Citat:
Ursprungligen postat av
Mindbusiness
Med tanke på hur fel våra professorer räknar, och hur träffsäkra killgissningarna här på Flashback är, har jag gett mig på att räkna lite jag också.
Om 7.1% av Stockholmarna varit eller var smittade vecka 18 blir det ca 175.000 individer.
Under denna vecka hade det totalt dött 1.330 stockholmare. Det innebär vad jag kan räkna mig till - en dödlighet på ca 0.76 %.
Låter kanske inte så farligt. Men räknar man med att större delen av svenska folket framöver får Covin-19 så innebär det mellan 35.000 - 45.000 döda...
Rätta mig gärna om jag har fel!
7.3% *
But you also have to take into account not only missed deaths but also the fact that the true percentages are so low true-positive prevalent that the false-positive rate of the tests sway the results dramatically. This can be seen when comparing Skåne and Göteborg — obviously the result of false-positives.
This means that only the Stockholm result is anything to look at, but even there it means that the 7.3% infection rate will be lower than the actual true-positive rate. It also means that results such as 65+ infection rate could range anywhere from 0.1% to the result given. Considering how bad the testing parameters are, we simply don’t know.
Also worth noting that since true-positive prevalence is so low, that the false-negative prevalence will be minuscule due to the number of true-positives. Low results like these will always lend themselves to a false-positive skew for that very reason. This is what studies such as the Gangelt pre-print have been critiqued for, so very important to remember when looking at low-prevalence antibody results such as these.
I imagine we’ll find around 5% infected in Stockholm at that period in time, and that adjusting for age and excess mortality (clinically diagnosed + non-tested deaths for example that aren’t included in the death toll) it’ll land somewhere around 0.9% to 1.3% — likely around 1.2%.
However it could be higher, which would be due to the pre-triage/triage undertaken. I don’t know what effects on the IFR that will have, but that degree of hospital collapse skewed Lombardy to 1.29%, so I would hope for no higher than this.
60% of IgG antibodies present after something like 10 days. 90% after 14 days and 99% after 21 days. I reckon taking the death tally twelve days after testing concluded (the sixteenth) and then work with that. If you use that to inform the IFR assuming the best-case 7.3% result and then stratify it for age, you’ll probably land at around 1%.
Also worth noting that other antibodies present faster and that IgG antibodies are the slowest to present. I believe they measured for IgM and IgG antibodies, meaning that waiting twelve days after results have concluded will give you a realistic insight into the death totals for this prevalence.
This is going by the median time to symptoms presenting after infection of 5 days. It’s also based on the 18.8 days to death after symptoms present estimate by the Imperial College studies.
23.8 days to death vs twelve days for the large majority of IgG antibodies and large majority of IgM antibodies = twelve day lag between that prevalence and deaths.
Mathematical approaches like the Bayesian Inference Method by Gellman can also help discern true-infection rate by calculating the test parameters against the result:
Examples, Sweden:
π = (p+γ−1)/(δ+γ−1)
γ = Specificity (0.977)
δ = Sensitivity (0.983)
p = Prevalence (0.05)
(0.027)/(0.96) = 0.0281
Which calculates a true-prevalence of 2.81%.
Another Bayesian example in short for Stockholm:
(7.3 + 97.7 - 100) / (98.3 + 97.7 - 100) * 100
Gives us a true-prevalence of 5.21%.
