Hacker News Reader

Comment by Protostome | Hacker News Reader

I work as a data scientist in one of the largest US healthcare companies.

We ran a similar analysis back in June and found comparable results based on our claims data. People with low vitamin D rates had higher likelihood of hospitalisation and mortality. Back there were less indications that vitamin D has any effect on Covid severity, and when I presented the analysis to our chief scientist he dismissed my conclusion.

View on HN

Ok... but why was your conclusion dismissed? This is very troubling.

Quoting him: "Our data is so noisy, you can't really infer such a relationship from claims data" Since all those analyses are retrospective, it's not easy to build a case that will show a causal relationship between the outcome (hospitalization/death) and the covatiates (vitamin D levels). Some will say that this is impossible and only prospective studies will provide an unequivocal evidence.

Therefore, it is not rare that retrospective studies are encountered with skepticism. What baffled me is that recommending people to take vitamin D has almost zero negative consequences, so why not? the data supports it, even if it's not a solid rock proof.

That makes sense to me. If the evidence is not rock solid, you probably should not be making medical suggestions to your customers. The idea of my insurance company suggesting I should take supplements where that isn't within the typical medical consensus or part of government advice would make me uncomfortable.

That sounds like one of the reasons why there were less indications.

Reply on news.ycombinator.com