so whats your true rate then, deaths divided by cases that are 4 weeks old?

i feel like that's not the best way of calculating it. it seems like there's a ton of variables that can cloud the authenticity of that data.

for example, one assumption you're making is that the population doubles every week. i've been reading from a lot of smart people that the population is doubling every 3 days. who knows what true rate is based on veracity of the strain (you may have diff ppl saying diff things) but tweaking that doubling rate i assume messes with your calcs a pretty strong amount.

We don't know the true rate, but:

- With population level sampling we know a an absolute lower bound, and it's 0.6%.
- We can discount at a minimum the last 2.5 weeks of infections in deaths, which gives 3x at least the current death numbers

We can't do the latter if we don't know the population infection rate, but we do in Korea and China now.

Like what? The only possibility is that the population-level tests aren't picking up a huge number of infected but completely asymptomatic people who test negative despite being infected. That's close to impossible for 3 different reasons, namely:

1. We aren't seeing breakouts that can't be chain traced.
2. Most close contacts of infected get infected; usually the whole family gets it
3. That's not how it works with other any other infectious agents that I know of (not a virologist).

If the population is doubling every 3 days, that makes the death rate far higher. If we assume a double every 3 days, you come out to a death rate of 12+%.

If we assume a doubling every week, you come out to a death rate of about 3-5% in Korea.

If we assume a doubling of every two weeks (impossible given the spread from Wuhan), then the death rate is around 1.5%.

You can't escape a >1% death rate no matter what you do.

Day 0 is today. I've got no interest in arguing numbers today. But let's come back to this post at the end of the month. If we're at around 200k-250k cases, what comes after is academic. And then you've got to worry.

i think i understand where i'm disagreeing with you now.

you're operating on an assumption that 5600 infections equals 5600 cases and that doesn't seem correct. they're only testing people that are showing symptoms afaik, and incubation rate is 10-14 days.

i just feel like its irresponsible to use this advanced past-projection dataset until you get a bigger sample on the full number of cases that have been in hospitals for 2 weeks (2 weeks incubation + 2 weeks of dealing with it).

Yeah in the US we'll never know with the fact that some people will be asymptomatic or have mild symptoms and never get tested.

I think South Korea where they are aggressively testing as many people as they can is the closest we're going to get to something close to reality.

Korea is testing everyone they can, >10,000/day. They're testing all 200K members of a sect for example, symptoms or not. That's population level sampling. And no, I'm using 2.5 weeks from symptoms to death, which is underestimating the time (it takes a while to go from sick to serious to critical to death). It's 3.5-6 weeks from first infection to death. I'm being conservative in all my numbers and am actually adding in underestimation biases in every number. I'm also being conservative in a week for doubling - it's probably less than 6 days. Every bias I'm introducing is strongly toward underestimating the death rate.

I'm not sure what your objection is. China has that sample if you need it. I've used at least four different streams of independent logic and data and they all lead to a similar result and >1% death. It's concrete at this point.

first case in korea was recorded jan 20, we're basically six weeks from that point.

so my issue with using that timeframe and establishing a meaningful dataset is... the cases that pop up the earliest tend to bottleneck in the death department. it's the same reason america has a very high death rate right now.

so if the strongest part of your samplesize is going 3.5-6 weeks back, i think it's going to overshoot death% since testing wasn't rigorous at the beginning (so you have a lot of cases that never got reported because they were mild flu symptoms) and the testing was primarily done on the older with pre-existing conditions.

i feel like you cant get a legit sample until you're hitting that 3.5-6 period with a sample of the population that's actually underdoing rigorous testing.

i'll leave it to other people to tell me if i'm crazy or not.

All of your objections are void by the fact that we have a population level sample. We're working back from the current level, exponentially, not working forward from zero. Your objections would be valid if we were using detections from 4 weeks ago. We're not.

We actually using the (biased to underestimation) notion that the current population level sample needs 2.5 weeks on average to die (a ridiculous underestimation which also underestimates death rate).

This is actually very straightforward. I'll let someone else explain it to you.

no America has a high death rate because they are only testing people that come in with symptoms and need treatment. Even then there have been instances where people were not tested because they didn't have the test.

ya america is the worst example but i'm operating under the assumption that a lot of countries, in the first week or two of dealing with it, had way too few tests and thus adopted a similar testing metric (only testing symptoms/critical)

It can also mean that virus becomes less deadly with time as the less deadly strains tend to be fitter than the more deadly ones.

Germans finally used the P word today. Could be a shift where governments don't want to be the last ones to get onside with this too. Could see a tone shift in the next week.

CDC is saying about a million testing kits out in a week or two and we should be able to test thousands a day by end of week. Expect confirmed case count to go up really fast.

Japan has tested 5,690 people (source is in Japanese but have confirmed it with a scientist in Japan). Population 126M. 45 tests per million people. Meaningless.

https://www.mhlw.go.jp/stf/newpage_09964.html

South Korea's early test numbers are insane, you really cant look at their data the same as other countries

There is no ****ing way the CFR is 0.1 to 0.5% Where the **** did they get that from ? It’s bullshit.

Retweet != endorsement, but saying the CFR at the end of 2021 will be 0.1 to 0.5% is quite different to saying the CFR thus far has been 0.1 to 0.5%.

But they are saying there is a lot of underreporting going on not that we will find a cure within a year.

It's almost as if this guy reads the outstanding analysis provided by posters in this thread.

Lol at world for not carefully analyzing Diamond princess data.

Diamond princess has 2,7 CFR% in closed cases judging by the available data. Much better than the 6% we are seeing in the rest of the world.

That is really good news if the data is accurate. It means that the actual CFR will end up around 1.5- 2% with undetermined old people bias because for some stupid reason we don’t have every relevant information from every single patient.

I had guessed 25 deaths in Diamond Princess. I think the under is clear favorite at this point 706 cases , 6 deaths , 212 recovered is my data.

With regards to Italy. Those deaths are following a similar pattern to the Chinese deaths.

The Chinese went from 106 to 232 in 3 days. The Italians are in 107.

Italy is on semi lock down. This is a huge natural experiment. They are going for aggressive control measures but not as strong as the Chinese. We will now know what happens.

Japan already apologized for botching Diamond Princess situation.

Website that tracks the mutations as they show up

https://nextstrain.org/ncov

cfr
3.4% from WHO
wuhan 1.4% from guan et al. (https://www.nejm.org/doi/full/10.1056/NEJMoa2002032)

SK cfr 34/5766= 0.59% https://www.worldometers.info/coronavirus/
italy cfr 107/3089= 3.46% https://www.worldometers.info/coronavirus/
diamond princess 6/706 = 0.84%
Japan 6/331 =1.8%

Once you factor in asymptomatic and light illness cases say 3x to 10x lower mortality, you're now at 0.3% to 1% total mortality rate. a 0.3% total mortality rate would be congruent with SK data and diamond princess data which I think we can all agree is cleanest and most precise.

0.3% is still 3x higher overall mortality than the flu. again, congruent with data available.


say 2 years before a vaccine:
7.5b people * 25% worldwide infection * 0.3% mortality rate = 5.625m deaths

worst case:

7.5b * 60% worldwide infection * 0.59 to 0.84% mortality rate (sk- diamond princess) =

26.6m to 37.8m deaths


this assumes generally top tier ability to provide treatment.

What adjustment are you making for lag-time from infection to death in the SK figures? ie. does this assume x% of currently infected people will die, or does it assume 100% recovery rate from anyone who is currently alive?