All of those are examples of exactly what I told you about: they take a group of people claiming to be sick, and go hunting for signals to claim as “significant”.

The MRI studies are particularly egregious examples of this. Just because you see a difference on an MRI does not mean that the difference is due to the thing you’re blaming. In fact, it almost never is.

> If your argument is that people are showing up with abnormalities, then diagnosed with Long Covid, then spurious biomarkers are associated to it - you are just wrong. Wrong multiple times. Demonstrably so.

I am? I have now followed every link. Literally every paper you posted is following this exact pattern. I don't know how you could possibly conclude otherwise, unless you just didn't read past the titles.

They each take a (typically small) cohort of people who self-identify as "long covid sufferers", they subject them to random combinations of tests, and report only what they find to be significant. It's literally the XKCD comic about jelly beans.

https://xkcd.com/882/

You are just ignoring the evidence, being unscientific, and unless you work for a top medical lab somewhere, plain arrogant.

The UK Biobank study scanned participants before and after infection with matched controls. The difference is measured against their own pre-infection brain. That is the opposite of what you're describing.

> You are just ignoring the evidence, being unscientific, and unless you work for a top medical lab somewhere, plain arrogant.

If you don't know how to interpret evidence, then I suppose it would sound like I am being overly critical. I didn't bother to pick on just one, but since you chose it [1]...

> The UK Biobank study scanned participants before and after infection with matched controls. The difference is measured against their own pre-infection brain. That is the opposite of what you're describing.

It is not. The longitudinal nature of the study is a distraction from the fundamental issues with the approach.

They did a longitudinal case-control study, one group of which had positive covid tests in the past, and the other one did not at the time of the second scan (2021). That's the entire evidence base that this study is built upon -- it has nothing to do with "long Covid", and it's only barely plausible that the control group is actually a control for the factors of interest.

Next, they took two scans for all participants - one from before the pandemic, and one made after (again, in 2021). They made over 6000 different images, and then cherry-picked the ones with differences for further analysis (~70). Ultimately only 6 of these fishing expeditions survived family-wise error correction:

> The main case-versus-control analysis between the 401 SARS-CoV-2 cases and 384 controls (Model 1) on 297 olfactory-related cerebral IDPs yielded 68 significant results after FDR correction for multiple comparisons, including 6 that survived FWE correction

So first off, no statistical correction can compensate for this fundamental bias. You cannot start with thousands of different samples - even if they're taken from the same people at different time points - and winnow that down to a handful by filtering on the outcome of interest, Applying a multiple-sample correction will not fix it. It's not even clear that there is such a correction that is valid for the underlying distribution of the data involved.

But setting that aside, the differences observed, even between longitudinal samples, do not have to be due to Covid! Even if they're not random (which we cannot grant; see previous paragraph) they could be due to everyone being locked inside during 2020. They could be due to factors completely unexamined by the study, like, say, increases in drinking or drug use, or lack of exercise. Or any of a million other things. We don't know. The authors don't know. They're just not intellectually honest enough to admit that they don't know.

I could go on, and point out more flaws (e.g. the "significant" results mostly disappear when you exclude hospitalized patients, yet oddly, the difference between "hostipitalized" and "control" cohorts is not itself significant, indicating inadequate statistics), but this post is already too long.

I'm sorry that you think this is arrogant, but this is how we actually read papers.

[1] https://pmc.ncbi.nlm.nih.gov/articles/PMC9046077

This seems to me like a performance at this point and not serious analysis.

It’s true I conflated this with long covid. It’s not a long covid study.

I am tired and done with this. You made several errors in this comment.

Your biggest error is the lockdown one.

This makes no sense whatsoever - the controls also lived through lockdown. If this is the rigorous analysis you’re bringing to the studies you read, I’m not surprised none of them pass the muster.

“No correction can fix it” is wrong because the olfactory IDPs were pre-specified. “Could be lockdown” is wrong because controls lived through the same lockdown. “Results disappear excluding hospitalized” is wrong because the paper says they persisted.

The statistical weaknesses you describe are in the papers own limitations section. You just read them back as limitations that can’t be surpassed while evidence based researchers in the field disclose them as meaningful but not exclusionary.

Unless you want to continue with debunking every other strong paper I’ve posted with similar limited and likely to be demonstrably wrong takedowns, then I can’t help you. You have unfalsifiable priors, are constantly ignoring evidence and seem to believe you know better than the top researchers in the field - people who are saving lives - because you catch some statistical limitations and imply that they debunk the entire thing, instead of accepting them as limits of incomplete research into a real condition that’s crippling millions of people.

> the controls also lived through lockdown. If this is the rigorous analysis you’re bringing to the studies you read, I’m not surprised none of them pass the muster.

You've missed the point. I'm not suggesting that the other factor or factors has to be "lockdown". I'm just giving examples that illustrate the idea: even if you assume that the differences between the control and the experimental group are non-random and significant, you still cannot attribute the longitudinal difference to the one factor alone. If you don't like my theory, it's easy to find another, if you're even a little bit imaginative.

> “Results disappear excluding hospitalized” is wrong because the paper says they persisted.

No. They lose all but one. The final "significant" result is teetering on the edge of insignificance. See table 4 [1]. Models 2-4.

> the statistical weaknesses you describe are in the papers own limitations section.

Yes, because they're real. It's great that they wrote them in the paper, but they're fatal flaws.

"We openly disclosed the reason our study is nonsense!" is not the damning indictment you're suggesting that it is.

[1] https://pmc.ncbi.nlm.nih.gov/articles/PMC9046077/table/Tab4/

Yes of course.

It’s lockdown and now no lockdown. Could be anything. All observational studies are wrong. The stated limitations are fatal flaws. You heard it here first in HN. All medical research is fatally flawed, says user “timr”.

Good luck with that.

> All observational studies are wrong...You heard it here first in HN.

No, but most of them are wrong, and all of them need to be treated with an incredibly high degree of skepticism. This is critical review 101. When you push on this paper, even lightly, it falls over.

Not all papers are bad, but this one is bad, and while there are a great many well-done studies in the world, the subject of "long covid", to date, has essentially ~none of them.

I knew this would be the conclusion. Again - good luck. You are always right.

If you’re right and everyone else is wrong about hundreds if not thousands of studies, then you should be writing a book, not comments in HN.

We started at “some studies have errors” and we ended in “an entire field of research is wrong”.

You have already decided the field has no valid studies. Even when given dozens of examples you picked one and made up a series of points about one study. You made mistakes, never admitted it, and now are calling into question an entire field of medical research.

Again. Good luck with that.

I’m not even sure you understand how evidence based medicine works.

Afaik evidence based medicine ranks mechanistic analysis near the bottom of the hierarchy — below controlled trials and systematic observation. I believe that ordering was a deliberate choice.

You seem obsessed with something that modern medical research often doesn’t focus on - by design. We still don’t know how lithium works 50 years past its introduction. We don’t know how the conditions that it treats - psychosis or bipolar - work either. Yet lithium is used all over the world- because the effects data and reports show that it works. Your mechanistic obsession isn’t just wrong - it’s directionally incorrect as far as a lot of medical research goes.