Comment by lucgommans | Hacker News Reader

Comment by lucgommans | Hacker News Reader

Parent Full thread

Paper on which the code is based: https://www.nature.com/articles/s41591-020-0916-2

The RIVM, the Dutch public health agency, referred me to this paper after I asked about combinations of symptoms (their website[1] only lists individual symptom occurrence percentages for seropositive and seronegative cases). Given that most people will not read a scientific paper and calculate that formula, I thought this might be helpful to others.

Disclaimer, as is also written on the site in bold and red: this is NOT validated or endorsed by the authors of that paper and this is NOT a replacement for a real COVID-19 test! The point is to give a rough indication using the same phrasing and exact model from the paper.

[1] https://www.rivm.nl/pienter-corona-studie/resultaten (Dutch, see "Tabel 1")

I ‘tested’ symptoms multiple times and found that effectively the one and only factor that could put me above the 50-mark was loss of smell & taste. Couldn’t really understand the significance of the large difference in score when choosing Male/Female either.

That is the main factor indeed, but the other questions do change the amount by which you go above 50 a lot, and there are other paths to get to being predicted positive without having loss of smell or taste (LOST). From what I read and if I remember correctly, different studies say somewhere between half and a third of the people did not report LOST while they did have other symptoms (there are also completely asymptomatic cases of course, but that's iirc more like 1/5 to 1/10 of young people). Even if you have no LOST, the model can return positive, despite LOST being very characteristic.

I don't know why male/female or age matters, my best guess is that the younger males in their sample were more likely to have a positive test result. That seems like a pattern which would shift over time as the media claims different age groups cause it, different measures are taken, etc., but I'm not really qualified to question their results or change the model without (new, or so much as the original) data.

Reply on news.ycombinator.com