[svnt]

There are some major qualifiers to this study:

1) The ages and BMI:

> The median age was 45 years (IQR 37–54) and median BMI was 29·8 kg/m2

2) The diagnosis of long COVID-19 was done by survey — this has been shown to be an unreliable method of assessing persistent symptoms of disease. This is in part because people will tell you they have symptoms if you give them a cause and ask. The diversity of the claimed presentation of symptoms is an indicator (e.g. > 50 different symptoms by article-in-support [1])

3) The hazard ratio 95% confidence interval extends to 0.99 when treatment is started within three days. This is not a strong result. 1.0 is no effect.

[1] https://www.medicalnewstoday.com/articles/symptom-burden-sur...

[mft_]

> 1) The ages and BMI:

> > The median age was 45 years (IQR 37–54) and median BMI was 29·8 kg/m2

Not sure what the problem is here? The age seems nicely representative without being too young (deal with COVID better) or too old (deal with COVID worse; and higher liklihood of comorbidities). The BMI is a little high, but then we know that larger people have a higher COVID risk, so maybe this makes sense? Either way, given this is a study of treatment effects, and the BMI was well-balanced between both groups, meaning it's reasonable to assume that it didn't affect the overall findings.

> The hazard ratio 95% confidence interval extends to 0.99 when treatment is started within three days. This is not a strong result. 1.0 is no effect.

The hazard ratio itself is 0.37, which is a pretty strong effect, and the effect is statistically significant. (Also, slight correction: the upper bound of the hazard ratio is 0.95 when treatment is started within three days.)

[svnt]

The issue I mentioned in reply to another comment below is that metformin, the medication they were having success with, is more commonly used as a treatment for type 2 diabetes.

[mikeyouse]

This is partly what I was talking about. Just odd dismissals of actuals scientific evidence.

It's a "major qualifier" that this study looked at middle aged overweight people? Or that they relied on a survey to assess symptoms but then quadruple blinded everyone? They added the long-covid assessment as a secondary endpoint to a group they were studying since before "long Covid" was a phenomenon.

And then to dismiss the result because you apparently misread the CI for the hazard ratio? For those on metformin, it extends to 0.89 not to 0.99 -- and is centered at 0.59 with a P = 1.2%. You can obviosuly quibble about subgroup analysis but for the 3-day group, it's 0.37 [0.15 - 0.95]. That's a fairly strong result!

[svnt]

The problem is always in the simplification as it is translated into lay conversations.

You cannot extrapolate from a median of obesity into the general population.

This is especially true when the medication you are having success with (metformin) is a treatment for diabetes.

What are they even measuring?

[mft_]

> This is partly what I was talking about. Just odd dismissals of actuals scientific evidence.

I've read (and heard first-hand) similar approaches by people with a (sometimes hidden) agenda to push. There's a lot of apparently "scientific" stuff out there, written confidently by people who ultimately aren't qualified, which is lapped up by those deep into the confirmation bias of whatever is topic is.