Comment by amluto

> "no increased risk of all-cause mortality"

My take is not quite as charitable as most of the comments, although my objection only barely applies to this particular paper. Biomedical research has long tradition of a very specific type of analysis: hypothesis testing. Roughly speaking, you make a hypothesis and test your data for compatibility with the hypothesis. In this paper, the authors are comparing two hypotheses: (a) there is an increased risk of all-cause mortality or (b) the increase in risk is zero or negative [0]. And the statement you’re quoting from the conclusion section sure sounds like it’s saying that the authors found that the data was consistent with (b) but not with (a).

Researchers love this. There are lots of papers with fancy-named tests with which one can do this analysis. Regulators often demand it.

And it produces papers that are correct in a fairly literal sense but miss the point entirely. For example, “we found no evidence that vaccines increase the risk of autism”. I, too, can look under my bed or study four people and find no evidence of anything at all about vaccines and autism, and I would be more or less justified in making that claim.

And because of this, you need to read papers very carefully to see what you can actually conclude. “No evidence” means a lot more in a large (“high powered”) study than in a small, weak study.

The right way to do this is something like “we looked at such-and-such data and found, with 95% confidence, that the increase in risk of X is <= 0.2%” or even that “the change in risk is <= -25%” (check out that minus sign!). If I look under my bed, I will find evidence that the increased absolute of autism caused by vaccines is <= 100%, and I didn’t need to study anything to confirm that :)

Fortunately, this particular paper has the silly throwaway starting you’re complaining about in a few places but is otherwise mostly on the right track. Quoting from the Results section:

> Vaccinated individuals had a 74% lower risk of death from severe COVID-19 (weighted hazard ratio [wHR], 0.26 [95% CI, 0.22-0.30]) and a 25% lower risk of all-cause mortality (wHR, 0.75 [95% CI, 0.75-0.76]), with a similar association observed when excluding severe COVID-19 death. Sensitivity analysis revealed that vaccinated individuals consistently had a lower risk of death, regardless of the cause.

IMO that should have been the headline. The latter sentence there is giving some indication that the result is robust and that the researchers didn’t flub it or get unlucky in quality of their data.

P.S. The fact that the vaccines seemed to reduce the risk is death from causes other than severe covid is interesting. I wonder how much is reducing death from complications of COVID other than “severe covid”, how much is behavioral changes (vaccinated whole socializing more and leaving their houses more), and how much is bizarre off-target effects of the mRNA vaccines. Here’s an example of a surprising off-target benefit that no one understands well:

https://www.science.org/content/blog-post/mrna-vaccines-and-...

[0] I think. I didn’t reread it enough times to rule out the other common test: is the data consistent with the hypothesis that the risk didn’t change at all, which is also sadly common. But I’m moderately sure I’m right in the context of this paper.