KNIB on the Finland Study

On Tuesday, April 7, 2020, 03:32:04 a.m. ADT, K.N.I. B. wrote in an e-mail to me (bolding mine):

Useful article (below). Particularly interesting that the Iceland study showed 50% of positives had no symptoms.

“However, there was a study in Iceland*, which tested [a large sample of its] population, and 50% of the people who tested positive had no symptoms.”
[*Iceland study:]

‘had no symptoms’ — does that mean not showing symptoms at the time, or had not shown and did not later develop symptoms? They probably would say the latter if they meant the latter, so it probably intends to say those people were asymptomatic through their entire experience of COVID-19.

The study covered 5,571 cases, of which 48 cases were positive. 48 cases is a large enough sample to say 50% with fair precision, so I wouldn’t be surprised if the true value was 47%, but quite surprised if it was 35% or 65%. So we can take it as pretty reliable that a large proportion of people once infected have no symptoms.
But 48 isn’t a lot of cases to ask more complex questions. Some questions can only be asked of positives and only when you have enough variation to ask the question.

What I’m keen to know is: how is variation in percentage (or, looking at individuals, variation in risk) explained by knowable factors, e.g:
– smoking
– good/bad diet
– underlying conditions
– activity levels
– how much VitC (or A, B, D, E…) they took
– did they use saline nasal rinse at the first sings of trouble?

And that could be asked if data from multiple countries were pooled, and presuming enough cases where there were data on the factors of interest. (If your analysis has, say, 9 factors [m=9], then a case with missing data for one of them will typically drop out of the entire analysis. And you’d need at least n ≥ 2m+1 cases to analyse, though more is always better. So it means a fairly complex analysis can be done with a fairly small number of cases. Those cases exist — but it seems nobody has put them together yet, though I’m sure they will.)
(For those that don’t do a lot of statistics and who might think a simple analysis is better, it might interest you to know that asking 9 questions together in a single 9-variable analysis will generally result in a more robust answer for each than could be had from 9 single-variable analyses.)