Has there been any ME/CFS treatment trials that adequately explored the "it works for a subset" concept?

[Andy]

When I see the claim that there are treatments for ME/CFS but they only work for a subset my assumption is that this based on a misunderstanding of the results of trials; that people making the claim don't take in to account the possibility of placebo effect or natural improvement. Is my assumption correct or has there been any treatment trials for ME/CFS that have been designed in such a way to allow us to see if there are genuine responding sub-types?

 

[BrightCandle]

It is mostly said about trials without controls and long term tracking. The data will show some small improvement on average and maybe 20% showing a response in a PROM measure. They tend not to have longer follow up periods so the averaging out that happens for a condition that varies also hasn't happened and without a control its therefore not a very strong signal of value for what is being tested in the trial.

There is not enough data in the study to identify what was different about the claimed subset and without a control it can just be the natural progression of a condition that tends to vary. Some percentage of patients will be improving at that time, some getting worse. Its just the usual slight positive bias of trials on qualitative PROMs and its an artefact that when tested (so far) disappears when a control and longer term tracking is tested in a better run trial.

 

[wallfish]

Some ongoing studies funded by MECFS Research Foundation have a priori defined subgroups (two based on groups identified post-hoc in earlier studies).

 

[fst]

I also think the subset claim is mostly used as a cop-out to justify failed trials.

I think it was definitely attempted many times in exploratory post-hoc analyses. In a way, trying correlate a marker with an outcome can be seen as an attempt to identify a subset.

For example, the protocols of the not yet published immunoabsorption trials read as if they wanted to identify subsets based on the autoantibodies that are present. But that would only work, if there actually were be a responsive subset that is characterized by this. So without working treatments and "biomarkers" how to even identify a subset?

In the Daratumumab trials, one could argue that they hypothesized a subset based on NK cell counts for their phase 2.

 

[Jonathan Edwards]

Is my assumption correct or has there been any treatment trials for ME/CFS that have been designed in such a way to allow us to see if there are genuine responding sub-types?

The blinded Fluge and Mella trials and probably some large phase III antiviral trials would have been well capable of identifying responder subsets if there had been an effect. Building subset analysis into trials is not particularly sensible or necessary if you are at the stage of trying to see if the treatment works at all. When we developed rituximab for RA there were subsets we knew might be relevant but rather than muddy the statistics we looked at these post-hoc. An obvious one was whether rituximab worked for people without autoantibodies - whether it worked for 'seronegative' cases. We were able to establish with pilot studies that they probably did not and that was confirmed in later formal trials. But we also learnt that the rule was not absolute. People who had no autoantibodies measurable but who had abnormalities of antibody subclass levels did respond - which makes sense for rather complicated reasons.

In the pilot daratumumab studies by F and M NK cell numbers appeared to indicate a 'responder subset'. The trial was not big enough to give a definite answer but the ongoing blinded trial will be able to provide responder subset indicators if there is a response.

You might think that subsets should be defined prospectively, but because there are several ways you might reasonably subset i don't think this is statistically very intelligent. You might say you could miss an effect on a subset but because of the multiple analysis problem I think it si pretty unlikely that you would do better by predefining.

There might be some obvious situations where a 'subset' makes more sense than taking the diagnostic group as a whole. Orthostatic intolerance is an obvious one. If you are wanting to see if drugs affecting heart rate are useful it only makes sense to give them to the subset of people with ME/CFS with abnormal heart rates. I am not sure that any good quality studies of that sort have been done though.

In short, the simple answer is that F & M's methodology has been good enough to pick out subsets but so far they have not shown the drugs actually work.

 

[SugarSquared]

I think the recent Paxlovid trial, while for Long Covid and not ME/CFS, looks at the subtypes quite well. Long Covid is an illness that requires subtyping because it is an umbrella term for multiple illnesses. They looked at three subtypes: cognitive, autonomic, and exercise. They had primary and secondary endpoints fitted for each subtype. And yet, this is what Amy Rochlin of the Complex Disorders Alliance said, and David Putrino agreed with her (thank you to @ChronicallyOverIt for finding this exchange).

In this case, the subtyping argument just sounds like this:

I also think the subset claim is mostly used as a cop-out to justify failed trials.

 

[Silip]

I think the recent Paxlovid trial, while for Long Covid and not ME/CFS, looks at the subtypes quite well. Long Covid is an illness that requires subtyping because it is an umbrella term for multiple illnesses. They looked at three subtypes: cognitive, autonomic, and exercise. They had primary and secondary endpoints fitted for each subtype. And yet, this is what Amy Rochlin of the Complex Disorders Alliance said, and David Putrino agreed with her (thank you to @ChronicallyOverIt for finding this exchange).
View attachment 34085
In this case, the subtyping argument just sounds like this:

It is ridiculous. Taking up the argument by Mr. Edwards from above, how likely is it that a group of true responders 'hides' so well that it doesn't show up in any(!) pattern in the data at all?

If I remember correctly in most trials where they claimed it "may help for a subset" they never found any significant similarities for their "responders", not in any basic things like sex, duration of illness, not in markers tied to the treatment not even in patterns of improvement like timing, improving symptoms, magnitude etc...

Not that the existence of such patterns without proven relevance would show that there is an existing subset, but that the absence of any patterns makes it really likely that there isn't one.

I'm stunned that some people still insist on the subset idea for their preferred theory or treatment after they weren't able to find any subsets even in post-hoc analysis.

 

[Kitty]

If I remember correctly in most trials where they claimed it "may help for a subset" they never found any significant similarities for their "responders", not in any basic things like sex, duration of illness, not in markers tied to the treatment not even in patterns of improvement like timing, improving symptoms, magnitude etc...

I don't understand how they can fail to grasp that it's quite likely a proportion will feel better after treatment. Some may even be able to return to activities they previously couldn't do.

This has to be expected with conditions that can improve (and occasionally resolve) without any therapeutic intervention. In the absence of objective markers, the only way to deal with improvement is to follow up at two years, and ideally again at three or five.

If adequate follow-up can't be built into a drug trial, surely the only reliable conclusion that can be drawn is that it doesn't work? It can't be said that it works for a minority, even if it does.