Consider the shown picture where you are the decision maker who can pull the lever of the train tracks to avoid the coming train from going straight. If you do not divert the train, one person, John, will get run over. He is elderly and suffering from many diseases. You know him personally and all his friends and family are watching you. They are all shouting at you to divert the train, claiming it is the moral and safe thing to do. You know that if you do not pull the lever, your life in the society you live in is over.
If you pull the lever, the diverted train will run over 50 random people from all over the world as the train drives through them, including people in your own country. Yet these people and their friends won’t know where the train came from that hit them.
What do you do?
And more importantly, because it is obvious what anyone with a modicum of self-preservation would do, what institutions can you think of that would lead to a different choice?
Over the fold I enumerate why I think this is roughly the tradeoff that has faced humanity over the coronavirus, where the options represent letting the virus rage unchecked (the train drives streight) or put whole countries into isolation, destroying many international industries and thus affecting the livelihoods of billions, which through reduced government services and general prosperity will cost tens of millions of lives (the diverted train).
If you don’t like my back-of-the-envelope numbers please provide an honest alternative numerical assessment: anyone can quibble with numbers of others but it only becomes a discussion if you give a reasonable counter-estimate.
Many believe we are currently saving tens of millions of lives via our response relative to taking no drastic actions. When pressed, some say the fatality rate of “letting it rage” could be 5%, which is then used to say we are saving 200 million people. Some even claim that the deaths from coronavirus would seriously disrupt the workforce and thereby the economy.
Let’s take the last issue before we address the question of death rates.
The fact that deaths in the working age population from a pandemic can have effects on the economy is well-established. The Spanish Influenza that wiped out maybe 50 million people in 1918-1920 had a large effect on the world economy because it took out so many healthy people aged 18-40.
What do we know about the fatalities from the corona virus?
The Italian statistics just released are probably the most comprehensive to date. As a new report of March 17th (https://www.epicentro.iss.it/coronavirus/bollettino/Report-COVID-2019_17_marzo-v2.pdf) showed, some 99% of the younger cases of coronavirus deaths were among people with pre-existing serious conditions. Furthermore, the average age of those who died was about 80.
It should be clear that people aged 80 with other serious conditions are not a significant fraction of the workforce.
Then the death toll of the virus in a “do nothing” scenario. Some use the figure of 5% death rates as the high-end estimate, for which one could cite World Health Organisation numbers.
The 5% is just highly misleading. None of the modelling experts expects anything like that.
The 5% is the high-end estimate of the death rate among individuals diagnosed with the virus. Yet, in most places, including the UK, only the very serious cases get diagnosed at all. Moreover, there are “silent spreaders”: people running around who have or had the virus without knowing it, having very mild symptoms or none. That number is very hard to know as our only estimates are model derived, but for every diagnosed case there could easily be 5 or 10 people running around who have it and never will know they had it (unless randomly tested). Researchers are trying to find this number now by looking at how many people have antibodies rather than whether someone is ill.
So my own best-guess for the total fatality rate for the coronavirus should the whole population get low-level exposure is 0.2%. That corresponds to the death rate on the “Diamond Princess”, a cruise ship with over 3700 people on it that got quarantined in Japan for a month when cases were found. All on board were probably exposed, a bit over 700 on that ship were found to be infected, and only 8 died. This suggests only 1 in 5 have it long enough to be detected and even among those detected via frequent testing (which will give you a much higher number of cases than in one-off random testing) only a little over 1% dies. *
One might object that a cruiseship like the Diamond Princess wont have the most unhealthy people on it, but then one should also say its passengers will be much older on average. And it is likely all the passengers will have had a somewhat high exposure. In a “do little” scenario, most people in the population will not get a high exposure, just as in a regular flu season not everyone in the population gets exposed. Since every year 1-2% of the population die, 0.2% is basically the death toll of 2 months.
These reasons are also why the measured death rate differs so much between countries: in South Korea, where they did a lot of random testing and hence picked up more of the very mild cases, the death rate was 1%. The current death rate in Germany is only 0.3%, again probably due to huge random testing. In Italy and Spain, where only the more serious cases ever made it to a test, the rate is 5% or over. Other differences are the structure of the population, with more old individuals in Italy than in, say, Wuhan. And of course, it is the case that good medical care reduces fatalities, or should we say, postpones fatalities to a later date.
Indeed, it has also been reported that in Italy only 12% of the death certificates claim direct causation by coronavirus (rather than being “one of the causes”). So even in Italy on closer inspection the fatality rate among identified cases is probably not much higher than 1%.
This is probably also why the actual numbers of deaths are so low relative to other major diseases or natural causes. Every day in the world, 3,000 people die of traffic accidents and maybe half a million people die from old age and other causes. The worldwide death toll of corona is even now no higher than a thousand per day, or 0.2% of the usual human death toll per day.1 Even in Italy, the corona virus fatalities (when properly measured) are dwarfed by the 2,000 or so “normal deaths” per day.
Failing to put corona deaths in such a context is part of the cause of the panic. Not just newspapers are guilty of this, also many health advisers and professionals who are not keeping a cool head. One needs to give reasonable middle-of-the-road estimates for how many people would die. People who claim 200 million might die from the virus choose the route of hyperbole. Not helpful, not safe, and not based on evidence.
Even reasonable death tolls from the virus are misleading because of the fact that the corona virus deaths are so heavily concentrated among individuals with very little remaining life left. Like the UK health system, that openly adopts QALYs (Quality adjusted life years) as the thing it cares about, we should look at “whole lives lost”, or “years of life lost”.
You’d need a good model to do this properly, but I can make a reasoned guess. Since life expectancy in Italy is 83 and the average age of death is 80, a simple rule of thumb would have it those who died had 3 more years left on average. This is obviously not a proper calculation1, but I don’t think it will be far off given how even among the old, it’s the relatively unhealthy that died. Indeed, if we’d adjust for the quality of life, 3 more normal years is a generous assumption.
So with these basic numbers in mind, let’s work out the likely tradeoffs being made on the corona virus, where the scenarios would be “no containment response whatsoever” (let the train roll) versus the reaction we have had.
Even if no containment would mean there are 200 million deaths from this virus, that would cost the average world citizen 0.08 years of life, which is 600 million years of life lost divided by the world population of 7.7 billion people.
If the economic damage from the panic and disruption caused by taking this virus so seriously via economic self-isolation costs 50 million whole lives, the average of the world under that scenario loses 0.5 year of life (6 months). Under a more reasonable estimate of 0.2% total mortality rate of “do nothing”, we’d be talking about 15 million death, or 0.007 years of life of the average human.
So if I compare my best-guess estimates, then the loss of life of “do nothing” would have been 0.007 (2 and a half days per human) and the cost of our actual response equals 0.5 (6 months). That is a ratio of 1:70 in terms of length of life. Diverting the train to save John costs a whole village.
But even taking the wilder estimates around of the loss of do-nothing, the expected loss of life from the economic panic dwarfs the loss of life in the worst-case-scenario.
This really does raise the question what else we could have done and how to do things next time. I understand the extreme stress of doing nothing as the train hurtles towards a loved one, but for humanity as I whole I believe our panicked response has been disastrous. We have lost perspective on the damage we are doing whilst staring at something closer by.
So the issue is whether there was a middle way and what structures we need to make it politically feasible next time around to take that middle way.
2 Update 21/4 Of course more passengers died since, as you’d expect from a boat with 3700 old people, muddying the water on what can be learned with later data from that ship. However, there is more research coming in on the likely death rate if the whole population were exposed. A recent German paper put the infection-fatality death rate at 0.37% based on finding that for every positively tested person there were 7 times more with antibodies. Yet, given how the more frail are probably also more exposed (see the Ionnanidis video John Walker linked to in the comments), and that there seem to be people who are not “infectable” anyway (the very young), the 0.37% is still an overestimate to what would happen if the whole population got a low exposure. Relevantly, a recent Standford study puts the death rate between 0.12% and 0.2%, essentially based on antibody findings in random samples in the US. So 0.2% is still looking like the right middle-of-the-road estimate of the more contemplative studies, certainly if we think of the fatality rate in a “do little scenario”, which will be far lower than the infection fatality rate].
1 Yes, there is a difference between life expectancy at birth and conditional on being a certain age, so I am taking an extreme short-cut to a much more complicated modelling issue (I have estimated duration models on mortality and taught them to PhDs). Still, the implicit argument that those who died had another 3 years to live on average seems rather generous if you look at the Italian data. 3
- UPDATE 21/04: Andrew Briggs kindly provided a simple excel tool based on UK life tables to calculate the average number of lost QALYs if one knows the age and comorbidity distribution of the victims (which he doesnt quite have: he has the age distribution). He shows the QALY loss given the current age distribution of victims in the UK is between 2.25 (if there are strong co-morbidities) and 4.90 (few comorbidities). Given the high prevalence of co-morbidities, my initial eye-balled guess of 3 QALY loss per death is still looking reasonable.