May 24, 2020
Impudent Thoughts on COVID deaths
Turns out that (a) lockdowns work great, and (b) ‘way more people have died of COVID-19 than we thought.
When someone dies, the government is supposed to mark down how they died — what’s called the proximate cause of death. This ignores all the other factors that may have contributed to death. However, when a death is listed as “COVID-related”, it can mean either (1) a death of someone who had COVID-19, or (2) a listing of “COVID” on a death certificate, or (3)something else entirely. There is unfortunately no international consensus, so it makes big data analysis even harder than it otherwise would be. (I had no luck finding information on how different countries define “COVID-related death” except for Italy.)
Because of this and other problems around defining what death by COVID actually means, politically minded people are having a great time lobbing rhetorical bombs at one another. This is a horribly stupid thing to do because it’s only a distraction while people keep dying unnecessarily.
There is, however, an interesting way to get around the ambiguity of determining whether a death is COVID-related or not, and it comes from the actuarial sciences.
Yeah, I know: I hate the insurance industry as much as the next person. But their morbid fascination with death statistics actually turns out to be very useful in thinking about just how bad the COVIDocalypse is. The technique is called “excess death analysis”; it’s a statistical instrument that gives a sense of current death rates with respect to normalized data from previous years. A nice summary of the technique is available at Our World In Data.
Let’s start with an article in the Financial Times (26 April). Here’s a key figure from that article:
Those red spikes show the excess death rates in March and April compared to average values. The black lines show the historical averages; the grey lines show data from recent, pre-COVID years. Except for Denmark, it’s quite evident that there’s a very strong correlation between outbreaks of COVID-19 and huge increases in overall death rates. Indeed, the increase is even worse in urban areas (you can see that chart yourself here).
You’ll also see that death rates are already decreasing, and you might infer from that that the worst is over.
But you’d be wrong.
To understand why it’s not over yet — not by a long shot — consider the excess death data charted at The Economist.
Here’s the chart for Italy:
There are three important points of interest here.
COVID-attributed deaths vs overall excess death. The black dotted line shows the historical average death rate. The beige area shows non-COVID deaths in 2020. The orange area shows only deaths attributed to COVID-19. The orange must sit atop the beige to show the total (COVID + non-COVID) death rate.
The scary part is that the non-COVID death rate also spiked. This means that either (a) something else is killing people in tight correspondence to COVID deaths, or (b) we’re significantly under-reporting the number of COVID deaths. Considering how large the non-COVID spike is, and how nicely it correlates with the COVID death rate, I really can’t accept option (a). To me, there’s no question that we’re underreporting COVID deaths.
And if you check that site yourself for other countries, you’ll see pretty much the same story all around the world.
Decreased death rate follows lockdowns. The Italian national lockdown went into effect on 9 March. At that time, there was already a significant increase in both COVID-related and non-COVID-related deaths. Lockdowns are not instantaneous; they will take time — at least a few days — to get into place and enforce. COVID’s incubation period is 1–14 days with a median of about 5 days.
And sure enough, death rates start to decrease a couple of weeks after lockdown just as one would expect.
You can check The Economist’s charts of other countries for yourself: death rates decrease between 1–3 weeks after instituting national lockdowns.
There’s an obvious conclusion: lockdowns work.
And to prove it, one might hypothesize that the earlier a lockdown is instituted, the less of a spike you’ll get. Guess what? That plays too.
The sooner you lock down, the better. This is particularly evident in the Italian data for the three regions shown below the main graph: Lombardy already had a significant spike when the lockdown went into effect, but Piemonte and Emilia-Romagna were essentially still unaffected on 9 March. The result is that the latter two regions suffered far fewer deaths, and a much shallower spike, than Lombardy did.
This clearly shows that the best way to prevent unnecessary deaths is to lock down before you see a spike. That would have been impossible in Wuhan, where COVID-19 started, but by mid-February, the writing was already on the wall. Anyone who understands exponential growth and the cognitive biases that afflict all human beings would have known we were headed for disaster. We should have locked down a month sooner than we did.
What does this mean for the rest of 2020?
Let’s be clear: the only reason we’re seeing a decrease in death rates is because of the lockdowns. The corollary of this is that if we relax the lockdowns in any appreciable way, we can absolutely expect a second wave of COVID-19.
(Indeed, Ontario has started seeing a 2nd wave already that started about five days after our vacuous Premier, Doug Ford, starting relaxing lockdown measures. Five days is the median incubation period for COVID-19.)
We don’t yet know, or will ever be able to predict, whether the SARS-Cov-2 virus will mutate into something worse. We don’t even know how the warming weather will affect it. While we can and should continue to study it vigorously, we simply cannot affect how the virus will change in the coming months.
The only thing we can do is change our behaviours to protect ourselves and our communities, and remain flexible enough to roll with the punches — because this virus very likely has a couple of really good punches left in it.
We need contact tracing, we need universal testing, and we need to get more people and resources involved in studying the behaviour of the virus.
For instance, regarding the high death rate in older people: is that due to the virus’s effect on older humans, or a function of their environment (e.g., shared HVAC systems between “tenants” in senior living facilities)? That’s not yet clear. It may be that locking down old people with lots of other old people is actually the worst thing to do if you’re trying to save their lives.
Considering how steep the infection rate was before lockdowns became the norm, the only thing stopping COVID from killing huge swaths of humanity — for now at least — is our willingness to stay away from each other.
And if you find that unacceptable, then I’ll remind you of Anne Frank, who spent two years in an attic; no Netflix, no Amazon, no Skip The Dishes, and a World Fucking War going on around her. Please stop pretending your life is so important that you can’t just stay away from people for the sake of the rest of the species.
