No, epidemiological models do not serve to predict the future

Different scenario envisaged by the Québec public Health at the beginning of the month of April.

Share

July 17, 2020 12h56

Share

No, epidemiological models do not serve to predict the future

Maxime Bilodeau

Agence Science-Presse

DETECTOR RUMORS / epidemiological models have become an easy target for those who want to deny the merits of containment : a number of these models, from Europe to North America, have “predicted” a number of deaths higher than what is actually produced. Are they to throw at the basket ?

On 16 march, the british epidemiologist Neil Ferguson and his team from imperial College, London, reveal an epidemiological model, which stipulated that in the absence of strict measures to curb the spread, the COVID-19 could cut about half a million people in the United Kingdom and over 2 million in the United States. Ten days later, a similar study of the same College was the predictions just as gloomy for many other countries. In Canada, it was a question of 326 000 deaths.

The worst-case scenarios

Four months later, it is clear that the imperial College has overestimated, by far, to the death. Even if the pandemic is not over, the current situation is light years ahead of the published figures.

Except that these models were presented in mid-march as a worst-case scenario : “If no action was taken against the epidemic, one might expect to… “. In other words, these are computer simulations whose goal was to predict what would happen if the various countries kept their borders open, undertook little or no containment, did little screening, etc

The same week they were released, the assumptions on which they were based were already no longer valid. On 18 march, the canada-u.s. border was closed. On 23 march, Quebec city ordered the closure of all businesses that are not essential. Result : two weeks later, the different simulations presented by the public health agency of Canada oscillated rather between 11 and 22 000 deaths by the end of the pandemic (mid-July, the Canada was approaching the 9000 deaths).

As for Britain, it must be remembered that this simulation has fallen at a time when the country seemed to always follow a controversial policy of acquisition by the population of a “herd immunity” — no containment, the normal pursuit of activities. The model Ferguson is often cited as the cause of the shift undertaken in the following days by the prime minister Boris Johnson.

Even before the containment measures were imposed around the world, its forecast “pessimistic” was not the only one to move. In the United States in mid-march, Dr. Anthony Fauci mentioned the “hundreds of thousands” of deaths. On 26 march, after two weeks of confinement in many u.s. States, a model of the Institute for Health Metrics and Evaluation, University of Washington, who went on to be cited by the White House, spoke of 100 000 to 240 000 deaths — in a scenario where a containment severe would be maintained until the summer in the country as a whole.

The oldest official forecast, available on the website of the Centers for disease control (CDC) is the 13 April, after a month of confinement : the scenarios ranged between 60 000 and 150 000 deaths by the end of may (the United States has passed the milestone of 130 000 people died in the 7 July, and could reach 150,000 before September).

In response to criticisms of the weaknesses in the computer coding of the simulation of Ferguson, the british journal Nature published on June 8, the comments of the experts who have tested and have found to reliable.

Designed to be contradicted

But they are pessimistic or optimistic, these models are always published with the hope of being contradicted, and not to quantify with precision the number of cases, hospitalizations and deaths to come. A study published last year and reviewed the epidemiological models published during the Ebola epidemic in 2014-2015, concluded in this regard that, because of the actions taken by the people who are always changing the game, these models are unable to predict the future beyond a horizon of one to two weeks.

Although it is too early to say, it is possible that this conclusion is also valid for the current scenarios déconfinement and second wave : all are trying to draw the outline of a possible future, but they can only do that though imperfectly, the big unknown being how will millions of people.

“All models are wrong, some are useful”, as is the saying, attributed to the statistician George Box, and the mathematician Norman Draper, in their book empirical Models of construction and surfaces of reaction (1987).

This is the paradox even of prevention in the context of public health, wrote in the June 9, Benoît Mâsse, School of public health, University of Montreal : if the measures ” are effective to control the epidemic, they give the impression that they were not needed, and we therefore leave it vulnerable to another wave of infection “. It was in response then to an “economic note” from the Montreal economic Institute, who questioned the merits of containment, accusing the government of being supported on the “mathematical model inaccurate” at imperial College London.

Anyway, in any science which produces models, these are based on the successes and the mistakes of their predecessors. The representations of the beginning give way to representations that are more complex. For example, we know a lot more now about the circumstances that influence the risk of infection.

Le Soleil