Better Estimates of Covid19 death rates

What is the real death rate of Covid-19?

I want to estimate the per-age fatality rate of different countries: China, South Korea and Italy. South Korea has 72 deaths and 8086 cases for an overall CFR of 0.89%. Italy has 1441 deaths and 21157 cases for an overall CFR of 6.8% and China has 3189 deaths and 80824 cases for an overall CFR of 3.95%.

The countries, however, have a very different age breakdown of their infected populations and it would be good to compare CFRs by age

South Korea>

Data here . Note, that this is an under-estimate because there are still a lot of cases to resolve.

South Korea has other data sources, however it has data of 7869 cases and only 36 deaths (under-counting 30 deaths from 2 days ago), so it doesn’t look too reliable.

Italy:

Unfortunately, it’s tricky to find day-by-day age buckets from both deaths and cases, so we’ll need to estimate this through certain approximations. Italy has a case age break down here . However our best breakdown of deaths is a small sample. Note, the data in the above link is NOT the CFR, but a % of all deaths who fall into a given age range.
The particular small sample did not include deaths under 50, so we don’t have a good estimate for those. If we combine the two age breakdowns we get:

Note, this is very approximate because are combining age breakdowns from different days, however it’s likely in the right ballpark. Note, that these are once again under-estimates because many cases are yet to resolve and the death rate can go higher due to even worse hospital overload.

China

One of the frustrating issues with the per-age death statistics that got shared out of china is that they used data from February 11th, when the CFR was 2.3%. Right now, the CFR is 3.95%, so to get the estimates, we can scale everything by 3.95/2.3. Given how many recovered there are, this is very likely to be close to accurate eventual death rate. It’s hard to trust China’s number in general, but this is the best we have.

Once again, I suspect these are also under-estimates because China is likely to understate deaths rather than cases.

As suspected, the age break down explains a large portion of the difference between South Korea and Italy, but not all of it. With the same age distribution of infections, South Korea would have about a 3% death rate instead of 0.8%.
Controlling for age, Italy’s death rate is better than China, however given that we are comparing death rates early in the epidemic vs late, this is likely change both due to current cases resolving and hospital overload. We could try to compare the death rate to the infection rate 1 week ago, which would up each of the death rates in South Korea by about 13% and death rates in Italy by ….186% (factor of 2.86).

I am going to omit the under 50 estimate, because I just don’t have the data to do an upper bound here. Yes, this is 57% for 80+ people. Yes, I realize that number is EXREMELY high, however I don’t consider this to be that implausible given hospital overload and medical triage.

Even the upper bound for Italy is not that different from out China estimates for 50-70, which somewhat validates the data for China, at least in this age bucket. However, there is massive uncertainty for the 70-80 and 80+ age buckets. Clearly, Korea is still doing something right with the medical care and noticing the spread of the disease for those cases in addition to being somewhat “lucky” in that the disease has mainly hit less vulnerable people.

Now, given the last time I have published statistics a lot of people tried to make a lot of wrong arguments about why the death rates are lower than my estimates. Many of them took the form: there is a lot of people not being tested / tests have false negatives/ there are a lot of mild and asymptomatic cases.

Now, it’s very likely that South Korea is testing almost everybody with the virus and Italy is not and thus that can explain some of the difference in per-age death rates. However, I suspect that the people not being tested are generally those who don’t have symptoms or whose symptoms are not as serious. So, this is an argument for a lower death rate in younger people and a lower overall death rate, but I suspect that the % of older people who have no symptoms at all is low.

There are also false negatives in tests, however it’s likely that there are also false negatives in tests after death. If a test just randomly fails 30% of the time after being administered on the living or dead people, then everything just scales by an equivalent amount. Now it is possible that the tests fail less on dead people than on living people due to the likelihood of higher viral load in dead people. However, this also has to be counter-balanced by the fact that we are likely to re-test living people more.

I suspect there might be a knee – jerk confusion between “mild cases” that are tested vs cases that are not. When we talk about 80% being “mild” cases we are already looking at the statistics of people having been tested, not the estimates of people who the testing misses.

Long story short, the most likely missed cases are likely to be in the demographics where the death rate is already low.

Some people have raised questions about Diamond Princess as an example of a situation where an “older” population has a low death rate so far (~1%). This is an interesting case, but it’s important to remember that there is a massive difference for the 60-70 bucket and 80+ bucket. Cruise ship goers are on average older, but many fall into 50-70 bucket and less in the higher buckets. I don’t have enough data to do a full estimate on Diamond princess cases. Yes, it’s somewhat encouraging, but we now have a lot more data from Italy.

The overall point is that the death rates are likely higher than we think, especially in the 70+ age-range. Many people got anchored on the early CFR from China with an estimate of 2% and kept repeating it. Even if the West mounts a South-Korea – level response, which it will struggle to do, with an Italy age profile, we will see an overall death rate of at least 3%. In reality it likely be a lot higher(>10%) if all the ICUs are overloaded.

While flattening the curve is great and all, we should ideally also be striving for Total Victory over the disease. Total Victory implied Total War and Total Mobilization.

#staythefuckhome #flattenthecurve #canceleverything #flattenthebug