It is very positive in that we can be more confident that our assumptions are close to reality. 0.5%*330M still equals 1.65M deaths, or (very roughly) 5M ICU beds. Spread out over 52 weeks, that's a lot of ventilators and ICU beds and PPE, but it is a target. A very big juicy target that we can and should aim directly at, and HIT.

So having killed 1.6 million in the US alone and run ICU at 100% for a year, you’re just going to hope people remain immune and the virus doesn’t mutate having passed through billions of people worldwide?

The alternative being...? We know that covid-19 will likely mutate, just like other coronaviruses, H1N1 and its mutations are still around every year; are you suggesting we must remain in quarantine until we both create a vaccine for covid-19 and then every fall quarantine until we're able to develop a vaccine for that year's bout of viruses?

It's more or less epidemiologically impossible that you will see an attack rate of 100% (as you just assumed). Once attack rate reaches >40~60%, the reproduction number drops below 1 because there just aren't enough uninfected hosts.

It will probably be significantly higher than 60%, since COVID-19's R0 without lockdown seems to be pretty high. The CDC has estimated it at 5.7 [0]. Herd immunity threshold = 1 - 1/R0, so the threshold would be 82.4%.

[0] https://wwwnc.cdc.gov/eid/article/26/7/20-0282_article?deliv...

That R0 assumes we are taking essentially no measures to stop the virus though, right? I don't no much about it but it seems like the R0 would drop significantly if we keep doing social distancing, expand testing, make people wear masks, etc.

It does, but the point of getting to herd immunity is that we don't want to do any of that other stuff.

The 5.7 number comes from wuhan, during lunar new year, and before viral identification.

The issue is you didn’t factor in immunity fading and requiring it again. So those that survive will need to roll dice at a random later date.

Right. Really this study is just confirming (with higher error bars than we'd like) what we already strongly suspected.

Note that as others are pointing out, that 0.5% isn't really extrapolated correctly. The tested case deaths are looking like they're only 70% of the total or so. Also using a current death count with contemporary infection counts is a mismatch, because it forgets that ~35% of the people currently in the hospital won't ever leave, which is about 5k more deaths.

So a better calibrated CFR might be 1% or so, which is even closer to the range we've all been assuming.

Except this data is also lagging by three weeks, depending on what antibodies are being checked. Some deaths today have come from infections in April, so the net effect could be a wash.

And where are you getting a 35% death rate for "ever hospitalized"? It's realistically somewhere around 15%-20%, though with major age skew. (https://www.thelancet.com/journals/laninf/article/PIIS1473-3...)

Per https://covidtracking.com/data#state-ny there have been 57103 hospitalizations in the state (not city, so not quite the same population), of which 15021 are still hospitalized. So that's 42082 cases that have left the hospital, of which 15740 died, or 37%.

Obviously there's some apples/oranges here, both due to the population being different from the city and the fact that some of those deaths were outside of hospitals. But it's close.

It's been covered in the past that NYC's total fatality rate has been good but its ICU survival rate has been anomalously bad.

Perhaps a lot of those deaths were outside hospitals (nursing homes, etc.)? Here's a report for all hospitalizations in NYC from March 1 to April 4: https://jamanetwork.com/journals/jama/fullarticle/2765184

21% of patients who discharged or died died.

IFR, not CFR

Yeah, still 150,000 death toll if everybody gets infected. And this is not considering the permanent lung and other tissue damage that's getting reported from "recovered" patients.

You are one order of magnitude too low at 150,000. It's much worse than that.

Yes, you're right.