Incidence age is bimodal for [ME/CFS], with higher severity burden for early onset disease, 2026, McGrath et al

[Kitty]

It's likely to be substantial contributor, but it doesn't explain the first peak. Srr figure 3E4 for DecodeME onset ages, infectious and nucleosis, other infections and non-infectious also show a an early onset spike.

Yep, it'd be possible to suggest explanations for pretty much any age peak, but it doesn't mean they actually apply.

It's interesting that the peaks are preserved across time despite significant social changes. When I got ME/CFS (mid-70s), the vast majority of the population left school and started work at 16. They had a level of financial independence almost immediately, often lived separately from their parents, and many had their children in their 20s. That timeline would look pretty different now.

 

[ChronicallyOverIt]

Anecdotally Mono really screwed me up from 16-17ish. I don’t think I would have qualified for me/cfs but I was sick for months if not a year pretty much constantly. I think I had strep 5-6x that year and was constantly testing positive for mono IGA. Then fine for 10 years. Looking at these peaks is quite interesting.

My identical twin brother was also hit by mono at the same time but did not have such a severe reaction during 16-17. He does not have ME either in adulthood.

 

[chillier]

Thanks, here's the graph:
View attachment 31313

And the equivalent for the EMEA data in the supplementary (no glandular fever):

 

[EndME]

Spain has drastically fewer glandular fever cases than the UK

On what is the based?

 

[Kitty]

On what is the based?

I don't know, but some countries do have low levels because social and cultural factors mean children are more likely to get infected with EBV very early. Under 5s often have few if any symptoms, and won't develop glandular fever as young adults.

 

[chillier]

On what is the based?

From the text in our paper:

Furthermore, while the incidence of glandular fever/infectious mononucleosis is 5 per
1,000 in the UK (44), it is an order of magnitude smaller (0.4 per 1,000) in Spain
(45), which in our dataset had the smallest early onset peak.

We cite NICE for uk and Merico-Coy et al 2020 for spain. Statista also provides an estimate here but we struggled to source it https://www.statista.com/statistics...ctious-mononucleosis-among-patients-in-spain/.

 

[Nightsong]

Not well enough to look into this fully at the moment but re the Statista statistics: try looking at the Spanish Health Ministry's SIAP/BDCAP data. I found this from 2012 on a quick search:



Link

 

[Nightsong]

Looking around quickly, it seems you can get more up to date case data out of the Ministerio de Sanidad's statistical portal (although it's a very unintuitive interface). You're looking for code A75, which is "Mononucleosis infecciosa". Hope this helps a little.



Link

 

[Utsikt]

Can the downward slope that starts around 40-50 years tell us something?

Does it mean that the risk of getting ME/CFS decreases once you reach that age?

To what extent could the data be confounded by e.g. underdiagnosis due to other health issues that might become more prevalent in the second half of life?

Could it be confounded by lower digital literacy and presence, creating a recruitment bias in the data?

 

[Kitty]

Does it mean that the risk of getting ME/CFS decreases once you reach that age?

And maybe that the risk of being diagnosed with ME/CFS decreases?

My family and friends have found there's a different attitude to ill health in the second half of life. The older you get, the more seriously symptoms are taken. Referrals to specialists to rule in / out clinical suspicions happen sooner.

So it's possible there are fewer misdiagnoses of ME/CFS after middle age.

 

[OrganicChilli]

Getting some publicity. From Nichtgenesen 4. Fachtag LongCovid, PostCovid, PostVac und ME/CFS

 

[Evergreen]

Huge congrats to @Simon M, @chillier and @audrey_ryback on the publication of this important piece, and thank you so much for all your work to make it happen. I love the story of how it came about.

Like @hotblack, I struggle to read the pdf with line numbers, so I will wait to gorge on the final version.

 

[LizWorthey]

very well done and interesting study! What I find particularly interesting is the lack of difference in age of onset peaks by gender. To me that suggests that whatever factor causes ME/CFS to be more common in women, it might not be a factor that is known to fluctuate with age. Which would also correspond with the bimodal peak, since it’s hard to come up with any kind of hormonal/developmental change that spikes at both ~16 and ~36 years old.

I was trying to puzzle this out also and couldn't come up with anything very good... Very interesting study!

 

[LizWorthey]

Getting some publicity. From Nichtgenesen 4. Fachtag LongCovid, PostCovid, PostVac und ME/CFS

And the equivalent for the EMEA data in the supplementary (no glandular fever):
View attachment 31319

So bimodality is present here only in the infectious, vaccine, and other categories? Which would support the two peaks as corresponding to 1) likelihood of infection as a child or 2) as the parent of a child as others suggested? Or for vaccines to 1) vaccination as a child and 2)?? (or maybe there is no later peak there in that case.

 

[paolo]

Interesting! Gammas skew right so the means will be higher than the peaks so that checks out.

In my analysis (I wrote SurveyME) I fit the distribution of age at onset with bimodal gamma and bimodal lognormal densities, then I selected the best fit. I initially excluded bimodal normal densities because gamma densities can approximate normal ones (as an example, see this, bottom of the page), while the contrary is not true. Also, when I looked at the distribution of age at onset for MS, I realized that it was skewed (see figure below, from my 2024 blog post, where I also fit a bimodal distribution for age at first diagnosis in 5809 Norwegian ME patients).



When I saw the paper by McGrath et al. I decided to test a normal bimodal distribution too (you find it in the repository) and I saw that the fit is worse than the one I got for gamma, and better than the one I got for lognormal. Note that I used the Kolmogorov-Smirnon test, which is probably more punitive than the approach used in the paper, and the fits are statistically significant only for males. Also, I only tested the entire sample (size 9,600 after cleaning).

The important point here is that if the gamma bimodal fit is better than the normal, then the proportion of patients who develop the disease at a younger age is greater than those who get the disease later (see table, from SurveyME).



With the normal bimodal fit, the density with the smaller mean is forced to have a small variance by the constraint to be very close to zero for negative ages and to be symmetric. This constraint is not present for gamma and lognormal densities, which are defined only for positive values of the random variable and are skewed.

 

[chillier]

In my analysis (I wrote SurveyME) I fit the distribution of age at onset with bimodal gamma and bimodal lognormal densities, then I selected the best fit. I initially excluded bimodal normal densities because gamma densities can approximate normal ones (as an example, see this, bottom of the page), while the contrary is not true. Also, when I looked at the distribution of age at onset for MS, I realized that it was skewed (see figure below, from my 2024 blog post, where I also fit a bimodal distribution for age at first diagnosis in 5809 Norwegian ME patients).

View attachment 31458

When I saw the paper by McGrath et al. I decided to test a normal bimodal distribution too (you find it in the repository) and I saw that the fit is worse than the one I got for gamma, and better than the one I got for lognormal. Note that I used the Kolmogorov-Smirnon test, which is probably more punitive than the approach used in the paper, and the fits are statistically significant only for males. Also, I only tested the entire sample (size 9,600 after cleaning).

The important point here is that if the gamma bimodal fit is better than the normal, then the proportion of patients who develop the disease at a younger age is greater than those who get the disease later (see table).

View attachment 31457

With the normal bimodal fit, the density with the smaller mean is forced to have a small variance by the constraint to be very close to zero for negative ages and to be symmetric. This constraint is not present for gamma and lognormal densities, which are defined only for positive values of the random variable and are skewed. This also makes them, in my opinion, a better choice in this case (age is always positive).

Very interesting that all makes sense to me. You are probably right that gammas are a better fit - and if you look at the glandular fever only plot from the decodeME data it looks quite like a gamma shape for that subset. Yeah using gammas would mean more people who in our paper are weighted towards belonging to the older peak would be weighted towards the younger peak instead, which would change the populations a bit for the severity comparisons but I strongly doubt would change the conclusions.

My instinct is not to be that bothered by the KS tests being insignificant (for most of the gammas and presumably for everything else too) because I think the goal (in our paper at least) is not to find the set of distributions that perfectly fit the variance in the data, but to assess evidence for bimodality and to estimate of what those modes are. There's lots of sources of additional variance both technical (ascertainment bias) and biological that could change the shape or add more variance to the peaks than can be explained by gammas alone (maybe a compound distribution like a gamma-poisson would fit even better? - they allow for overdispersion).

I'm surprised that normals fit better than log-normals though, if normals fit worse than gammas. I thought that log-normals and gammas were very similar in shape.

 

[paolo]

Very interesting that all makes sense to me. You are probably right that gammas are a better fit - and if you look at the the glandular fever only plot from the decodeME data it looks quite like a gamma shape for that subset. Yeah using gammas would mean more people who in our paper are weighted towards belong to the older peak would be weighted towards the younger peak instead, which would change the populations a bit for the severity comparisons but I strongly doubt would change the conclusions.

My instinct is not to be that bothered by the KS results being insignificant (for most of the gammas and presumably for everything else too) because I think the goal (in our paper at least) is not to find the set of distributions that perfectly fit the variance in the data, but to assess evidence for bimodality and to estimate of what those modes are. There's lots of sources of additional variance both technical (ascertainment bias) and biological that could change the shape or add more variance to the peaks than can be explained by gammas alone (maybe a compound distribution like a gamma-poisson would fit even better? - they allow for overdispersion).

I'm surprised that normals fit better than log-normals though, if normals fit worse than gammas. I thought that log-normals and gammas were very similar in shape.

When I compared males who got the disease earlier vs those who got the disease later in life (using my gamma bimodal fit to define the two populations), I found that the former are generally more severe: higher severity, sensitivity, dizziness, and sleep problems. But less pain. See table below.

This corresponds to table 4 of my repository.