The Canadian federal government recently announced it would fund at least one million blood tests to track the novel coronavirus over the next two years.  This is a step in the right direction.  But is it enough?

How do we know if we are testing the right number, and the right people, especially as the pressure to ease up on the lockdown and isolation rules increases?  The answer depends critically on why we need test results.

A sophisticated sampling strategy is the only path forward at present.  There simply are not enough resources to test every one or even perform a simple random sample.

Perhaps half of those infected do not show more than very mild symptoms, if any. These individuals have greatly complicated efforts at epidemic control.  If everyone infected with COVID-19 had symptoms, we could simply require them to self-isolate. Instead, we have to keep two meters from everyone and wear masks, because we cannot tell if they are infected but not showing it.

Public health agencies clearly need to continue and expand testing for high-risk populations, including front-line health care workers, personal support workers (PSWs) working in nursing homes, retirement residences and in home care, as well as the public health workers doing contact tracing.

But we need to do better, with reliable real-time data on how many people in the population are, or have been, infected and where they are.  Tests of current infections, with a lag of a few weeks, signal impending hospitalizations. That’s no longer good enough.

It is essential to detect and isolate infected individuals quickly, and as many of their contacts as can be traced, if we want to relax the restrictions as quickly as possible.

Blood tests, as the government has just announced, can tell us how many people have been infected (though the amount of resulting immunity remains unknown).  But these will likely be far below those needed for herd immunity, so low that significant relaxation of physical distancing would result in a surge of new infections — straining health care resources, causing more deaths, and requiring the reintroduction of draconian controls.

To monitor adequately, the tests cannot cover only symptomatic individuals since this will miss the large asymptomatic or pre-symptomatic portion of the infected population.

It matters who is tested.  If it’s mainly individuals who live alone and are careful about physically isolating, most tests will be negative.  For PSWs or meat processing plant workers, though, the same number of tests could find much higher rates of infection.

To provide valid and useful results, testing needs to be based on sophisticated sampling, simple random samples will not work.  A highly controversial Santa Clara study of how many residents have been infected shows the perils of poor sampling.

For example, nursing homes need their own samples; indeed, every resident should be tested periodically for the time being.  For the general population, though, a multi-pronged effort is needed, starting with new clusters of infection, including key groups such as front-line workers in shops that are re-opening.

We also need to distinguish geographic regions within provinces, for example, different cities.  Even though most of the public discussion has been about policies at the provincial level (and state level in the U.S.), proper sampling, and relaxation policies, will need to target very real differences within provinces.

In sum, we need a sophisticated sampling strategy for testing, and one that can adapt to evolving circumstances, not least as physical distancing and related policies are reduced.

These data need to be accessible for statistical analysis not only locally, not only provincially, but also nationally, and of course, they need to be securely handled and protected.

Testing results, based on proper and extensive sampling, are fundamental to improving the model results shown on TV and used by governments to inform the relaxation of restrictions.  This would allow us to return to a new normal – more quickly, and with lower risks of serious mistakes.

Michael Wolfson is a statistician.

Image credit: Shutterstock.com