[jtsnow]

Ioannidis continues to push this evidence-based approach, but I have not yet seen an adequate response to critics such as Nassim Taleb or Yaneer Bar-Yam. They argue that one doesn't need raw evidence to act if the statistical evidence shows that the risks are catastrophic. They published a paper on January 26th which lays out arguments for why conventional risk management approaches are inadequate in these situations. [1]

This paper was met with disregard on HN [2], but the persistent reach of Ioannidis shows why Taleb's arguments have value.

Some on Twitter argue that the WHO statement on January 14th shows where Ioannidis' approach fails: "Preliminary investigations conducted by the Chinese authorities have found no clear evidence of human-to-human transmission..." [3] Authorities were looking for evidence before taking any action.

[1] https://necsi.edu/systemic-risk-of-pandemic-via-novel-pathog...

[2] https://news.ycombinator.com/item?id=22154333

[3] https://twitter.com/WHO/status/1217043229427761152

[avsteele]

I find Taleb's argument specious. The risk inherent in a reaction can be pretty catastrophic as well; for example we are now seeing some project 20%+ unemployment in the next quarter.

In addition, we do have data on the possible harms from Coronovirus, since we have more CFR data at this point. This puts bounds on possible harms. For more detail see John's other article in statnews. The results of the Diamond Princess cruise ship also are telling [1].

It looks like Iceland has about 1% of its people infected [2]. This didn't happen overnight, and Iceland is doing OK.

[1] https://twitter.com/maximlott/status/1241718453700038658?s=2...

[2] https://www.government.is/news/article/2020/03/15/Large-scal...

[rglullis]

If you are driving along the coast and you hear on the radio some report of a earthquake by the ocean, do you wait until you know how big the tsunami will be or do you just do your best to get the fuck out of there?

Also, how hard is it to understand that if Taleb's argument was turned into action on January 26th (57 days ago!) the number of people being impacted would be an much smaller fraction of what it is today?

Also, today we have a virus that is "only" killing 1% of its people, and you seem to be okay with it. What about the next ones? Do you think we should take a "wait-and-see" attitude for the next epidemic that might turn out to kill 5% of the infected? What if it turns out to kill 10%?

[yongjik]

Diamond Princess (pop. 3,711) was stationed right next to Japan (pop. 126M) and could use its full medical resource. That's not a good model for an actual country hit with the virus.

[natalyarostova]

Well, consider if we reacted late January when he wrote that article it would have been very cheap to contain...

[Barrin92]

Then the immediate question is, the next time we react early and it turns out we overreacted, is that something you're going to register as well? In particular in authoritarian countries like China which actually have a history of vast overreaction based on a 'security-first' mindset.

Ioannidis actually points to one other example in the article. There are several corona strains already in circulation with fatality rates as high as 8% among the elderly. If this reaction is rational, are we irrational not locking down everything every winter?

[natalyarostova]

Maybe we can just lock things down when we sport extremely dangerous new viruses. It doesn’t seem that tricky to me. I was following this from late December, and it was clear it was a potentially huge risk.

[pmachinery]

> It looks like Iceland has about 1% of its people infected.

Apologies if I've misunderstood something but your link says 473 cases, which is more like 0.1% of the population.

[avsteele]

fro the article:

"deCode has published the results of a total of 5 571 tests. Those have yielded 48 positive results (0.86%)"

'deCode' is the new series that will test everyone.

[pmachinery]

Sorry, I just saw this now. Misunderstanding I was. Thank you.

[ttul]

Iceland’s serology will be interesting to follow.

[guscost]

> I have not yet seen an adequate response to critics such as Nassim Taleb or Yaneer Bar-Yam.

Is that even necessary? Taleb's school argues that empiricism has fundamental shortcomings because it must be "incomplete" in a world of imperfect (read: statistical) evidence. I agree it's a big problem, one that we are not likely to solve. But they go on to say "therefore, we must apply the precautionary principle to XYZ" which is frankly nonsense. Instinct might be important in Taleb's world of non-ergodic black swans, but that absolutely does not prove that his instinct is better than mine or yours!

[acqq]

Taleb's point is that high uncertainties which can result in catastrophic loss are worth of spending money to protect against.

In our present case, the real nonsense is US allocating funds for a 1.5 trillion dollar plane, but totally avoiding investing in the response of pandemics... before it can be too late.

https://www.newyorker.com/news/news-desk/how-long-will-it-ta...

"science can meet the challenges, but there is lots of attrition” before any vaccine gets to the point of licensure. The problem is twofold. First, there may never be a market for a vaccine at the end of the development process, because the epidemic is contained, or never comes to pass. Then, traditionally, if there is an epidemic, it may take hold in a developing country where the costs of research and development cannot be recouped. “The resources and expertise sit in biotech and pharma, and they’ve got their business model,” Grant said. “They’re not charities. They can’t do this stuff for free."

Were there early enough proper funding (surely insignificant compared to 1.5 trillion dollar) this pandemic could have been avoided and, additionally, in the case it couldn't have, the vaccine produced faster. (Bill and Melinda Gates tried to motivate others to do something about that, for years).

Once the virus spreads, it is totally irrelevant where it started.

Edit: and to answer to the message below: I don't have to prove anything. The exponential spread will do its work, independently of all of us. That exponential spread is not something that happens just with coronavirus, it was for decades a known fact. That's the nature of pandemics. The humans in charge ignored the fact at the humanity's peril. There's nothing that can disappear because your political beliefs are different. Even more directly, we're where we are exactly because the political beliefs resulted in the ignorance of the facts.

[guscost]

Surely you must see that now we're having a moral/ethical/political/religious argument, very far removed from anything resembling science?

This is exactly my point. You may believe that you have identified all of "the real nonsense" perfectly well, and your suggestions might be better or worse than mine. But good luck becoming any more certain than you are right now, or proving your case to anyone else.

Put another way: If you believe that X is a "serious enough" risk and I do not, but I believe that Y is a "serious enough" risk and you do not, how do we resolve that disagreement? Who gets the money?

[paulryanrogers]

> Put another way: If you believe that X is a "serious enough" risk and I do not, but I believe that Y is a "serious enough" risk and you do not, how do we resolve that disagreement? Who gets the money?

Worst possible outcome for overreacting could be severe economic hardship and domestic unrest.

Worst possible outcome for under-reacting could be millions dead, leading to severe economic hardship and domestic unrest.

If the evidence isn't beyond all doubt I would hope the choice is clear which worst outcome is worth paying to avoid.

EDIT: "paying to avoid" not "paying for"

[guscost]

Let's not be naive here, the worst possible consequence of overreaction is a war.

[fcurts]

In that case, why wouldn't the same hold for underreaction?

[__s]

Pascal's Wager

[cma]

Non-ergodic means the virus isn’t randomly bouncing around with the same parameters on the way down. It has very different dynamics on the way down than on the way up, due to contact tracing and more, as seen in China and S Korea etc.

The UK was using a model that said after suppression it would just immediately rise, as if other measures wouldn’t be taken to hold it down.

[rumblecat]

Risk management is the correct way to go when uncertainty is high. Containment was the correct approach at the time.

When evidence starts coming in, then you can start applying evidence based approaches.

> Mortality rate: Mortality and morbidity rates are also downward biased, due to the lag between identified cases, deaths and reporting of those deaths [1]

> Among 3,711 Diamond Princess passengers and crew, 712 (19.2%) had positive test results for SARS-CoV-2 (Figure 1). Of these, 331 (46.5%) were asymptomatic at the time of testing. Among 381 symptomatic patients, 37 (9.7%) required intensive care, and nine (1.3%) died (8) . . . As of March 13, among 428 U.S. passengers and crew, 107 (25.0%) had positive test results for COVID-19; 11 U.S. passengers remain hospitalized in Japan (median age = 75 years), including seven in serious condition (median age = 76 years) [2].

Based on the second source, who can still seriously believe that the naive death rate is too conservative, because all the people in intensive care just have not died yet?

Look at the deaths/recoveries in Singapore and Hong Kong for more evidence [3][4].

Whereas, if you compare fatality rates reported by Germany, SK, HK, Singapore and other high testers vs China, Italy and Spain, it's pretty clear the latter are under-diagnosing mild/asymptomatic cases, which increase their fatality rate by a factor of 10 or more.

[1] https://necsi.edu/systemic-risk-of-pandemic-via-novel-pathog...

[2] https://www.cdc.gov/mmwr/volumes/69/wr/mm6912e3.htm?s_cid=mm...

[3] https://en.wikipedia.org/wiki/2020_coronavirus_pandemic_in_H...

[4] https://en.wikipedia.org/wiki/2020_coronavirus_pandemic_in_S...

[cameldrv]

Right, so do the math. 5% of positives required intensive care. If 197 million Americans get this, That means 10 million people go to intensive care. There are 60,000 ICU beds in the U.S. If 10 million people need the ICU, effectively none get a ventilator and they all die.

Now it's true that the cruise ship passengers skewed significantly older, but on the other hand, they were all ambulatory and healthy enough to be taking a cruise. There are populations that are at significantly higher risk than the cruise ship passengers.

Also, Chinese experience was that about half of the people admitted to the ICU eventually died.

[dvsfish]

>Now it's true that the cruise ship passengers skewed significantly older, but on the other hand, they were all ambulatory and healthy enough to be taking a cruise.

While yes this is true technically, I'm not sure the bar for "healthy enough" is as high as you're making out it is. In my experience (apologies for the anecdote), significantly obese people are quite capable of going on a cruise almost always.

[cameldrv]

I'm talking about people in nursing homes, people in hospitals, people on immunosuppressants, for example, after a transplant (who wouldn't go on a cruise because of norovirus etc), people with other immune diseases, etc. There are a lot of these people out there, and these people would be very hard hit if we just let COVID run through the population.

[guscost]

But on the other other hand, they were all traveling, eating cruise ship food, and probably drinking, all of which could weaken their immune systems. We can add speculative adjustments all day long, but there's no way we're going to get a randomized double-blind study out of it.

Also you can't conclude much of anything based on a linear extrapolation, even if you have good data.

[cameldrv]

What do you need a randomized double blind study for? You're not sure the people on the ship died of COVID?

As for adjustment factors, if you just adjusted for age, you'd get about 50% less mortality if the ship had the same age distribution as the country. So that's 5 million dead. However there are over a million people in the U.S. that are medically compromised and would have a very high fatality rate with COVID.

I also don't see what the problem with a linear extrapolation is.

Finally, I only accounted for deaths due to lack of ventilators. There also wouldn't be enough hospital beds, and that would lead to millions more deaths.

There is simply no reasonable alternative to suppressing the disease. We're talking more deaths than the Holocaust here.

[guscost]

> What do you need a randomized double blind study for? You're not sure the people on the ship died of COVID?

Er, you're not trying to figure out how the ship victims already died, you're trying to predict how many other people might die of the same cause. To do that kind of thing well, you need a hypothesis, and then you need to test it properly.

> As for adjustment factors, if you just adjusted for age, you'd get about 50% less mortality if the ship had the same age distribution as the country.

You can't "just adjust for age" or "just adjust for" anything, you're going to miss something! That's why people do clinical trials.

> I also don't see what the problem with a linear extrapolation is.

Basically, an epidemic is not a linear system, so you can't model it with linear functions. Look into the "SIR model" for a standard way to do that kind of thing. I'm not trained in this field so I'd look for a medical/science forum if you have questions.

https://mathworld.wolfram.com/SIRModel.html

[cameldrv]

> Er, you're not trying to figure out how the ship victims already died, you're trying to predict how many other people might die of the same cause. To do that kind of thing well, you need a hypothesis, and then you need to test it properly.

What would be the randomized double blind trial that you would run, and what information would it give us?

> Basically, an epidemic is not a linear system, so you can't model it with linear functions. Look into the "SIR model" for a standard way to do that kind of thing. I'm not trained in this field so I'd look for a medical/science forum if you have questions.

I'm familiar with the SIR model. What you'll find is that if R0>1, the SIR model converges to a state where S=1/R0, I=0, and R=1-1/R0. In this epidemic, R0 is approximately 2.5, of course depending on conditions. That means in the U.S. population, 60% will end in state R, which means 60% of people will get the virus. That's the 198 million number from above. It's actually a little worse than that because the SIR model doesn't have a "Dead" state, so more than 60% of the population has to get the virus in order for 60% of the end state population to have recovered.

[joe_the_user]

Whereas, if you compare fatality rates reported by Germany, SK, HK, Singapore and other high testers vs China, Italy and Spain, it's pretty clear the latter are under-diagnosing mild/asymptomatic cases, which increase their fatality rate by a factor of 10 or more.

South Korea has 1% fatality rate at the end of their epidemic, they showed .5% in the middle of this. Germany has .2% rate but it has crept up to .4% and I suspect it will continue to creep to 1%, and if they get overwhelmed it could go higher. China has a less than 1% rate outside of Wuhan, since outside that area, the health care system wasn't overwhelmed [1]. The extra deaths in Wuhan could be attributed to the health care system getting overwhelmed rather than under counting - 20 or 10% of those infected require intensive care. You quote 10% of the infected on the Diamond Princess as requiring hospitalization. With an overwhelmed health care system, that might be the death rate.

Which is to say that we have more evidence but that evidence seems to point to a desperate need for containment.

[1] You can compare all the statistics at: https://covid19info.live

[guscost]

> You quote 10% of the infected on the Diamond Princess as requiring hospitalization.

10% of symptomatic cases, not all cases, and definitely not all infected.

[ellyagg]

It's weird because I'm just as afraid of a black swan of unforeseen, runaway financial collapse due to pandemic countermeasures. People seem awfully glib about how many will die from the pursuant depression.

[rootusrootus]

I'm becoming considerably more afraid of the social and economic consequences from the pandemic countermeasures than of the virus itself.

[wwweston]

We had our chance to get this together years beforehand. We even had a chance to have lower impact interventions be effective 2-3 months ago.

There are no chances for light economic impacts now. Use lighter mitigation methods and productivity and demand will get hit by the illness and death itself, both the first order effects that hit those infected (which would be wider), and the second order effects as people improvise their own rightly fearful responses. The main difference you can count on is that the timing would move closer to the peak. AFAICT nobody has a clear model of whether it would be better or worse.

[ClumsyPilot]

You do realise that, given no countermeasures, there will be deaths of millions of people, plus half the country I'll. It will collapse the economy.

Tou cant save money by ignoring the disease.

[amiga_500]

My main fears:

1. They find it does effect kids, it just takes longer

2. They find reinfection is common

[rapidslowness]

Do not fear.

1. All evidence shows children have very low risk of severe illness. It can happen but is very very unlikely.

2. Immunity for corona viruses lasts months or years. Why should it be different for this one? Opposing news stories are mixtures of early discharges from hospitals and false negatives test results.

Studies that test also asymptotic cases from Italy and Heinsberg (Germany) (e.g. https://www.faz.net/aktuell/gesellschaft/gesundheit/coronavi...) indicate a R0 much higher than 2 or 3 (more between 5 and 10). We do not know infected people(!!!). We just know cases. My belief and hope is that virus is so infectious live can return to normal in weeks not months because almost everybody already has or had it (in areas with community spreading). Don't panic. Distance and wait.

[jhanschoo]

Taleb's note was probably more relevant back when it was published. It's clear that current measures are measures devised against epidemics, and not "conventional risk-management approaches" for everyday business, which he seems to be criticizing. There's a lot to criticize about the note [1] but its conclusion was appropriate in that a response tailored to the particular epidemic based on conventional risk-assessment can't properly evaluate the harm of an epidemic until it is too late.

But today, governments are responding in the typical fashion of responding to pandemics. Closing borders and quarantines are precisely the measure that dampens the more frequent so-called "superspreader" events, and social distancing is the exaggerated response that decimates the potential impact of these "superspreader" events.

It's obvious and unfortunate that the authors didn't do their research on public health policy, since public health advisors and experts actually already knows the right measures to take; it's a matter of convincing the decision-makers that this is the right way to respond to a new pandemic rather than the risk-assessment models that were tailored to more frequent events.

> Ioannidis' approach fails: "Preliminary investigations conducted by the Chinese authorities have found no clear evidence of human-to-human transmission..." [3] Authorities were looking for evidence before taking any action.

I don't think this is the right conclusion. I think many flu strains occasionally transmit from animal to humans, but fail to spread from human to human, and imposing drastic measures upon this is too much of an overreaction, since they aren't usually much worse in public health impact than the seasonal flu. On the other hand, as soon as there is a spike of cases in a local area, that is enough evidence to be wary. That tweet just reads like a poor excuse by Chinese officials that doesn't make sense, something that the rest of the world have come to expect from them.

You also have a misreading of what Ioannidis is trying to push for. He isn't advocating against taking action before good evidence comes out. Rather, he is highlighting that our current lack of good evidence about the epidemic is necessitating a greater reaction than may actually be necessary if we had better evidence. He is advocating for better evidence to be published with higher standards, so that if this pandemic is actually less dangerous than it actually is, decision-makers would continue to trust the public health experts to make decisions should a worse pandemic come about.

[1]

e.g.

> Standard individual-scale policy approaches such as isolation, contact tracing and monitoring are rapidly (computationally) overwhelmed

Computationally ???

[sdenton4]

/You also have a misreading of what Ioannidis is trying to push for. He isn't advocating against taking action before good evidence comes out. Rather, he is highlighting that our current lack of good evidence about the epidemic is necessitating a greater reaction than may actually be necessary if we had better evidence./

This is kinda the core problem, though: We don't have access to the full evidence (yet) and things already look somewhere between very very bad and mildly catastrophic. (I'll reserve full catastrophe for Giant Asteroid.) If you don't plan for the 'catastrophic' case and it's on the table, you look pretty bad if the error tends in that direction. By the time you KNOW you're in the catastrophic case, it's too late to deal with it.

I've seen the Ioniaddis pieces showing up in a couple places, and he really comes off as a bit of a crank, more concerned about his Stanford-supported stock portfolio than considering the Actually Available evidence. I don't give a fsck about the initial wrong reports in China... Italy's got overflowing hospitals and a very exponential-looking death curve right now. Not enough testing means we're getting better numbers, leading to a sharp spike... but the numbers are still reflecting mostly the worst cases, coz that's who tests are available to. And the numbers of deaths and very bad cases are climbing very, very fast.

Here's a model you can tweak, with plots of available data and estimates of available hospital beds, etc... Just looking at the death curves (can't hide a body, amirite) the 'Fast/North' scenario looks like it fits well for the US. Moderate-to-no mitigation is then modeled at O(3MM) deaths, and strong mitigation drops to O(1MM). So, the error bars that we're playing with are measured in millions of lives.

https://neherlab.org/covid19/

[jhanschoo]

Read that quote more closely. I'm of the impression that the article acknowledges that until better evidence comes in, "taking [drastic] action" is the right decision. But researchers should gather evidence quickly to see if the drastic action is an overreaction, and loosen measures once there is evidence that it is safe to do so.