Covid-19 Points of Leverage, Travel Bans and Eradication
Covid-19 has become a major topic for discussion with talk about many different interventions and ideas that might help, from 3-D printing parts for respirators to drafting medical students into hospitals to thorough hand-washing procedures.
However as rationalists we should be asking which actions have the highest expected utility, not which actions have some positive utility. In an exponentially growing process, the actions with the highest expected utility are those actions which intervene early in the process, and actions like drafting medical students which intervene late in the process when the disease has already grown to a huge size are “nice to have” but by that point most of the damage has been done.
Proper and Prompt Travel Bans do Work
As early as January 26th, I called for cancellation of flights to limit the spread of covid-19; there was some pushback based on the idea that travel restrictions don’t work which upon closer examination was actually the idea that late or half-hearted travel restrictions don’t work:
During the height of the SARS outbreak in 2003, he had a colleague who wanted to return to the UK from Toronto, one of the cities most affected by the virus. So she caught a domestic flight from Toronto to Vancouver, then boarded a flight to London. “When she arrived at Heathrow [airport] and authorities asked her, ‘Have you been to Toronto,’ she said no and walked right through.”
A policy that allows people to travel from an infected area to to an uninfected area is not a travel ban. It’s containment theater. A real travel ban would be grounding all international flights and stopping passenger trains and boats until the disease had been eradicated or at least very well contained, as well as aggressively tracking down and contact tracing people who slipped through before the lockdown, for example using cellphone data from intelligence agencies. A key point here is that mopping up a small number of cases that slip through is in fact possible.
It would have been expensive to do all this, but the cost of not doing it is that the developed world is now on lockdown, the stock markets have fallen by around 33% and we have about 10,000 deaths at the time of writing. And we have ended up implementing the travel bans anyway!
#Don’tFlattenTheCurve
The optimal strategy to defeat the disease is currently the subject of much debate. Several strategies have emerged, and a popular meme right now is #flattenthecurve. The idea of flattening the curve is that if we increase the duration of the pandemic, the number of people infected at any one time will be lower and our ability to treat people properly will be increased. People put a lot of time into creating convincing memes and diagrams showing how this works:
Unfortunately people didn’t put much effort into getting the numbers right. Every single one of these diagrams is a steaming pile of nonsense because the line for “Healthcare System Capacity” is about 20-50 times too high, which was first pointed out by Joshua Bach. That tiny red line right next to the x-axis is our health system capacity:
(taken from The Imperial College COVID-19 Response Team’s latest report ).
The UK government’s “herd immunity” strategy was another possible way forward, but the government reversed course on this when they realized it would involve at least a few hundred thousand deaths.
Contain and Eradicate
In my opinion, the correct strategy to beat covid-19 whilst minimizing losses from this point forward is a contain-and-eradicate strategy. The New England Complex Systems Institute’s writeup on this, written by Nassim Nicholas Taleb of Black Swan fame outlines the strategy:
Since lockdowns result in exponentially decreasing numbers of cases, a comparatively short amount of time can be sufficient to achieve pathogen extinction, after which relaxing restrictions can be done without resurgence. …
Finally, the use of geographic boundaries and travel restrictions allows for effective and comparatively low cost imposition and relaxation of interventions. Such a multi-scale approach accelerates response efforts, reduces social impacts, allows for relaxing restrictions in areas earlier that are less affected, enables uninfected areas to assist in response in the ares that are infected, and is a much more practical and effective way to stop otherwise devastating outbreaks. …
A few other issues are of importance: They ignore the possibility of superspreader events in gatherings by not including the fat tail distribution of contagion in their model. This leads them to deny the importance of banning them, which has been shown to be incorrect, including in South Korea. Cutting the fat tail of the infection distribution is critical to reducing R0.
Basically:
- Close borders and limit internal travel, lockdown and hygiene to drive R0 below 1
- Ban large events to cut off the long tail of the R0 distribution
- Use aggressive testing and contact tracing to clean up any remaining holdouts, and eradicate the virus on a region-by-region and country-by-country level.
- “Green” regions can return to mostly normal life, albeit without large events and travel. That means that people can go back to work and we can reverse the economic damage.
Contain-and-eradicate probably results in both less loss of life and less economic damage than any other strategy, and we can see this as a consequence of taking an exponential process and fighting it in the low orders of magnitude rather than the high ones. Flatten-The-Curve is bad because a flat curve that lasts for a long time is still, in log-terms, almost at the maximum power of the virus and therefore it can do huge amounts of damage. Herd-Immunity and Deliberate-Infection are bad for the same reason. The only other sensible plan I have seen is the idea of rushing a vaccine as quickly as possible, but that is beyond my expertise.
Travel bans and restrictions during a pandemic
Why do borders need to be closed now when the virus is already everywhere? Because there is still uncertainty about where the virus is and in what numbers. The virus wins when people with different amounts of virus mix, because areas of high virus can spread to areas of low virus whilst the reverse process doesn’t work.
Similarly, if you were certain about who had the virus, this would almost be trivial because all the infected could be moved to containment facilities and everyone else could get on with running the economy.
The virus wants to maximize entropy (virus spread everywhere), humanity wants to minimize it (all virus in one place), for a given total amount of virus.
The need for borders is a result of this combination of uncertainty and mixing being bad. And as Taleb points out in the NECSI review, travel bans and restrictions during a pandemic should be multilevel.
As we approach the “endgame” where testing is ubiquitous and virus numbers get closer to 0, borders become more important, because adding 500 cases to an area with 1 case is much worse than adding 2000 cases to an area with 1000 cases (you have to think in logarithms).
Though even when numbers are high, closing borders is still useful and we should still do it; principally because the decision to close can lag behind rapidly developing facts on the ground, or worse the decision to close borders to a particular area could leak, at which point people start actively helping the virus to spread as they flee from the soon-to-be locked down area. Of course nobody would be stupid enough to leak that information, right?
What if we do Mitigation instead?
If we do go down the mitigation path—letting most people get the disease—there are some important “dice rolls” that will determine how it goes:
The rate of long-term complications amongst covid-19 survivors,
The rate at which young & otherwise healthy people die when hospital treatment is denied due to overcrowding
Whether new pharmaceuticals like Chloroquine and Remdesivir are both effective and scalable, and how quickly covid-19 evolves resistance to them
Whether summer weather substantially slows the spread
Whether covid-19 picks up a mutation that makes it less lethal, or more lethal
A Test of Rationality
When competent Muggles make decisions, they’re usually very empirical about it. They build a chair with one leg, it falls over, and then they don’t do that again.
Covid-19′s exponential dynamics, asymptomatic carriers and long lag time between infection and death punished the try-it-and-see approach very hard.
By the time it became absolutely obvious to people who build one-legged chairs that this was a big deal and needed attention, the virus had increased both its numbers and distribution most of the way to its goal of infecting every human being on the planet.
Covid-19 was a rationality test as well as a competence test. China failed on rationality but passed on competence. The West failed hard on rationality and is on course for a F+ on competence as well. Vox and the other mainstream media who either mocked those who took it seriously early, or got on a soapbox talking about racism (which is bad, but was not even remotely the most important thing at that time) should take a reputational hit. The various government agencies that dithered throughout February should be investigated, particularly in the USA.
To start, the severity estimates that Joshua assumed were worst case and are implausible. The very alarmist Fergeson et al paper has much lower numbers than Joshua’s claim that “20% will develop a severe case and need medical support to survive.”
I also think you’re wrong about the likely course of the disease, for a couple reasons.
First, as the overload gets worse, therapeutic drugs will become more widely used. I expect that at least a few of the candidates will be at least moderately effective in treating cases, and even though we’ll run out of Remdesivir quickly, production will be ramping up. Chloroquine will be made available widely as well.
Second, R_0 will drop significantly with the community distancing / flattening the curve measures. The line in your diagram is typical capacity—but if the spread is slowed enough to bring extra ventilators and emergency response capacity online, the situation is much less disasterous. Yes, it will be bad, but the worse it is, and the more news coverage there is, the more distancing will happen on its own.
Third, the seasonal component is very uncertain, but is almost certainly non-zero. If spread is slowed due to distancing, R_0 could certainly drop below one by the time the health system is getting overloaded.
For all of those reasons, I think your prescription is alarmist. Good Judgement’s Dashboard has less than a 20% chance of over 350k deaths—that’s a 0.1% population fataility rate. (Full disclosure: I’m forecasting for it, but am currently less pessimistic than the average.)
I’ll address my claims about why not to call for bans or eradication yet in another comment.
I believe the 20% figure comes from the WHO joint report which says
There are a lot of modeling assumptions that go into this, and the true number is probably lower, but not so low as to invalidate Joscha’s point.
In that report, 13.8% had at least one of those symptoms—that doesn’t imply than many or most would require ICU support to survive.
And event if we assume they would all die, which is wrong, Wuhan was an unlikeley-to-be-repeated worst case scenario—not just because of the medical overload with no warning, or the significant under-diagnosis of lightly symptomatic younger patients inflating the severe case percentage, but because they didn’t realize this was a severe disease for the elderly until at least weeks into the spread. Elderly people globally are now being kept largely isolated, and will be treated aggressively when they get sick initially, instead of treating it like influenza until they are nearly dead.
If Wuhan was a worst case scenario, how come Italy has now overtaken it in deaths? (3400 Vs 3200)
The population of Italy is several times higher, and the death rate per case is still significantly lower.
The population of the Lombardy region is slightly smaller than that of Wuhan, and they are reporting 3x as many daily deahts as China did during their peak.
It’s changed substantially since you wrote this BTW. It’s now 54%
Right, meaning that the population fatality rate looks like it will end up close to 0.1%, so saying 20% would need medical support to survive is incredibly alarmist.
Thanks for your comment, David.
The diagram with the red line is from Ferguson et al, and is labelled “Surge critical care bed capacity” in their paper.
Why? If we follow a “Flatten the Curve” approach we will take significant casualties as well as significant economic damage. Flatten the curve in Ferguson et al assumes that lockdown lasts until September. The strategy that I endorse is to zero out the curve and eradicate the disease, and I think that some people could be back at work in 6 weeks if we do that, albeit with certain restrictions on travel and events. I would have to look at this a bit more, but it seems to me that an eradication plan will be much less economically costly.
Yes, GJP is predicting 326,000 dead in expectation in the US alone. But I think a large input to that is people rationally expect that if it gets that bad there will be a massive lockdown that lasts for a long time. Lockdowns and deaths are two different costs we are paying, we want to minimize the sum of human plus economic costs.
Yes, the diagram is based on waiting another full month—April 20 - before starting any interventions to reduce this, and assuming an Rp of 2.2 or 2.4 until then. That’s not happening, because they’ve already started much of the proposed interventions, and given that, the curve will already be far lower.
And “flatten the curve” can and will be used as an interim strategy if it is ineffective—this paper assumes that they would have 6,000 people in ICUs before anyone starts asking whether they should start more.
True, these things will limit the damage in lives if all goes well.
But I think there is a bit of a strategy bifurcation here: if the drugs allow the damage from covid-19 to be minimized to such an extent that we’re happy with the “flatten” strategy—meaning a partial lockdown that helps to slow the virus down but still lets it go through the whole population, then at that point we should probably take some extra casualties and just allow the virus to go through the population as quickly as possible, thereby avoiding months and months of costly lockdowns.
If we are not happy with doing that, we should probably go for eradication.
I don’t see a middle ground where a rational decision-maker would want to run the “flatten” strategy, though we may muddle through to there through indecision and incompetence.
Yes, I’ve noted elsewhere that treatment options might make increased spread more likely, - and it’s unclear that this will be net positive in fact, because as you say, decision-makers will muddle through, and use the existence of treatment as an excuse not to limit spread enough, potentially increasing total deaths despite partially effective treatment.
But your claim that we want to ” just allow the virus to go through the population as quickly as possible ” seems wrong. Imagine (very generously,) that the available treatments reduce the percentage of critical cases by 80%. That means that health care systems can stay under capacity with a flatten strategy, but not with your suggested strategy. For “as quickly as possible” to make sense, we’d need a 95% effective treatment—which is implausible to the point of impossibility with the types of drugs currently being considered.
If the drugs save 80% of critically ill people from dying, then even if the remaining 20% overload the hospital system it might still be worth not trying to flatten the curve, just to avoid the economic damage from the lockdowns.
One would have to do a detailed analysis, but right now I seem to be getting the impression that the drugs actually aren’t that good.
Anyway I still stand behind the point that there is likely to be a strategy bifurcation where it’s best to either go pretty all-in on containment or go pretty all-in on herd immunity/deliberate infection, depending on just how bad it would be.
A 0.1% chance of death is worth about 1-3 months of lockdown, but age-weighting of deaths towards older people and lockdown damage skews this.
Without any change in the numbers I definitely still support containment.
It looks like widespread border closures are inevitable now, and border policy will become even more visibly important if/when community transmission is brought under control in a country (e.g. as in China today where ~100% of new cases outside Hubei are imported). So I don’t think advocating for border closures is high leverage at the moment.
I agree that it’s super high leverage to get the public and policymakers to understand that it’s not too late for eradication (R0 < 1) through strong social distancing, and that it may be feasible to keep secondary epidemics controlled at social cost far below that of continued lockdown. [ETA: as far as I can tell there is near consensus on this point among vocal rationalists on Coronavirus Twitter, but I have seen no public official or advisor state or signal that they are looking in this direction at all.]
An important part of that will be running the numbers on something like Taleb’s proposal or your “Basically” paragraph. Ferguson et al. got quantitative results from their model & set of assumptions, so far the response has been mostly handwaving and pointing at case studies (where we have 2 months of data and are making claims around sustainable policy on 1-2 year scale).
No, I agree that it’s not super-high leverage, but still worth saying. I’m just emphasizing that I called for it back in January when it was super high-leverage.
I would love to do this, but someone will have to pay me because I don’t have loads of time/money to spare. Alternatively someone else, perhaps a professional, will do this. Ideally they should already be doing it.
In general, I think calling for interventions that would work but aren’t politically feasible is low value, and mostly about signalling. This is made worse by the fact that the current projection aren’t catastrophic, just very bad—but even in the worst case, it’s a waste of time.
For example, as you suggested, we could have called for groundings on February 1, and if super-strict, it could have been mostly successful—but wouldn’t have been enough. If we had banned all air travel on Feb 1, we’d still have had community transmission that had started earlier than that.
But let’s say we did it. Everyone involved would be looking for a new job by February 3rd, and the decision would have been reversed—and the people knew it. Perhaps we’d now be more upset about the reversal, but that wouldn’t have made it work, and you would of course have many people blaming the initial overreaction for why the containment failed. So I think Vox called this exactly right—you can’t implement these measures early enough, even if in the counterfactual world where people did try, and even if in that counterfactual world it would work. And as I said at the time, I didn’t think it was going to work in practice.
BUT I think that calling for eradication in the US now. We should have gone for suppression earlier, and let the CDC tell seniors not to fly, etc. But it’s unclear we could manage eradication at this point, with the spread where it is—and calling for it is a waste of our time. But don’t worry, they’ll call for more drastic measures in another 2 weeks anyways, even though it’s already too late. And then you can say you told them so. At this point, arresting everyone who has an event with more than 10 people is arrested and everyone there is fined heavily, which I think is the right strategy everywhere that can manage it—isn’t feasible in a country like the US or UK. This is for the same reason I thought banning flights on Feb 1 would be a bad idea. I don’t think the population will listen, COVID is widespread already, and authorities aren’t willing to do something so unpopular.
NOTE: I’ll likely be writing a post-mortem of my reactions and thoughts in a couple months. I was wrong to think the government was starting to handle it decently, or that they would get their act together quickly enough—I wasn’t pessimistic enough about how badly the current US administration screwed things up, or how long it would take them to let public health people actually take over managing the response—I’ve stopped hoping they will start doing that at all, despite the fact that it’s insane they haven’t.
This is a misuse of the word can’t. People can, but people don’t want to.
If I don’t say what the correct answer is because I and others believe that people won’t listen, then I’m not doing my job as a rationalist. My top loyalty is to the truth, even if I am 99.99% sure people won’t listen.
That might be the main point of disagreement—I’m much more interested in effective altruism in pandemic preparedness than it making true claims that are irrelevant to decision-making.
I think these true claims are highly relevant even if there is a very high chance that none of the authorities will follow them.
Covid-19 is a comparatively mild test of humanity’s capacity to fight dangerous diseases. It’s not the “real thing”, the disease X that could kill hundreds of millions of people or bring our civilization crashing down.
As such I think it’s very important for rationalists to build up a track record of making the right calls.
My only regret is that I didn’t express a strong opinion even earlier in January.
I agree strongly with this sentiment.
How much of this changes if you choose to use smaller units than the entire United States? Right now, the North East is really bad, but the Mid-west is in great shape. What about even smaller, NY is in a dire situation, but Maine is fine. If you zoom in even more, NYC is a train wreck, but with the exception of two counties, upstate and western NY are fine.
If a travel ban in/out of NYC were strictly enforced, would they emerge from this in a few weeks with effective herd immunity and the virus unable to spread further there?
All the analysis I read keeps assuming the US is the base unit when calculating total infected or ICU capacity. I think it would change the calculations significantly if you add a time distribution, where it would be possible for places without rapid spread like Maine, Montana, and other low case areas were to send Medical personnel and equipment to NYC for a few weeks, then as the NYC cases pass the peak, those resources move to the next hot spot. Does this reduced population (just NYC), combined with more ICU/Ventilators, keep it below the healthcare system is overrun (much better ratio of ICU availability per person)?
What is the reason to ban travel between two already similarly affected countries, like Spain and Italy, except the fact that people will be infected during the travel (in planes)?
1. The decision to close can lag behind rapidly developing facts on the ground, or worse the decision to close borders to a particular area could leak. This actually happened in Italy.
2. Mixing + uncertainty is bad. Even if you think the rates are equal, they might actually not be.
3. At this stage everyone should be on local lockdown, so for most people there is no benefit to traveling anyway.
One more idea: “Synchronised pulsing closing everywhere”.
Simultaneous closing of everywhere in the world is more important than desynchronised closing of everything. If all the world will close for 6 weeks simultaneously, it will kill the virus wave without destroying the economy. The world then will open again for 2 months and then may close again, if a new wave appears. Such pulsing closing could control the epidemic, but will be predictable for economic activity and could happen several times until vaccine (and ubiquitous UV) appear.
Even across a single country there will be variation in the effectiveness of local measures. When you have this dispersion in R0 what happens is most places go to 0 like you hoped, but some places don’t.
Since that dispersion is unavoidable, you need a fractal system of borders—strong borders around countries, medium borders around states and cities, and weaker social distancing measures locally. Then the problem areas don’t spread much, and they can be defeated in detail.
Look at the map of South korea after the superspreading event: https://imgur.com/l7RD345
98 cases in Seoul (pop: 10 million) and frickin’ 4000 in Daegu (pop 2 million).
Joscha’s article seems to overcorrect in the other direction. Messing with ad hoc models I get a mitigated peak 3-5x hospital capacity, not 20x+.
It doesn’t matter hugely whether hospitals are overloaded by 3-5x for a long time or 20x for a relatively short time. In the former about 75% of people are unable to get treatment, in the latter 95%. 3-5x is better but isn’t as much better as it might seem.
The UK government has claimed that hygiene/social distancing can flatten the curve by 20% but 20% doesn’t make much difference unless you happen to be just over capacity beforehand.
Oh, yeah don’t know how the differences would play out down stream in the model or in real life. Just thought it was worth pointing out since 3-5x vs 20x+ is very large and thus it’s worth investigating the model to see what’s causing that difference.
The Ferguson et al paper on page 8 shows a lot more than a 3x to 5x hospital overcapacity. It’s something like 30x for uncontrolled and 10x for their maximum mitigation plan.
https://www.imperial.ac.uk/media/imperial-college/medicine/sph/ide/gida-fellowships/Imperial-College-COVID19-NPI-modelling-16-03-2020.pdf
If this is true then we should see another spike in Hubei as they lift restrictions right?
No, because the Chinese are being smart and driving the disease to 0. That’s whey they dismantled their temporary hospitals.
Why would you want to ban travel indiscriminately once testing has become ubiquitous? You can instead bar entry only to the tiny minority of travelers who test positive.
Because people don’t test positive immediately after getting infected but only after the virus reproduces a bunch of times.
What is the lag between infection and feasible detection? Without knowing the answer to this question, I’m skeptical this consideration should suffice to justify indiscriminate travel bans. South Korea has largely contained the outbreak mostly by extensive testing and isolation, and without imposing significant travel bans. And we are assuming a scenario where tests are even more widespread, and deliver results more quickly, than currently in South Korea.
It depends on the test.
If you had an instant test with a very low false negative rate, and in quantities such that you could test every traveler, then there would be no reason to ban travel because you have created a situation where the virus cannot move, but using tests instead of banning human travel.
My impression is that you could do something almost as good with well-managed quarantine that lasted long enough for you to be sure about the test results.
Most travel is not worth the hassle of a 7-14-day quarantine though
One thing regarding the herd immunity response is that it seems to me to become a lot more plausible if you could direct who gets infected and becomes immune.
I did the numbers on the population of Sweden, combining our age demographics with the estimates from the Imperial College report.
With a population of around 10 million, if everyone here got infected, 831 000 patients would require hospitalization and 281 000 patients would require intensive care, which would completely unmanagable. However if only everyone below the age of 40 got infected, 65 000 patients would require hospitalization and only around 3 200 would require intensive care. And that constitutes pretty much exactly 50% of our population. So if it was possible to keep the age of the infected down substantially, it seems possible for a large part of our population to get immunity without completely overburdening the health care system.
So the downside of this approach is that if everyone below 40 gets the disease you are entering a situation where the disease is all over the place, and it will be hard to keep the 40+ people fully safe. For example many older people need to be _in a hospital_ for various reasons, which is exactly where the virus is.
In addition you might be inflicting death and long-term disability on quite a lot of those under-40-year-olds.
And I think at the end of it, under 40s immune might not even be enough for herd immunity. You need something like 80% I think.
What if we assume that the COVID pandemic was the intended result?
The healthcare system capacity shouldn’t be a flat line, though I admit that the reports I’ve seen suggest that not nearly enough effort has been devoted to ramping up to deal with the emergency. But obviously if there is an upward slope to capacity (and there are efforts to increase production of ventilators, to pick one of the most troublesome restrictions), that increases the benefit of curve flattening efforts.
There could be a downward slope as healthcare workers get sick and various resources run out.
I’d put money on healthcare capacity being at an increased level throughout the pandemic, if you can figure out how to implement this as a concrete prediction. Perhaps total ICU-equivalent beds available?
Can we look at what happened in Wuhan, but subtract away any outside help they got?
Wuhan got outside help but they were less prepared. There’s some learning about how to treat cases going on as time passes. China manages to ramp up mask production.
But I mean do we know what fraction of Wuhan Healthcare Workers were rendered useless by exhaustion and/or disease? It looked pretty bad, and I imagine it would have been much worse if the *whole* of China was going through the same thing as there would have been no outside help.
It’s almost a shame that some country isn’t doing mitigation because I would like to see just how much of a clusterf**k is turns into and how optimistic assumptions are crushed by reality. Of course in reality I don’t want that because I think it would be very bad
Western countries seem to be all doing mitigation but we will see if a country like Afghanistan manages to do it. Will the Taliban who just made their deal to have their land back accept the knowledge of outsiders, that it’s important to do mitigation?
Are there any other organizations that should get a reputation boost from this event? I’m thinking about organizations like HEB supermarket and USPS.
I rag on the California government a lot, but big positive update on their early response. Some will want to credit the big tech campuses for closing fast, but this is more a function of how cheap it was for them to do so relative to other businesses.