What the science says about the sensitivity of coronavirus tests

(Leah Hogsten | The Salt Lake Tribune) Brenda Vijil, a Diné medical assistant at Utah Navajo Health System's Monument Valley Clinic, records an incoming patient's personal information waiting in their vehicle before admittance into the triage unit to take vitals and administer a coronavirus test, Aug., 24, 2020.

A debate among researchers has been picked up by coronavirus skeptics — could tests for COVID-19 be a little too sensitive?

That was the thesis of an article published by The New York Times with the headline “Your Coronavirus Test Is Positive. Maybe It Shouldn’t Be.”

You may remember my explanation of how most COVID-19 tests work using a process called PCR. Here’s a quick recap: The PCR process doubles the small amount of viral genetic code found in a swab sample over and over again until there are many millions of copies. At that point it can be detected, usually by a machine.

It turns out that the number of cycles it takes for the viral RNA to be detected can be reported on each test. And it’s a decent proxy for how much virus was in the sample to begin with — the more doubling cycles it takes to make the virus detectable, the less there was to begin with. If it doesn’t take very many cycles, there was probably a lot of virus in the sample. That number of cycles it takes to detect the RNA, typically between 15 and 40, is called the cycle threshold.

Most tests used now in the U.S. give up after either 37 or 40 cycles. If they find the virus before that, the test comes back positive, and if not, the test comes back negative.

The Times article quoted scientists who believe those numbers are too high. If it takes 37 or 40 cycles to find detectable virus, the likelihood is that the virus detected was trace fragments of dead virus, found either in the community or after a bout with COVID-19 from weeks or months ago. And in particular, one of the doctors, Dr. Michael Mina from the Harvard T.H. Chan School of Public Health, would prefer the cycle threshold to be 30.

According to The Times, 85% to 90% of positive tests in Massachusetts had cycle thresholds between 30 and 40, so if we changed the test parameters, it would make a huge difference in our number of cases.

This article has been pounced on by COVID skeptics, who assert that a large amount of our positive tests are essentially misleading. I wanted to know if they had a point, so I looked up the available research.

It’s clear that cycle thresholds are meaningful. There are significant correlations with the cycle threshold and various virus metrics. One study showed a correlation between low-cycle threshold and deaths, seven studies out of 11 found a correlation between low cycle threshold and severe disease. Lower cycle threshold numbers were correlated with lower white blood cell counts, which means the immune system wasn’t working as well.

And here’s the thing about correlations: They usually work but not always.

Let me show you what I mean. A study of coronavirus positive tests in England from January to May found that those samples with high-cycle thresholds mostly had dead virus, with no actual infectious virus in their sample.

Cycle thresholds compared to whether or not infectious virus was found in the sample. As cycle threshold values increase, the likelihood of infectious virus decreases. (https://www.eurosurveillance.org/content/10.2807/1560-7917.ES.2020.25.32.2001483#html_fulltext)

Just 8.3% of people who had a cycle threshold between 35 and 40 had infectious virus in their sample. And 28% of people who had a cycle threshold between 30 and 35 had infectious virus. These are pretty low numbers! There’s clearly a correlation between cycle threshold and whether someone can infect someone else.

And yet, let’s say you knew someone who had a 10% to 30% chance of actively having infectious coronavirus. Would you want them teaching your children? Would you invite them to dinner with your family? Would you want your grandparents to shop in the same store? Would you trust them with delivering health care to someone with preexisting conditions? Your answers to those questions probably depend on your individual situation.

From the same study, here’s another example of how messy it can get. They tried to look at cycle threshold vs. when symptoms occurred. Would it change during the course of the disease?

It turns out the two were correlated — as people’s immune systems got to work, the amount of virus dropped and the cycle thresholds frequently went up. Makes sense! But as you can see, it certainly wasn’t a perfect correlation. A couple of days after showing symptoms, the study showed some people who had a cycle threshold of 18, and others who had a threshold of 40. It’s about the same 10 days after symptoms. The data is messy, and a single test cycle threshold couldn’t really tell you where you are in the course of disease.

There's a correlation between the amount of time someone has had symptoms and cycle threshold values — but the data is pretty wild. (https://www.eurosurveillance.org/content/10.2807/1560-7917.ES.2020.25.32.2001483#html_fulltext)

Or take another study that looked at asymptomatic coronavirus cases. Among those, the cycle threshold looked to be correlated with whether they eventually developed symptoms. Asymptomatic people who later developed symptoms had an average cycle threshold of 34, where the ones who didn’t develop symptoms had an average cycle threshold of 39.

But then again, it wasn’t consistent. And the study showed that the two groups were shedding virus for a similar amount of time, seven days vs. eight days.

Let me add just one more complication: Comparing cycle thresholds isn’t an apples-to-apples situation.

Coronavirus tests are pretty different — how much virus a sample collects might depend on whether someone gets a saliva sample, a throat swab or a nasopharyngeal swab. Those samples are then diluted into a solution that isn’t always consistent from test to test, then read by various machines that don’t all work to the same specifications. Cycle thresholds could be quite different depending on those variables.

With all of those caveats, I can understand why coronavirus testers aren’t in the habit of giving out cycle threshold counts — in fact, none of the major testing providers does. I’m always on the side of more information, but given some of the poor decisions people have been making with far more solid evidence, I understand why these cycle thresholds aren’t usually reported with a positive or negative result.

But while I don’t agree with Dr. Mina that the cycle threshold cutoff for positive cases should be at 30 or lower given the evidence above, I do agree with him when he says medical professionals should consider the cycle threshold when evaluating their patients — especially those who have had symptoms over a long period of time. In an article for the journal Clinical Infectious Diseases, Mina and a colleague argue that patients with symptoms that have fully resolved who see two tests with cycle thresholds over 35 shouldn’t be retested. They’re likely good to go.

That being said, even if you’re a medical professional, it can be difficult to find a test that will report the cycle threshold count. That’s a shame. While this information shouldn’t be broadly available, it can be a useful diagnostic tool in the right hands, especially with repeated testing.

Still, with how long tests are taking to come back — a majority of tests take more than 24 hours in Utah — giving positives for even low virus amounts makes sense, because low viral amounts can become high viral amounts in the course of a couple of days at the beginning of infection. In those cases, a high cycle threshold value is a warning sign.

Scientists may continue to debate this issue, but for most of us, we should stick to the basics. If your coronavirus test is positive, it means that you either could be or are infectious, and you should probably quarantine.

Andy Larsen is a data columnist who is focusing on the coronavirus. He is also one of The Salt Lake Tribune’s Utah Jazz beat writers. You can reach him at alarsen@sltrib.com.