Why it’s a bad idea to test everyone for COVID-19

“Everyone should be tested!” they cry.

“It’s the only way to know who has the virus and who doesn’t!” the calls ring out.

“We’ll never be safe unless we all get tested!” the thinking goes.

That’s not exactly true. Here’s why:

When I was in grad school studying epidemiology I had the very good fortune of my advisor being in a management position at the State Hygienic Laboratory in Coralville, Iowa. He helped me get connected to a part-time job that reinforced some of the core principles I was learning about in class: sensitivity, specificity, predictive value positive, and predictive value negative.

These are not terms most people hear every day, so here’s a general definition of each.

• Sensitivity: assuming you’re infected, what’s the likelihood the test will be positive?
• Specificity: assuming you’re not infected, what’s the likelihood the test will be negative?
• Predictive value positive (PVP): if the test is positive, how confident are we that the person is truly infected?
• Predictive value negative (PVN): if the test is negative, how confident are we that the person is truly not infected?

So why is this important? Within the context of COVID-19, here’s the problem with poor test performance on any of these things.

• Poor sensitivity: a lot of infected people will not know they’re infected and will spread the disease to others.
• Poor specificity: a lot of non-infected people will think they’re infected, causing them to be quarantined, stressed, missing work, separated from loved ones, etc.*
• Poor PVP: among people who test positive there will be a lot of wasted time and resources, which is a problem when you start trying to test millions of people.
• Poor PVN: again, wasted resources.

*Yes, this is similar to what’s happening now for many millions of people, but a person with a false positive test result would be truly quarantined. Locked at home. Unable to go literally anywhere but their house. That’s like social distancing on steroids and we should avoid that as much as possible.

“But wait, what if a test is 99% accurate?”

Even tests that are very, very good, are wasteful if everyone is tested. Here’s an example that reflects the current situation in the U.S.

Today, April 21, the CDC website says there are almost 800,000 cases in the U.S. Let’s overestimate and say that means there are actually 8,000,000 people who have the virus (assuming that for every 1 person diagnosed there are about 10 people undiagnosed). There are about 330 million people in the U.S., so we can start to make a simple table of who’s infected and who’s not.

Clearly there are some missing figures. Let’s fill those in, shall we? Using the question above, let’s assume that the test in question can find 99% of infected people. (99% sensitivity)

That looks pretty good, although it means we have 80,000 people walking around with no idea that they’re infected. Now, let’s also assume that 99% of the non-infected people end up with a negative result.

Not bad, but I think you can probably see some issues. First off, we’ve told 3.2 million people that they are infected when they’re actually not. That’s 3.2 million people we now have to lock in their homes. They can’t go to the store. They can’t take a walk. They can’t get back to work. They’re stressed out and panicking because they think they have a serious infectious disease…but they don’t. Beyond that, whatever they were contributing to the economy (think: work productivity, restaurants, home improvement, decorating splurges) is gone. So it’s not just that those 3.2 million people are suffering, which is terrible enough, but their absence from society may have far-reaching effects.

In a similar way, this scenario means our healthcare system has just run 318 million tests to tell people what they probably already knew or assumed: they’re not infected…for the moment. Will we have to test them all again in six months? A year? Our test result tell us that “up until that point” the person was not infected. The numbers get a lot bigger if we have to run all of these tests again. And again. And again.

And this is where predictive value positive (PVP) and predictive value negative (PVN) become useful tools. The PVN in this case is pretty good. It’s the number of true negatives divided by the number of negative tests: 99.97%. This means if we get a negative result we can be almost certain that the person is really not infected. PVP, however, is poor. It’s the number of true positives divided by the number of positive tests: 71%. We can only be 71% confident that a positive result is accurate.

At the lab we had a saying: “Garbage in. Garbage out.” The idea is that if what you’re starting with is garbage, it doesn’t matter how good the rest of the process is. The result will always be unreliable.

So how do we do better? I mean, I said 99% sensitive and specific is as good as we can expect and even if we had perfect tests (which only exist in theory) we’d still run millions of tests for people who have no indication that they might be infected. We need to find a way to keep the garbage out of the testing process.

The solution: test the people who have some indication that they might be infected.

This starts with symptomatic people, which is obvious, but many people infected with the COVID-19 virus may not have symptoms. So let’s assume we can identify about 100 people with some chance of contact with the 800,000 known cases in the U.S., either because they’re family members, close co-workers, or friends at an ill-advised dinner party, etc. These people don’t have symptoms, but they do have a reasonable chance of being infected. So instead of testing 330 million people we only test 8.8 million.

By focusing the testing strategy on people who are most likely to be infected, we will probably still find almost all of the infected people and we dramatically improve the performance of a test. We’re still finding 99% of the truly infected people, and we’re still making sure that 99% of the non-infected people are sent back to their lives, but we’ve also saved millions of tests for future use and cut way back on the number of people we quarantine unnecessarily.

This shows up in the PVP. When we tested everyone, we had pretty low confidence in the positive results. Only 71% of the positive results were trustworthy. Now that number is 91%.

Obviously, this is a contrived example but the point is this: when you’re dealing with limited resources, you have to use those resources well. Testing people with no reasonable chance of being infected wastes time and resources, and hampers our ability to trust test results. And the last thing we need right now is people not trusting test results.