COVID-19 Test Accuracy: When is too much of a good thing bad?

More accuracy sounds like a good thing, and in the abstract that is true. The analytical accuracy of a PCR test (the short name for a polymerase chain reaction lab technique) is too much of a good thing when used to determine an individual’s infectivity — which is what we really need to know to stop the epidemic. PCR tests cost too much and take too long, too few will be tested and the results will arrive too late to act on them. Counterintuitively, less sensitive antigen tests are more effective at identifying those who could otherwise become super-spreaders.

There is a growing consensus that these inexpensive, while-you-wait tests performed every day are the key to controlling the pandemic. This insight was first published last June (as a pre-print on medRxiv) by Michael Mina, of the Harvard School of Public Health, and by Daniel Larremore and Roy Parker of the University of Colorado Boulder. Helpful summaries were recently published in The New England Journal of Medicine and in The New York Times.

The most accurate tests for the SARS-CoV-2 virus that causes COVID-19 use PCR to detect the presence of its unique RNA signature. This test is exquisitely accurate — a single copy of the virus in a test sample can be detected. Unfortunately, this test typically costs about $100, and because the sample has to be transported to a central laboratory to perform, it takes a minimum of one to two days for the results to come back. In fact, due to overwhelming demand over the summer, results were routinely taking one to two weeks.

Clinical diagnostics are designed and authorized to inform physicians how to treat a patient with symptoms. Knowing whether the symptoms are caused by COVID-19, an influenza virus or the common cold requires a high degree of accuracy. Waiting one to two days for the most accurate result is acceptable because the treatment in those first two days is primarily observation to detect evidence of serious complications. The US health care system — physicians, hospitals, the FDA — is built to ensure that tests are accurate enough for individual treatment to be based on their results.

It has become clear over the last three months that the virus is airborne and that patients are infectious for about four days — the two days before symptoms appear and the next two days after that — of the average 14 days or so from exposure to full recovery. If testing is only done when patients notice their symptoms, or more likely a day or two later, whatever onward transmission is going to occur is then past history. Those tests can still be useful to guide individual patient care, but the virus has marched on past them.  A compounding factor: At least 40% of COVID-19 patients never develop symptoms at all but are just as infectious, and a further 25-30% with minor symptoms are unaware they have COVID-19. These two groups are rarely tested and therefore continue to expose others.

We know that herd immunity will be the point that the epidemic is contained, i.e., when 60-75% individuals have recovered or been vaccinated, and the virus runs out of new people to infect and dies out. The best estimate is that by the end of June 2020, only 9.3% of the population had immunity through past infection. In the three months since, a further five million cases have been reported, implying that current prevalence has reached about 20%. This is a long way from herd immunity — the remaining 80% of us are still vulnerable and the goal is still about two years away, i.e., sometime around mid-2022, less whatever a vaccine rollout adds in 2021.  Until then our only weapons are masks, distancing and frequent testing with isolation.

If PCR is too expensive and slow, what is the alternative? Antigen tests provide while-you-wait results and are much cheaper to make (currently available and approved tests cost $5-30) but with accuracy from 70-95%, below the 95-100% of PCR. These tests are cheap and simple enough to be used several times per week, enabling a nationwide fast-frequent testing paradigm.  But are they accurate enough? The answer is yes. Explaining why requires some investigation of SARS-CoV-2 infection dynamics.

The reason PCR is so accurate is that it locks in on segments of the viral genome, and if it finds any, it amplifies the result exponentially to make it easier to detect. Different PCR tests have variable ability to detect virus, but even the very worst will give a positive result when there are about 540 detectable fragments in a microliter (one millionth of a liter). Antigen tests detect the numerous copies of the viral proteins built from the viral genome, but their technology cannot amplify the result, so about 1,000-times less sensitivity is the result.

Knowing that the virus is present is not what we really want to know. What we need to know is: Can this person transmit COVID-19?  Test “accuracy” should be defined in terms of this objective, not mere viral presence. A viral culture test is the gold standard that measures exactly this: Can the virus in a sample infect human cells or not? These tests report how many cells can be infected by the viral load in a sample (expressed in PFU/ml — plaque-forming units, or TCID50/ml —tissue culture infectious dose required to infect half the cells present). These tests are too slow and expensive to be used routinely for patients, but they are the standard against which all tests are judged for infectivity. It takes 7-10,000 viral copies to infect one human cell. PCR’s greater accuracy detects viral load far below this level, and thus incorrectly identifies many individuals as infectious when they are not.

On the plus side, PCR is better at detecting the early hours of infection as the virus becomes established in a human host — the start of 4 days of peak infectivity. On that first day, viral load is exploding from undetectable to trillions of viral particles in hours. Even if a less accurate antigen test misses the early hours of infectivity, it will definitely catch all cases by later that first day and on all subsequent days of infectivity. Over the next five to 10 days as patients recover, viral load declines only slowly, so PCR continues to be positive throughout this period. This is a serious disadvantage because these PCR positive individuals are “false positives” in the sense they can no longer transmit the virus to others. Recent work carried out at Johns Hopkins on behalf of BD Life Sciences (a vendor of Veritor Plus, one of the approved antigen tests) clearly confirms that the “lower” sensitivity of antigen tests provides a much more “accurate” answer about infectivity than the “higher” sensitivity RT-PCR test.

PCR is just too sensitive to tell us accurately whether an individual’s breath has the capability to infect others. Antigen test sensitivity is more appropriate to identify those who need to be isolated to prevent ongoing transmission.