Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

who.int

Post reply on HN
Print view View original

341–350 of 400 posts

Re: Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

#341

Post by andbberger »

joe_the_user wrote: post #336
andbberger wrote: post #334

Earlier quoted context omitted.

The seroprevalence study was indeed awful, but this isn't so remarkable on its own. Bad science gets published all the time. The emails suggesting a foregone conclusion make Ioannidis look even worse. Nonetheless, the work stands on its own and should be evaluated as such. You would be forgiven for going in with the prior that Ioannidis has a poor track record in this area - but your prior alone is not enough to dism…

Science isn't a matter of simple data. Science is a matter of trust. A scientific cannot "stand on it's own" without the assurance that it was produced by someone with a minimal amount of integrity. There are too many elements in the process of research that can be "fudged" to allow this.

> Science is a matter of trust

Well that's simply not true, by definition.

That's like the whole point of science. You can do something and write a report on it, and if I don't believe you I am free to try and reproduce it and publish my own account.

Track record is an excellent reason to be skeptical. Skepticism alone is not enough to invalidate a study.

Re: Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

#342

Post by nobody9999 »

dx87 wrote: post #62
beervirus wrote: post #19

Earlier quoted context omitted.

> National health policy should not be based on personal outcomes. I'm not sure I understand what that even means. What should national health policy be based on?

What's best for the nation. Having blanket lockdowns with no timeline for lifting them, while also wrecking the economy because people don't want their 80+ grandma to die of covid isn't necessarily good for the country, even if it makes people feel better in the short term.

>What's best for the nation. Having blanket lockdowns with no timeline for lifting them, while also wrecking the economy because people don't want their 80+ grandma to die of covid isn't necessarily good for the country, even if it makes people feel better in the short term.

To where, specifically are you referring? I'm not incredibly well informed about policies outside where I live, but I'm not aware of any place in the US where every business is closed, everyone is required to stay indoors, all economic activity has ceased and police are using criminal penalties to enforce such behaviors.

The above is what you mean by a "blanket" lockdown, yes?

I'm not sure where you're talking about. I live in one of the most densely populated (27,000/sq mile) areas in the world (NYC), and we don't have (and haven't since June) had anything that could plausibly considered a "blanket lockdown."

Actually, even when NYC was experiencing the worst, and a "lockdown" was in place, some businesses were open and people were absolutely not required to stay in their homes.

Even so, the restrictions put in place were effective in reducing the exponential spread of COVID and allowed us to relax those restrictions.

In fact, except in a few areas with infection rates that are 5-8x surrounding areas, schools and almost all businesses are open (with some restrictions like indoor mask wearing/limiting the size of indoor gatherings).

There were pretty strong lockdown measures in the Mid March-early June timeframe, but those are long gone.

We are able to do this because we do a huge amount of testing and surveillance and have data-driven rules (>3% positive test results for seven consecutive days, for example) for addressing case clusters.

Lockdowns (like social distancing, wearing face coverings and improved hygiene) are just one facet of an appropriate response.

Large-scale testing, tracing and surveillance are required to ensure that infection clusters don't spread into larger populations.

No one wants lockdowns, but unless we utilize the other tools available to us, we will likely see R0 growing beyond our ability to control it.

And if infection rates skyrocket, lockdowns become the only way to minimize the spread of infection.

But making all that happen requires the cooperation of the vast majority of us. Where there are lockdowns, that's evidence of failure to execute on all the other mechanisms we have to combat this virus.

Re: Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

#343

Post by joe_the_user »

andbberger wrote: post #341
joe_the_user wrote: post #336

Earlier quoted context omitted.

Science isn't a matter of simple data. Science is a matter of trust. A scientific cannot "stand on it's own" without the assurance that it was produced by someone with a minimal amount of integrity. There are too many elements in the process of research that can be "fudged" to allow this.

> Science is a matter of trust Well that's simply not true, by definition. That's like the whole point of science. You can do something and write a report on it, and if I don't believe you I am free to try and reproduce it and publish my own account. Track record is an excellent reason to be skeptical. Skepticism alone is not enough to invalidate a study.

I should clarify science does require data, does descriptions and so-forth. But if scientists are free to fake data until caught and to essentially be untrustworthy, it becomes impossible to make progress.

That someone has engaged in bad faith and bad methology previously doesn't invalidate their findings. It doesn't prove they're wrong. But it makes people justified in ignoring them.

The world is full of, uh, bullshit, full of unjustified claims on this and that. These have to be ignored because otherwise you waste all your time. Being a credible scientists engaging credible research is a reason to take someone out of the this category and pay attention to them. But once someone has discredited themselves as a scientist, they're back in the bullshit category and no one has an obligation to look at their stuff. Sure, maybe their stuff is true, who knows.

Re: Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

#344

Post by nopzor »

Barrin92 wrote: post #330
mkolodny wrote: post #226

Earlier quoted context omitted.

My main concern with this study is that it uses reported Covid-19 deaths to infer the fatality rate. According to the paper, the fatality rates in the US are far higher than the rates in China and India. While the inferred fatality rate in the US is as high as ~1.3% (Louisiana), in many other places like China outside Wuhan and in India, the inferred death rate is close to 0.0%. I highly doubt that's actually the cas…

> in many other places like China outside Wuhan and in India, the inferred death rate is close to 0.0% it's very plausible that it's Singapore in particular is probably a high fidelity case. It has one of the highest testing rates on the globe, likely clean data and they've registered 28 deaths on >50k cases.

singapore has a special factor that you didnt mention.

the very large majority of their cases were in foreign worker dorms. almost none of these workers are over 40, and are generally in good health.

so the demographics are very different.

Re: Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

#345

Post by andbberger »

joe_the_user wrote: post #343
andbberger wrote: post #341

Earlier quoted context omitted.

> Science is a matter of trust Well that's simply not true, by definition. That's like the whole point of science. You can do something and write a report on it, and if I don't believe you I am free to try and reproduce it and publish my own account. Track record is an excellent reason to be skeptical. Skepticism alone is not enough to invalidate a study.

I should clarify science does require data, does descriptions and so-forth. But if scientists are free to fake data until caught and to essentially be untrustworthy, it becomes impossible to make progress. That someone has engaged in bad faith and bad methology previously doesn't invalidate their findings. It doesn't prove they're wrong. But it makes people justified in ignoring them. The world is full of, uh, bullsh…

Okay that's fair, I agree with that. No one has time to disprove quacks.

I do think that some mistakes should be recoverable from though. Outright data forgery should be career ending event full stop - but the burden of proof should be high.

Re: Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

#346

Post by bonzini »

timr wrote: post #229
bonzini wrote: post #188

Earlier quoted context omitted.

450 deaths a day was not the total, those were only those from confirmed COVID patients. Excess mortality was more like 650.

Citation required (I've looked, and all I can find are annual mortality rates per capita, which are higher, certainly, but not double the rate for a year). Even if you're right, a doubling of baseline mortality per day is certainly something plausible for influenza in specific scenarios. Italy as a whole is now seeing new cases per day roughly twice that observed in the spring, and yet deaths are up a tiny fraction o…

Source: https://www.corriere.it/dataroom-milena-gabanelli/covid-risc... (in Italian). 45000 excess people are estimated to have died in about 60 days, for an average of 750 per day so I was even remembering fewer than the actual value.

Testing was awful in the spring, serological surveys were made in June and estimated that only 15% roughly of the cases were caught and other surveys estimated even lower percentages (as low as 6%). The territorial distribution is also much more even this time, so it is easier to cope for the healthcare system.

My point is that any a priori estimate of the IFR falls apart if the healthcare system fails and the purpose of lockdown is to avoid that. You don't lock down because it's the only way to keep the IFR down; you lock down when you realize that tracing is failing to capture and/or isolate many cases, and therefore lockdown is the only remaining way to keep the IFR down.

Re: Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

#347

Post by makomk »

mrkurt wrote: post #264
makomk wrote: post #166

Earlier quoted context omitted.

One of the more... interesting things about Covid-19 is that the countries everyone thinks are stuck with indiscriminate lockdown as their only tool due to lack of testing capacity actually have the most testing capacity, and the one everyone points to as a mass testing success story has about an order of magnitude less in per capita terms and is likely severely limited in their ability to actually detect cases as a…

Do you have details? A quick glance here shows low positive test rates in the countries that seem to be doing the best. Low positive test rates seem like adequate testing. https://ourworldindata.org/coronavirus-testing

Low positive rates are an indicator that a country is having to do more testing to locate their cases than a country with a similar amount of cases, but higher positive rates. Nothing more, nothing less. At best, it indicates that the country has a relatively low infection rate (which is why countries that are doing better have low positive test rates), at worst it's because they're testing the wrong people.

Think about it this way. Currently, countries like the US or much of Europe tend to offer testing to anyone with mild potential symptoms that are caused by many common diseases. If they ever reach the point where Covid-19 cases make up such a substantial proportion of people with those generic symptoms as to affect the number of tests required, the country is in really deep, Lombardy-level trouble. The idea that the level of testing required depends on the size of the outbreak is nonsense created to continue the narrative that Western countries like the UK and US are still failing on testing and that's why we have so many cases well past the point it should've been put to bed. It made sense as a metric of testing aggressiveness in February when everyone was looking for cases linked to travel from China, since it tended to indicate how wide a net countries were casting when doing this, and a little bit back in March and April when many countries were only testing people with serious and obvious symptoms. Not so much these days.

(Note also that South Korea doesn't routinely offer testing to people with mild symptoms. Anyone can get tested if they pay out of pocket for it, but it's discouraged and at a tenth of the testing capacity of most Western nations I don't think they could handle many people demanding it. Which means they can't reliably detect cases not linked to ones they already know about, and of course those unlinked cases grow exponentially... Makes meaningful comparison of case figures and test positivity figures hard.)

Re: Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

#348

Post by zby »

noja wrote: post #85
rossdavidh wrote: post #2

For comparison, from what I've read the typical annual influenza fatality rate is around 0.1% (don't know what the under-70 rate would be). So, this puts covid-19 solidly in the uncanny valley of viral mortality rates; not low enough to be "just like a flu", not high enough to justify shutting down the world. No wonder opinions on it are so divided.

Apple and oranges. That flu fatality rate is without social distancing.

Social distancing reduces the number of infections - but it does not change the course disease once a patient is infected. So out of the infected (for IFR or out of the reported case for CFR) there should be the same fraction of fatalities with and without social distancing. Fatality rate is that fraction.

Re: Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

#349

Post by zby »

SomeoneFromCA wrote: post #97
rossdavidh wrote: post #2

For comparison, from what I've read the typical annual influenza fatality rate is around 0.1% (don't know what the under-70 rate would be). So, this puts covid-19 solidly in the uncanny valley of viral mortality rates; not low enough to be "just like a flu", not high enough to justify shutting down the world. No wonder opinions on it are so divided.

What you need to look is not fatality rate, but total fatalities caused by the disease. Flu kills 60000 a year max in US, Covid kills 5x that number. And it is unclear what will be a fatality rate for younger people whose organs are damaged by previous infection.

Actually you need to look at both. IFR shows you how dangerous the disease is once you contract it. The yearly fatality rates shows how much risk you have from the disease in a given year. Covid is much more contagious than flu - so there is bigger risk that you will contract it eventually, but once contracted it might be not much worse.

We still don't have comparable stats on that - even for flu the stats are really all over the place: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3809029 """ There is very substantial heterogeneity in published estimates of case fatality risk for H1N1pdm09, ranging from 10,000 per 100,000 infections (Figure 3). Large differences were associated with the choice of case definition (denominator). Because influenza virus infections are typically mild and self-limiting, and a substantial proportion of infections are subclinical and do not require medical attention, it is challenging to enumerate all symptomatic cases or infections.2, 45 In 2009, some of the earliest available information on fatality risk was provided by estimates based primarily on confirmed cases. However, because most H1N1pdm09 infections were not laboratory-confirmed, the estimates based on confirmed cases were up to 500 times higher than those based on symptomatic cases or infections (Figure 3). The consequent uncertainty about the case fatality risk and hence about the severity of H1N1pdm09 was problematic for risk assessment and risk communication during the period when many decisions about control and mitigation measures were being made. """

Re: Infection fatality rate of Covid-19 inferred from seroprevalence data [pdf]

#350

Post by timr »

bonzini wrote: post #346
timr wrote: post #229

Earlier quoted context omitted.

Citation required (I've looked, and all I can find are annual mortality rates per capita, which are higher, certainly, but not double the rate for a year). Even if you're right, a doubling of baseline mortality per day is certainly something plausible for influenza in specific scenarios. Italy as a whole is now seeing new cases per day roughly twice that observed in the spring, and yet deaths are up a tiny fraction o…

Source: https://www.corriere.it/dataroom-milena-gabanelli/covid-risc... (in Italian). 45000 excess people are estimated to have died in about 60 days, for an average of 750 per day so I was even remembering fewer than the actual value. Testing was awful in the spring, serological surveys were made in June and estimated that only 15% roughly of the cases were caught and other surveys estimated even lower percentages (…

OK, first, I need to say this: that news article has a number of false claims. Most notably, it claims that the fatality rate (# deceased / # infected) is 1-3%. Regardless of your opinions on the paper being discussed here, no credible source believes that the IFR for this virus is over 1%. That information is simply wrong.

That said, the claim for 45,000 excess deaths in March and April appears to come from this:

https://www.thelancet.com/journals/lancet/article/PIIS0140-6...

With this table having the details:

https://www.thelancet.com/cms/10.1016/S0140-6736(20)31865-1/...

The 45,000 number in that table is for all of Italy, whereas Lombardy specifically had excess mortality of 25,212 in March and April, with another ~700 in May. So that's 420 excess deaths a day in March/April, over a baseline of 275 (16,480 deaths in Lombardy, on average, for March and April of 2015-2019). This is nowhere near the 650 excess deaths per day you claimed in the GGP comment, but is a factor of about 2.5x over baseline.

For whatever it's worth, here's a paper that makes a claim of a much lower excess mortality figure of 5740 for Bergamo, and 3703 in Lombardy in the first four months of 2020, using better-controlled models for mortality in the regions:

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7520169/

I think it's somewhat pointless to debate the exact number of people dying every day, because we'll never know, and in any case, the virus was clearly quite deadly in that place at that time. However, both of these sources note that excess mortality spiked in March and April, and by May, had returned to below normal levels. So whatever happened in Lombardy, it was a statistical anomaly, and we should be careful extrapolating from it.

Did the virus cause significant excess mortality in Lombardy in March and April? Yes. Could the flu cause similar levels of excess mortality in a naive population? It can, and it has. The 1958 pandemic killed about 116,000 people in the US, which is well above the 12,000-60,000 people we see per year in modern times, and worse on a population-adjusted basis:

https://www.cdc.gov/flu/pandemic-resources/1957-1958-pandemi...

https://en.wikipedia.org/wiki/United_States_influenza_statis...

People like to make comparisons to the 1918 pandemic, but if anything, Covid-19 appears to be on par with the 1958 pandemic in terms of overall severity.

Post reply on HN