UK: MRC and NIHR announce ME/CFS workshop, November 2019 & ME/CFS Biomedical Partnership FAQ

[Jonathan Edwards]

I am not worried about some dilution of the cohort with misdiagnoses. If 60% of the 20,000 ME/CFS cohort did not in fact have ME/CFS, maybe not having PEM, I think the chances that it would prevent picking up a genetic link would be small. You would just be diluting the ME/CFS genetic signal a bit.

What worries me more is that getting a cohort through social media-based volunteering could very easily introduce spurious genetic bias.

Paradoxically, there is an argument for saying that what the study really needs to do is to get samples from at least 60% of PWME in the UK - say 70,000. It may be that people below 18 are not eligible and that might make it 50,000. The ideal would be to get 85%+. Having a high proportion of all cases would reduce the risk of recruitment bias.

I think there would be a case for putting out recruiting material indicating that the study wants to study every single person with ME/CFS in the UK above 18. National newspapers and television services could be asked to make a story about it. I suspect not that many more than 20,000 would respond and it might then be possible to back check using the sort of methodology the ME Biobank used to see where biases in recruitment lay and try to offset these in a sensible fashion. (Another approach might be to limit recruitment to a geographical area with only 20M people, but this would probably have more downsides than upsides.) As long as all that is done before a statistical analysis is undertaken it wouldn't fall foul of multiple analysis problems.

 

[ME/CFS Skeptic]

There seems to be the assumption that there is only one illness to discover (that is, everyone either has ME or a known illness). We don't know if that is true.

Not really, I tried to be careful in how I worded it. It's about a potential difference with the way ME/CFS is defined in most research.

Most ME/CFS studies focus on patients with clinically defined confirmed ME/CFS. So patients did not simply fill in a questionnaire, they saw a doctor who probably did a lot of tests and took his time to figure out what might be causing the symptoms. That's usually a long and expensive process which makes it hard to recruit a lot of ME/CFS patients. So researchers like Leonard Jason or James Baraniuk have been trying to get to a questionnaire to help with the diagnosis of ME/CFS. I turn out that this isn't so easy because ME/CFS symptoms are not very selective. So questionnaires aren't usually used to make the diagnosis, only in helping to make the diagnosis for example by standardizing how symptoms are assessed or by forming a first selection filter with high sensitivity before going over to clinical examination.

So if the GWAS does use a questionnaire (+ a reported diagnosis of ME/CFS made by a clinician) it could select a patient group that is significantly larger than the patients researchers have thus far included in their studies. I think most studies on self-reported ME/CFS diagnosis have reported prevalences of over 1% (which is weird cause it's higher than prevalence estimates and you would assume that many ME/CFS patients haven't been diagnosed yet). So if the questionnaire has low specificity, it's possible that quite a large group of patients are included in GWAS that would have been excluded in normal ME/CFS research. So this is unrelated to whether ME/CFS is one illness or many.

What I'm worried about is that this group could influence the results of the GWAS and suggest a connection that wouldn't be there in patients with clinically confirmed ME/CFS. Given the size of GWAS, I suspect it will have a large influence on how researchers will view ME/CFS and how they will conduct further research.

So what I would prefer is to see a sort of test of the proposed diagnostic procedure first. How does it compare to clinically confirmed ME/CFS as used in current research? If the populations are similar and there's only a small difference (say 20% of patients selected with the GWAS procedure did not have a ME/CFS diagnosis after clinical examination) that would probably be no problem, cause our current method of selecting ME/CFS patients probably isn't that good either. But if the difference is large (say more than 50% of patients selected with the GWAS procedure did not have a ME/CFS diagnosis after clinical examination) then perhaps things should be reconsidered.

 

[ME/CFS Skeptic]

If 60% of the 20,000 ME/CFS cohort did not in fact have ME/CFS, maybe not having PEM, I think the chances that it would prevent picking up a genetic link would be small. You would just be diluting the ME/CFS genetic signal a bit.

What about the risk of picking up wrong signals, signals that have little to do with clinically confirmed ME/CFS?

 

[Hoopoe]

I think there would be a case for putting out recruiting material indicating that the study wants to study every single person with ME/CFS in the UK above 18. National newspapers and television services could be asked to make a story about it.

I like this.

 

[Jonathan Edwards]

What about the risk of picking up wrong signals, signals that have little to do with clinically confirmed ME/CFS?

Like you, I think the risk of picking up wrong signals is more of a problem than missing because of dilution. However, in that case the higher the proportion of true ME/CFS cases picked up the better. If you keep pushing until you have a figure close to 0.2% of your total pool you are more likely to recruit more bona fide cases than more misleading cases I think.

 

[mariovitali]

During my visit at the CureME team in March, we tried to see whether Machine Learning was able to differentiate Healthy Controls vs PwME based on questionnaire answers. The algorithms identified a single feature that was able to differentiate HCs vs PwME with very high precision/recall (unfortunately i do not remember the exact numbers)

Perhaps we could re-visit this work so i will contact the CureME team to see if we can pursue this further.

 

[Hoopoe]

If like me you were wondering where the UK prevalence estimate of 0.2% comes from.

Nacul et al arrived at 0.2% prevalence by searching electronic records of 142000 people for indicators of possible ME/CFS, then applying CCC or Fukuda criteria. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3170215/

They wrote that GPs without an interest in ME/CFS would often fail to diagnose it, which is relevant for this GWAS study.

We were careful to choose practices with GPs with experience in diagnosing ME/CFS. The effect of this approach on representativeness of cases was judged to be justified, otherwise we would have severely under-estimated the true disease occurrence, as GPs without an interest in ME/CFS would more often fail diagnose it.

However it's worth noting that in the US, Jason et al's study found a prevalence of 0.42% with methodology that was I think more reliable from this point of view (random households were contacted and assessed for possible ME/CFS).

 

[Medfeb]

They wrote that GPs without an interest in ME/CFS would often fail to diagnose it, which is relevant for this GWAS study.

At least in the US, I don't think this assumption is valid.

The US states execute a yearly patient-reported survey of diseases and risk factors. The CDC has piloted two additional survey questions on CFS in 5 states - a) has a doctor ever diagnosed that you have CFS and b) do you stlll have CFS. The findings were that 1.6% of people had been diagnosed at some point with CFS and 1.2% still have it (suggesting 1/4 recovered)

Remarkably high numbers given that a) it takes ME patients a long time to get a proper diagnosis, b) the IOM report noted one study showed up to 91% are undiagnosed, and c) a 25% recovery rate is considerably higher than typically reported (except of course by PACE).

It's really hard to know what's going on but one possible cause is that "CFS" has been used - at least in the US - as a wastebin diagnosis used by doctors for patients with any cause of chronic fatigue that the doctor can't otherwise explain.

 

[Adrian]

During my visit at the CureME team in March, we tried to see whether Machine Learning was able to differentiate Healthy Controls vs PwME based on questionnaire answers. The algorithms identified a single feature that was able to differentiate HCs vs PwME with very high precision/recall (unfortunately i do not remember the exact numbers)

Perhaps we could re-visit this work so i will contact the CureME team to see if we can pursue this further.

I don't think this is the right question.

Can you differentiate someone with MS but diagnosed with ME or differentiate people with depression vs ME with the questionnaire values.

It would be interesting to do some plots based on features from various diseases (and healthy controls) based on the questionnaire responses. Also do some feature engineering to look at which features seem to be IID for a given group and which seem to just be random across the whole set - which is essential if you are doing unsupervised clustering (or anomaly detection).

 

[Medfeb]

Can you differentiate someone with MS but diagnosed with ME or differentiate people with depression vs ME with the questionnaire values.

And can you distinguish between ME and patients who have unexplained chronic fatigue from other causes? Presumably a reliable PEM screen would help to the extent it is differentiating

 

[mariovitali]

I don't think this is the right question.

Can you differentiate someone with MS but diagnosed with ME or differentiate people with depression vs ME with the questionnaire values.

It would be interesting to do some plots based on features from various diseases (and healthy controls) based on the questionnaire responses. Also do some feature engineering to look at which features seem to be IID for a given group and which seem to just be random across the whole set - which is essential if you are doing unsupervised clustering (or anomaly detection).

Sure,in any case the questions should be posed by medical experts, not myself. So to rephrase : we have certain analytical techniques that can be put to use in order to investigate further the differences between these patient categories and under the right guidance i would be happy to help.

 

[Simon M]

My blog is now live (thanks, @Andy, for checking). Probably nothing new in it for anyone who has read this far in the thread.

Bold plans for two big UK biomedical research projects
November 8, 2019 Simon McGrath

​

The ME/CFS Biomedical Partnership, led by Prof Chris Ponting and Dr Luis Nacul, plan a huge genetic study and a major expansion of the UK ME/CFS biobank. The partnership will give patients and their representatives a major role in planning and running the project. The genetic research, a genome-wide association study, would need to recruit 20,0000 people with ME/CFS – and the researchers know they can only do with the support of the patient the community.

A new research team is hoping to boost UK biomedical research with a proposal for a very large genetic study and a major expansion of the UK ME/CFS biobank. The new ME/CFS Biomedical Partnership is headed up Professor Chris Ponting, and Dr Luis Nacul who leads the CureME and UK ME/CFS Biobank team. The partnership plans to submit a grant application for the two studies to the Medical Research Council (MRC) and the National Institute of Health Research (NIHR) early next year.

Not only are the partnership’s plans for research studies ambitious, but it is also setting new standards for involving patients and their representatives in research.
...

Read the full blog

 

[Chris Ponting]

I can only skim but has an age limit been mentioned? I was on Nacul's ME Registry but was too old for his biobank.

When we discussed this as a team, we thought that the only limit (because of consent issues) was the lower limit of 18 years, but we were not going to apply an upper age threshold. So: you'd be welcome to participate (if we are funded)...!

 

[Chris Ponting]

I shall be involved as an advisor. My view would be that it would be a serious mistake to try to extend recruitment beyond the UK at least for an initial definitive GWAS cohort. The biggest problem with the study is going to be recruitment bias. There is likely to be bias associated with all sorts of social factors - which may well knock on to racial imbalance and so on. That would immediately throw up genes like MHC, associated with race, which would be spurious.

If the sample is 20,000 of the maybe 120,000 in the UK then at least there are some constraints on how bad the bis may be. If the recruitment is extended outside the UK then bias is likely to go haywire. Populations from there countries might be useful as further comparators but a dilute sample from all over the place is likely to be the most misleading.

Thanks. Yes, we've been discussing a multi-stage process, with the first step being in the UK. Standard GWAS analysis takes account of population stratification, but I agree that we need to take care of residual bias and think about exploiting multiple different control cohorts. We plan to use the BOLT-LMM Bayesian mixed model association method which allows for substantial flexibility to adjust for co-factors, covariates or random effects, such as potential population structure and/or kinship among cases and/or controls.

 

[Chris Ponting]

A spit-and-post is basically how 23andme and others do it and they have international reach but it gets sent to their US locations. DNA seems stable enough for the couple of weeks it would take.

I don't know whether the rates of participation will justify it but I would definitely take part if international samples are accepted. Probably much easier within the EU, especially as there are EU-wide collaborators that already certainly know the people at the biobank. But sending the requests through the various researchers around the world would definitely speed up reaching 20K samples.

One of our inspirations is the “snowball sampling” spit-and-post approach of Genes for Good project (https://genesforgood.sph.umich.edu/) in the USA which uses social media to engage a large, diverse participant pool in genetics research and education.

 

[Chris Ponting]

I think it may also be important to map out what happens after the initial findings from this study appear.

For example will there be a budget earmarked to follow up say the top 5 findings - e.g. take a small sample of blood from patients who have the variants highlighted and perform proteomics and WGS/Sangar sequencing to determine if in fact the genetic marker has an effect on the resultant protein that the gene has a relationship with? Or at least a path to get that budget if there are positive findings.

Otherwise we will have data, but then have to wait years for funding to follow-up on it, and all the momentum will be lost.

There is obviously an awful lot of effort being put into preparing this proposal, which is not usual. Would be nice to make the most of it.

Yes, we are going to ask participants to consent to be recalled into new studies. This provides the exciting prospect of "recall-by-genotype" studies, and for the UKMEB to target individuals for blood samples that would be particularly informative. Importantly, the funds that we are applying for are not competitive with any other ME/CFS research projects so we are not hindering any other projects being put forward. Quite the opposite, because the genetics and biosamples should enhance most future studies.

 

[Andy]

One of our inspirations is the “snowball sampling” spit-and-post approach of Genes for Good project (https://genesforgood.sph.umich.edu/) in the USA which uses social media to engage a large, diverse participant pool in genetics research and education.

Thread on this project here, https://www.s4me.info/threads/genes...ch-via-social-media-2019-brieger-et-al.11898/

 

[Chris Ponting]

I don't think they will need to recruit healthy controls. They can use existing healthy controls data from other studies that do the same GWAS.

Yes, this is correct. Initially, for controls we will use the 500,000-strong UK Biobank study (not to be confused with the ME Biobank at the LSHTM).

 

[Chris Ponting]

I am not worried about some dilution of the cohort with misdiagnoses. If 60% of the 20,000 ME/CFS cohort did not in fact have ME/CFS, maybe not having PEM, I think the chances that it would prevent picking up a genetic link would be small. You would just be diluting the ME/CFS genetic signal a bit.

What worries me more is that getting a cohort through social media-based volunteering could very easily introduce spurious genetic bias.

Paradoxically, there is an argument for saying that what the study really needs to do is to get samples from at least 60% of PWME in the UK - say 70,000. It may be that people below 18 are not eligible and that might make it 50,000. The ideal would be to get 85%+. Having a high proportion of all cases would reduce the risk of recruitment bias.

I think there would be a case for putting out recruiting material indicating that the study wants to study every single person with ME/CFS in the UK above 18. National newspapers and television services could be asked to make a story about it. I suspect not that many more than 20,000 would respond and it might then be possible to back check using the sort of methodology the ME Biobank used to see where biases in recruitment lay and try to offset these in a sensible fashion. (Another approach might be to limit recruitment to a geographical area with only 20M people, but this would probably have more downsides than upsides.) As long as all that is done before a statistical analysis is undertaken it wouldn't fall foul of multiple analysis problems.

"the study wants to study every single person with ME/CFS in the UK above 18". I like this idea! Thanks.

 

[Chris Ponting]

What about the risk of picking up wrong signals, signals that have little to do with clinically confirmed ME/CFS?

We are going to look computationally to see whether genetic signals for diagnosed ME/CFS are shared with those for other traits using LD score regression. We hope to be able to find genetic signals that are only ever seen for ME/CFS and not for other inherited traits.