Hacker News Reader

Comment by azakai | Hacker News Reader

That's why the study looked at a control group, to try to account for that bias (as the abstract says, the control group was chosen to match "personal and clinical characteristics").

It's always possible they missed a factor, that's always a risk in such research, but they worked hard to address that possibility.

View on HN

Not matching for hospital admission status is a pretty big factor given the bold, highly publicized top-line numbers:

> We selected controls from the general population rather than matching to non-covid hospital admissions to determine the increased risk after hospital admission for covid-19 versus no hospital admission for covid-19 (that is, compared with the expected risk for people with similar personal and clinical characteristics in the general population).

(I disagree, see discussion above: https://news.ycombinator.com/item?id=26672461 )

Essentially any contact with a hospital in that population produces eye-popping statistics. Having worked in this area, I’m of the opinion that the presentation of this work is misleading.

But the opposite would be misleading, too, in that the hospital contact was induced by COVID.

It's all confounded. If you compare to other hospital admits, you're finding an even-less-healthy population (because they had a reason to show up to the hospital -without- COVID). If you compare to the general populace, you risk inadequately controlling the comparison.

But if you want to know the increased risk people have subsequent to COVID hospitalization, perhaps the latter makes the most sense to measure.

My beef isn’t with the measurement, but with the presentation thereof. This problem you have recognized is so common that a framework was developed [1]. I don’t feel the work addresses all element of that framework satisfactorily. Others may disagree, and that’s okay.

[1]: https://en.m.wikipedia.org/wiki/Strengthening_the_reporting_...

Reply on news.ycombinator.com