Monday, March 30, 2020

A statistically sound way to count the death toll from covid-19 (EN)

Mr. Lemaitre lives in Brussels, Belgium. He is 70 and five years ago underwent surgery for lung cancer. In mid-February his carcinoma recurred and he was admitted to hospital. Subsequently his condition worsened and in the beginning of March he entered a coma. Less than a week later, his temperature spiked and he was tested for the novel coronavirus. By the time the result came back positive, Mr. Lemaitre had already passed away.

Mr. Peeters lives alone in rural Flanders. His age is the same, 70, and he has a chronic heart condition. But he is in decent health otherwise and he makes sure to regularly visit his cardiologist. Good monitoring, a healthy diet and a quiet lifestyle could easily give him 10-15 more years of quality life, said his doctor. At the beginning of March Mr. Peeters had a couple of bad days, but then was better. He didn't worry too much, it had happened several times in the past and the doctor had adjusted his treatment in response. He would anyway see his doctor in two weeks for his regular appointment.
As the days pass, Mr. Peeters begins to stress about the coronavirus epidemic. He knows he is in a vulnerable category, at his age and with his chronic heart disease, so he limits his contacts and spends most of the time indoors, watching the news. The epidemic is growing in his region, with more and more cases reported every day. He knows stressing is not good for his heart, but cannot help it. He has another bad day and he takes meticulous notes of all the symptoms, to tell his doctor when he sees him in a couple of days.
Then one day and a half before his scheduled appointment, the government on TV announces a lockdown, aimed at slowing the spread of the virus. All non-essential outings are banned. All regular medical appointments are canceled. Mr. Peeters panics. During the night, he has a heart attack and by dawn is dead in his bed.

While Mr. Lemaitre and Mr. Peeters both died at about the same time, only one of them died due to the coronavirus. Mr. Lemaitre was already on his deathbed when he contracted the virus; it might have accelerated his demise, but it certainly wasn't the primary cause. On the other hand, Mr. Peeters' death is obviously caused by the coronavirus pandemic, even if he might not even have been infected; indeed, were it not for the cancellation of his cardiology appointment and for the stress of the situation, chances are Mr. Peeters would be happily tending to his garden in the spring sun, and for years to come.

Still, in the figures announced in the news bulletins, it's Mr. Lemaitre's death that is counted among the victims of the virus, and not that of Mr. Peeters.

Both Mr. Lemaitre and Mr. Peeters are fictional characters, but you get the point. As a real life example, I could refer to the hundreds of people who died in Iran after drinking methanol based on the false belief that it would protect them from the virus. Does the pandemic play a role in their deaths? Most certainly, but you will not find them counted among its victims. Also forgotten will be the lives lost due to the economic collapse provoked by the lockdown measures to stop the pandemic.
The covid-19 statistics that so many of us follow daily, which make the media's frontpages and determine the policy responses of our governments, are flawed.
Different countries count fatalities in different ways and test to various extents for the virus. We have no clue how many people are infected, how many recovered without ever being tested, and we only make guesses as to the death rate. Official or quasi-official estimates are multiples of the headline confirmed figures. We scratch our head at the widely diverging death rates in Italy (very high) and Germany (very low) - but we don't even know whether the data is comparable, given the differences in testing and in counting fatalities.
For all the talk about the science's comeback, about evidence-driven policies, in this unprecedented crisis our leaders are to a large extent flying blind. And the stakes couldn't be higher. The measures being taken these days are historic: state of emergency, lockdowns, border closures, huge fiscal packages to keep the economy - and indeed the society - afloat. One would wish that the figures informing these decisions would be reliable, statistically sound.

But is there any way to have accurate estimates of the death toll of the coronavirus?
The answer is yes, and it was used before - but not yet in this crisis.

The key to meaningfully measure the death toll of the coronavirus, in a way that counts the case of Mr. Peeters (who died from a heart attack  triggered by the pandemic) and of the hundreds of Iranians who died poisoned from fear of the virus, while leaving out Mr. Lemaitre who would have died anyway, is based on a simple principle: compare the number of people who died during the pandemic with the number of people who would have died during the same period if there was no pandemic, with all the other factors unchanged. The difference will then show the net toll of covid-19, counting all the additional deaths due directly or indirectly to the virus.

And how can this be done?
By comparing the number of people dead during the period in question with the "usual" number from the similar period in previous years, controlling for known variations due to other factors (these days for instance one would not include in the comparison fatalities due to road accidents or workplace accidents, which have declined significantly due to the lockdown).

In summer 2003, much of Western Europe suffered an unprecedented heat wave. The most affected country was France - not because it had the highest absolute temperatures (it didn't), but because it had the largest, and most sustained in time, deviation from the multi-annual average temperatures for the period. Its population wasn't prepared to cope with such conditions, especially many elderly people left alone in the cities, in apartments without air conditioning, while younger relatives went on holidays.

As a result, it is estimated that during the first three weeks of August 2003 around 15,000 people in France died because of the heatwave. The estimate is based on calculating the "excess deaths" during that period compared with the same period in previous years (see the study here, in French).
The immediate cause of death of course varies, all we know is that some 15,000 people would have not died were it not for the heat. In fact, for the most part we don't even know who these 15,000 are among all those who passed away during that period. But we know with a high degree of statistical certainty that they were close to 15,000.

This method of calculation isn't really appropriate for limited, localised and more straightforward impacts, like those of smaller epidemics such as SARS (another coronavirus that made the headlines in 2002-2003). But it is by far the best possible method to measure the impact of large-scale phenomenons, which affect in many ways, directly and indirectly, a significant part of the population, for which we cannot even begin to measure at the micro-level every consequence. Like the 2003 heatwave in Western Europe. And like the global coronavirus pandemic that we are currently living through.

To perform a similar calculation for covid-19 in Europe, one already has the month of February and soon will have the complete month of March to compare with the "normal" numbers of dead for these months. For more granular results, one could compare the numbers by age cohort - will probably not find much of a difference overall, but might detect a change in some countries (Italy, Spain) among the elderly. In any case, such a calculation would offer a far more accurate image of covid-19's actual death toll than any of the numbers making headlines these days.

So why hasn't anyone done it yet?

No comments:

Post a Comment