Like many other countries, Canada is generally testing a narrowly defined group of people for COVID-19 — people showing severe symptoms, those at risk due to other health problems and health-care workers. B.C.’s testing criteria, in fact, explicitly instructs health-care workers not to test patients with mild symptoms.
Testing the sickest among us is useful to protect patients and front-line workers from catching and spreading the disease. However, this information is also used to estimate the spread and severity of the disease, which in turn guides our country’s response. At best incomplete, this data can only provide us with a severely skewed picture of how this disease works. Canada needs to adopt a very different approach, one currently only Finland appears to be considering, to get the data we need.
Although the specific criteria for testing vary by province (and are, in truth, a little vague), the focus on those most seriously infected means we only ever see the effects of the novel coronavirus within this group. What we won’t see is how the coronavirus plays out in those who have milder or no symptoms. As a consequence, the mortality and severity rates we hear about every day do not reflect COVID-19’s actual impact on the broader population.
It’s difficult to overstate the possible magnitude of this bias.
At one extreme, it is entirely possible that the coronavirus truly does make people extremely ill — in which case, testing only the very ill would give us an accurate sense of mortality and severity rates. At the other extreme, the coronavirus may make only a very small fraction of people seriously ill. The problem is that testing only this fraction of the population for infection would result in exactly the same statistics, making it impossible to determine which of the situations is truly the case.
“If Canada continues to test in this way, it could face a similarly enormous gap between estimates and reality.”
This bias is understood by epidemiologists, but the extent of the bias, and the relatively straightforward solution, seem to have been paid little attention by policymakers.
To illustrate how big of a problem this is, consider Italy. At the time of writing, Italy had confirmed 139,422 cases of COVID-19. However, using recently published fatality estimates and accounting for Italy’s age distribution, the total number of deaths in Italy (17,669) suggests that over 1.2 million people may have actually been infected.
That means there could be over one million infected people with symptoms not apparently severe enough to be tested. A report from Imperial College suggests this figure could be closer to six million. If that’s the case, the 139,422 confirmed cases may in fact represent only the most severe two per cent of cases.
If Canada continues to test in this way, it could face a similarly enormous gap between estimates and reality.
The second, less-obvious consequence of Canada’s current testing strategy is that it likely underestimates the difference in the severity of the disease between young and old people. Already, the coronavirus appears very much more likely to sicken and kill older people than younger people.
Data from the CDC, for example, indicate (upper bound) hospitalization rates of 70 per cent, ICU rates of 29 per cent, and mortality rates of 27 per cent for those 85 and older. Compare that to the rates reported for those under 20: 2.5 per cent hospitalized, with zero ICU admissions and deaths.
While these statistics almost certainly overestimate the severity of the disease in both the young and old, somewhat counterintuitively, the overestimate is probably larger for young people. Testing only the very sick means proportionally fewer young people are tested, resulting in a bias likely even larger for this group than in older age groups.
A possible solution
There is (at least in principle) a simple method of testing that would provide an unbiased estimate of the effects of the coronavirus — testing a random sample of the population and observing the consequences for those who test positive.
Testing this way would give us an accurate sense of the consequences of the coronavirus, because our sample would represent everyone, not just those who are most sickened or most at risk.
There would be other benefits, too. Random testing would give us an accurate picture of the number of active cases in the population (not total cases, because some people have recovered, and there is a different test for that). Random testing would also provide more accurate information about the effect of the disease on people in different age groups (or any other demographic group), because different age groups would now be appropriately represented in the sample.
A second approach to get accurate information is to use widespread testing, as Iceland is currently doing. The Ontario government seems to be pushing for something similar. However, this approach can lead to biased estimates, too, if the people who are tested first are different in some way that affects how they respond to the disease. The bias will progressively decrease as the total number of tests increases, but, to put this in perspective, at Ontario’s stated goal of 13,000 tests a day, it would take three years to test the entire Ontario population.
Have an opinion you’d like to share on HuffPost Canada? You can find more information here on how to pitch and contact us.
Also on HuffPost:
- What are the cases of the new coronavirus in Canada? Take a look at our map.
- Wondering which financial assistance to apply for? Check this out.
- What’s the difference between the coronavirus and the flu?
- You’ve probably been hearing a lot about PPE. What it is — and how to donate it.
- Things are changing quickly: a cross-Canada look at which services are open and closed.