Multi-method phenotyping of Long COVID patients using high-dimensional symptom data, 2024, Green et al

Multi-method phenotyping of Long COVID patients using high-dimensional symptom data

Background
Long COVID, characterized by symptoms that remain or emerge in the months after infection with COVID-19, has complex and highly variable patient presentations, with myriad seemingly disconnected symptoms.

Methods
We apply three different machine learning techniques to identify groups of patients with similar symptoms in a large patient-reported symptom dataset with the aim of identifying robust Long COVID phenotypes.

Results
All three methods produced clinically plausible symptom clusters which are technically valid partitions of the high-dimensional symptom space. However, concordance across methods was low. Some features did recur, such as low-symptom count clusters having the highest average age and lowest proportion of women, and specific recurrent clusters or subclusters across pairs of methods.

Conclusions
The high sensitivity of observed patient clusters to algorithm choice has implications for other studies reporting Long COVID phenotype clustering, as it suggests that a single method may provide an incomplete or unstable partition of the cohort, particularly in studies with fewer symptoms observed. With the 162 reported symptoms considered here, patient presentations vary smoothly and segmentation, while internally consistent, was not reproducible across methods; this suggests that the complexity of LC symptom presentation may easily be missed by clustering approaches that use insufficient data or overly-simplistic clustering methods. Future work would likely benefit from semi-supervised approaches matching patients to pre-defined phenotypes or diagnoses, or from the inclusion of additional patient data. Overall, our multi-method analysis highlights the importance of assessing clustering robustness and considering the full scope of patient symptoms when evaluating treatments.

Link | PDF (Preprint, open access)

 

Not sure what the added value is of machine learning techniques in this context.

EDIT: I have little to no experience with this so would be happy to hear if someone could explain the use case a bit more.

 

I think this is from the patient-led team that sprung out of the Body politic group that published the first LC paper. It looks excellent and just about confirms what other studies that have tried the same: you can find clusters of symptoms, if you want, but they aren't much useful, as neither are symptoms on their own.

This is all consistent with the obvious patterns of wildly fluctuating symptoms that vary from person to person and over time with individuals. The 2nd Body politic paper had a very good rundown of symptoms and their timeline and was strongly indicative of that. Symptoms also tend to come and go and vary wildly when you assess them at different time points.

So what we remain left with is the fact that differential diagnosis is the cornerstone of medical assessment, it's utterly useless here, and there is no plan B because the assumption is that if you can't do differential diagnosis, then there is no disease, symptoms are non-specific and indicative of "somatization", or whatever people want to believe.

Which is... not good. Because it means that medicine has to acknowledge a fundamental flaw in its primary process of assessment, and it has literally created an entire pseudoscientific ideology out of everything in that gap. A gap which is enormous, probably amounts to 1/3 to 1/2 of all health problems. Something that cannot be walked back without exposing the biggest scandal in the entire history of professions, all backed by governments and greedy powerful interests.

A scandal that will inevitably be exposed. It's just a matter of time. But they have already sacrificed tens of millions of lives, and likely hundreds of millions of lives-equivalent, and seem to have no ability or interest in doing so. So in the meantime, the stranglehold holds.

 

It can find patterns when they are too complex for humans to find. Machine learning is really perfect for this as it's built out of the same problems: connections and directions in highly multidimensional (as in thousands of dimensions) fields.

But there really aren't useful patterns here. Nothing predictive. Symptoms alone aren't enough for this, there is too much missing data.