12/06/2021 Update: This simple Iceland covid predictive model from early 2020 held up surprisingly well. My mean predicted Iceland IFR was 0.17%. As of today the actual total IFR is 0.179% (36 deaths / 20,044 confirmed infections). My predictions for the US were not as successful (off by a factor of ~3), mostly because I overweighted this Iceland based model. Iceland is a more homogeneous (and high vitamin-D) population than the US.
A private company in Iceland, deCODE genetics, has provided valuable insight into true COVID-19 prevalence by PCR testing a random-ish sampling of Icelanders. You can clearly see the difference in the data itself: deCODE’s tests have a positive rate of ~0.9% which is about 10x lower than the positive rate of NUHI (a state hospital), as the latter is using a more standard biased testing strategy. This suggests that at least 0.9% of the Icelandic population has been infected with COVID-19. (The PCR test can’t reveal individuals who now have low viral levels.)
That’s at least 3.2K infections (0.9% of their population of 440K), and more realistically 4K to 5K.
Iceland has only 2 deaths so far for a naive IFR in the range of 0.04% to 0.2% to (we can probably ignore false negatives for deaths – as they are harder to miss in Iceland). Iceland’s cumulative case count is clearly in a linear growth regime (past midpoint of sigmoid). They have 6 patients in ICU (Iceland data), which has about a 30% fatality rate, and 19 in hospital with a 10% fatality rate so we can estimate the future total death count from this cohort in the 2 to 8 range.
This results in a mean predicted IFR of 0.17% (6/3500)and a range of 0.04% to 0.4% (2/5k to 8/2k), similar to influenza but potentially a bit (2x) higher. The uncertainty range will eventually tighten as we know more about survival in their current hospitalizations.
This agrees with the Diamond Princess data which rules out IFR much higher than influenza. (see my analysis here, or a more detailed analysis here) In that same post I also arrived at a similar conclusion by directly estimating under-reporting (the infection/case ratio) by comparing the age structure of confirmed cases to the age structure of the population and assuming uniform or slightly age-dependent attack rates similar to other viruses. That model predicts under-reporting of ~20X or more in the US, so it’s not surprising that the under-reporting in Iceland is still in the ~4X range.
Roughly let’s conservatively guess that a large fraction of the recovered were hospitalized previously, for 40 total hospitalizations. That’s a hospitalization rate of 1% (40 / 4,000), which is a little less but close to the CDC estimated influenza hospitalization rate of ~ 1.7% (500K hospitalizations / 29M infected for 2016).
This also puts bounds on how widespread C19 can be – with IHR and IFR both similar to influenza, there couldn’t be tens of millions of infected in the US as of a few weeks ago or we would be seeing considerably more hospitalizations and deaths than we do.