significant, both having a p value well below the statistical significance threshold of p ≤ 0.05, at p < 0.0000000000000002 (that is, p < 2 x10-16).

Six of the nine other countries were also individually statistically significant, though Switzerland and France (which had the smallest sample sizes), and the Netherlands, were not.

Next we wanted to work out where the peaks were and describe their features. We used a technique called a Gaussian Mixture Model to do this, and the results were striking, as you can see below.

The early peak is marked in orange, the late peak in blue. From Figure 2 in the paper.

read the full blog

 

[Simon M]

Please do post any corrections,

And can anyone advise me on how to include small photos of team members in the original Wordpress blog? When I tried to add them, they appeared huge on phone screens. Thanks. DMs appreciated.

 

[Creekside]

Interesting. It could be due to non-medical effects, such as age-related decisions to go to doctors for a complaint, or doctors ignoring the ME option, but that's for statisticians to judge. I think it's reasonable for a complex disease to have multiple modes or mechanisms of triggering.

I think it's less a clue and more of a means for weeding out some theories. For example, is there a known age-related change in the body that would fit the chronic viral infection theories?

 

[Yann04]

This is definitely interesting. And obviously Simon and co have made a great contribution to the field to bolster the evidence behind this. So thank you for that.

It is remarkable that this seems to appear across countries.

I’m still not entirely convinced we can be certain of the dual age peaks though. It’s a widely accepted fact that most people with ME/CFS haven’t received the diagnosis. Which adds a massive selection bias.

Then there is the issue of the voluntary EMEA survey being a large selection bias as well. I doubt it’s representative. Same thing for who got diagnosed and recruited into DecodeME (since this finding was replicated in DecodeME data).

For this reason I hope there is a replication attempt in a random population diagnosis criteria screening study. I think that could help bolster the evidence. Age of diagnosis or onset is a pretty basic variable. Some already done population diagnosis criteria screening studies probably have a dataset that could be used to test replication.

 

[Kitty]

Great work, @Simon M!

I've never seen Wordpress, let alone used it, but usually images are constrained by inserting them into boxes or placeholders of some kind. They then scale along with screen size. Hopefully someone with practical experience will have something more useful.

 

[Hoopoe]

The link "genetic association related to the immune system has been demonstrated in the early onset peak" is broken.

 

[Simon M]

It is remarkable that this seems to appear across countries.

That was what stunned us, and makes the case more compelling. The data looks very different for Sweden and Norway (the former has a much smaller early peak), yet all the countries have very similar peak onset ages.

I’m still not entirely convinced we can be certain of the dual age peaks though. It’s a widely accepted fact that most people with ME/CFS haven’t received the diagnosis. Which adds a massive selection bias.

Yes, that's likely. But how would that selection bias produce such an unusual pattern, and how can it be so consistent between countries?

Then there is the issue of the voluntary EMEA survey being a large selection bias as well. I doubt it’s representative. Same thing for who got diagnosed and recruited into DecodeME (since this finding was replicated in DecodeME data).

As the paper acknowledges, this is a significant factor. The survey was promoted by ME organizations, so if you don't have a diagnosis, you are very unlikely to have seen this survey.

It's also worth noting that the Norwegian data registry study (Bakken, 2014), gives a very similar pattern (with a strong early peak) even though it covers all diagnosed cases, rather than the much smaller number who saw the survey. That suggests that at least in Norway, the selection of the minority who completed the survey did not introduce substantial skewing bias.

For this reason I hope there is a replication attempt in a random population diagnosis criteria screening study. I think that could help bolster the evidence.

I hope so too.

Age of diagnosis or onset is a pretty basic variable. Some already done population diagnosis criteria screening studies probably have a dataset that could be used to test replication.

Sadly, I don't think it is a variable that is captured, so I would be surprised. I'm pretty sure if the Norwegians did have onset age in their registry, they would have used it. Hopefully, it will be captured in future, and could help make sense of messy data - perhaps early and late onset cases behave differently.

If anyone does know a a population linked study with onset age, please do say. That would be awesome.

 

[Yann04]

That suggests that at least in Norway, the selection of the minority who completed the survey did not introduce substantial skewing bias

Yes. So it seems the source of bias to eliminate to make this finding go from likely to essentially proven would be who gets a diagnosis.

Yes, that's likely. But how would that selection bias produce such an unusual pattern, and how can it be so consistent between countries?

ME/CFS is a niche concept promoted by a few medics in most of the world. Not only the biases these medics have given they often cross communicate, but also the biases of who is recommended that special doctor, who figures out they have ME/CFS online and then searches and finds the “ME/CFS doctor”.

It doesn’t take much of a bias at all to have an outsized effect on the data. Say for example only 1 in 6 people with ME/CFS are properly diagnosed (I think this is likely much smaller in most of the world). It only takes a slight bias in who that 1 of 6 is compared to the general population with ME/CFS for it to have an outsized looking effect on the epidemiological data.

 

[Axel]

It is possible that in France, a misdiagnosis of spasmophilia may have masked the onset of ME/CFS. This diagnosis was handed out like hot cakes and suddenly seemed to disappear, so I suppose it covered a range of conditions.

 

[Simon M]

ME/CFS is a niche concept promoted by a few medics in most of the world. Not only the biases these medics have given they often cross communicate, but also the biases of who is recommended that special doctor, who figures out they have ME/CFS online and then searches and finds the “ME/CFS doctor”.

That's true in some cases, less so in some others, such as the Norway. They diagnose about 1,500 a year from a population of around 5 million (equivalent of about 18,000 a year in the UK). And we know from NHS data (Samms & Ponting) that there are 100k diagnoses in the NHS HES database. That isn't a few specialists deciding - though it was the case in the UK historically. Mostly in the UK diagnoses is from the general clinics set up post 2000, and they are definitely not specialists. Yet the UK and Norway have the largest early peaks. And the same pattern was reported in the UK in the 1990s before diagnosis scaled up from specialist to generalist.

Biases could be a factor, but two specific age peaks is hard to explain - especially when they are so consistent between countries. I'd like to see a more coherent alternative explanation for peaks in mid teens and late 30s to say biases are likely to cause two peaks. (I think it's important to test all theories, including the explanation we put forward).

 

[Amw66]

That's true in some cases, less so in some others, such as the Norway. They diagnose about 1,500 a year from a population of around 5 million (equivalent of about 18,000 a year in the UK). And we know from NHS data (Samms & Ponting) that there are 100k diagnoses in the NHS HES database. That isn't a few specialists deciding - though it was the case in the UK historically. Mostly in the UK diagnoses is from the general clinics set up post 2000, and they are definitely not specialists. Yet the UK and Norway have the largest early peaks. And the same pattern was reported in the UK in the 1990s before diagnosis scaled up from specialist to generalist.

Biases could be a factor, but two specific age peaks is hard to explain - especially when they are so consistent between countries. I'd like to see a more coherent alternative explanation for peaks in mid teens and late 30s to say biases are likely to cause two peaks. (I think it's important to test all theories, including the explanation we put forward).

For the early peaks , depending on where you live , it may not be clinics but paediatricians diagnosing .
Diagnosis will therefore be dependent on on the criteria they work to .
Until recently PEM was not understood well ( arguably at all ) , so general post viral fatigue will be a confounder , which may also underpin the promoted recovery rates for this group.

 

[Simon M]

For the early peaks , depending on where you live , it may not be clinics but paediatricians diagnosing .
Diagnosis will therefore be dependent on on the criteria they work to .
Until recently PEM was not understood well ( arguably at all ) , so general post viral fatigue will be a confounder , which may also underpin the promoted recovery rates for this group.

There will certainly be some post viral fatigue cases in this sample. We also see the same pattern of age peaks in mid teens and late 30s it in DecodeME, which also recruited via social media but had a decent question to screen for PEM in addition to requiring a diagnosis. And the same pattern has also been reported by specialist clinicians. So the finding appears to hold up aside without simple post viral fatigue.

There are biases in this EMEA sample that will have some effects, biases in all the age peak studies, Usually in ME/CFS research we are looking at a single or a few parameter comparing controls with patients and it's easy to see how biases can confound such differences. I find it harder to see how biases cause this very specific bimodal pattern (thought they might well modify it). Perhaps this is just my biases talking.

 

[Evergreen]

Great to read another of your clear blogs, @Simon M. They cut through the fog.

@Yann04's questions and your replies also very helpful.

Spotted a rogue "n":

And only those with n a diagnosis were likely to find the survey, which was promoted by national ME organisations.

I wanted to ask you +/- @Jonathan Edwards about arthritis. My understanding is that arthritis in children is usually considered different from arthritis in adults. If we take osteoarthritis out of the picture and stick to JIA vs rheumatoid/spondylo/psoriatic in adults (not sure what to do about Still's), are there age peaks for arthritis?

I found this interesting in terms of thinking about how prognosis of early onset vs the rest might compare. Here are some quotes:

“[Juvenile Idiopathic Arthritis] represents a group of very different diseases that evolve differently in adulthood,” she says.

To prove the point, Dr. Oliveira-Ramos and her colleagues analyzed data from a Portuguese rheumatic disease registry. Most patients in it had been diagnosed with JIA around age 10 and had arthritis for 22 years. The researchers found that in adulthood:

• Almost all patients with systemic JIA were classified as having Still’s disease (the adult version of systemic JIA).
• More than 90 percent of patients with childhood RF-positive polyarthritis and more than half with RF-negative polyarthritis were diagnosed with rheumatoid arthritis (RA).
• Some JIA patients with oligoarthritis were diagnosed with RA, and others with spondyloarthritis (SpA) – a type of arthritis that attacks the spine, especially the low back.
• Almost all kids with enthesitis-related JIA were reclassified as having SpA.
• Patients diagnosed with childhood psoriatic arthritis (PsA) had the same diagnosis as adults.
• Twenty percent of kids diagnosed with RF-negative polyarthritis or oligoarthritis never received an adult diagnosis.

Dr. Oliveira-Ramos found that about 72 percent of patients in her study still needed some type of disease-modifying drug to control their symptoms. But other studies have more optimistic findings. For instance, Norwegian researchers report that 30 years after being diagnosed with JIA, 60 percent of people were in complete remission off medication. A small number were in complete remission with arthritis drugs. The rest had active disease. Most had the same type of arthritis at 30 years that they had 15 years after diagnosis.

Greek researchers also looked at long-term outcomes in JIA patients. They found nearly half were in complete remission 17 years after they were first diagnosed.

These studies also looked at why some former JIA patients did better in adulthood than others. They found that kids who were diagnosed early in life, had polyarthritis or were treated with corticosteroids had worse outcomes than others. Research has shown that regular exercise for kids and adults with arthritis can help overcome some of these disadvantages.

 

[Alinda]

Regarding bias, do we know if things like medically unexplained symptoms, somatic symptom disorder etc. Also have a similar peak of onset?
Might not help a lot because I expect a lot of people with me/cfs to be included in those groups, but I think they are the most prone to be impacted by bias.

 

[Mij]

FWIW. Not a study, the ME doctor I saw in Canada in the early 90s with over 20 years experience seeing pwME told me that the average age for women (he didn't mention males) was 25-30 years of age. Most (90% patients) of his patients with post-infectious onset were referred to specialists and dx with something else.

The relapsing/remitting group didn't recover. I was 29.

 

[AliceLily]

FWIW. Not a study, the ME doctor I saw in Canada in the early 90s with over 20 years experience seeing pwME told me that the average age for women (he didn't mention males) was 25-30 years of age. Most (90% patients) of his patients with post-infectious onset were referred to specialists and dx with something else.

The relapsing/remitting group didn't recover. I was 29.

I was 30 years old at onset and was diagnosed at 37 years old.

 

[Yann04]

I'd like to see a more coherent alternative explanation for peaks in mid teens and late 30s to say biases are likely to cause two peaks

In my view as of now we neither have enough evidence to say it’s biases or it’s conclusively proven.

The first peak coinciding with mono is definitely an interesting mechanistic proposition, which in my opinion gives extra weight to the bimodal finding. But the second peak seems currently without plausible biological explanation. (Given it holds in men, so probably not menopausal).

Since mechanistically either way seems at least partially unexplained, I don’t think the biases explanation needs one to stay a possible interpretation until we can correct for diagnostic bias over a random population sample dataset and prove (or cast doubt on) the bimodal age peaks.

But it could be that alternative diagnoses are more popular in late-20s early-30s, burnout depression, migraine, ibs, Functional Neurological Disorder, anxiety, Somatic Psychobabble disorder, Peristent psychobbable syndrome etc.

In the sense that even small biases in how many pwME receive those diagnoses instead of ME/CFS at that age of onset could create an artefact in the data.

 

[Yann04]

As an alternative question. If I understand correctly you propose that it could be possible that ME/CFS is two different illnesses correlating with the different age of onset peaks. I find this a fascinating idea.

I wonder, do you think it would be worthwhile to subgroup and run analyses on biological data by age of onset? Is that possible with the decodeME dataset? (Or did I miss this was already done?), I think in the paper you already ran the age of onset (bar the genes) on the decode ME dataset?

Because finding that the effects increase/decrease and show different hits for different age of onset groups in the decodeME SNPs would be a very big piece of evidence towards bimodal age peaks and also make it likely that they are quite biologically relevant.

 

[ScoutB]

Is that possible with the decodeME dataset?

I had the same question, and people shared lots of interesting thoughts in the DecodeME thread if you haven't seen it.

 

[PageofME]

This is interesting and congratulations for your findings. However, I don't think that this will lead to understand the pathomechanism. Because you are looking at the triggers and it can safely be said they are independent of the pathomechanism because they are so manyfold and the illness looks always the same in its main charateristics.

Explanations might not be too complicated: What if the first statistical concentration at around nineteen represented EBV infection at its peak too, also known as the kissing disease. I'm not sure whether that's a concentration point for EBV infection. But I think I remember from when I had it that late teenagers usually have a more severe course of the illness as when infection happens earlier.

The second concentration could be stress related. I think that in the late 30ies it all can begin to add up for many people.

The covid-trigger could then all even this out.