One constant during the chaos of the COVID-19 pandemic has been the call for more testing. That’s no less true now, even as the United States has greatly expanded its capacity from the earliest stages of the outbreak, when technical problems at the US Centers for Disease Control left the country well behind its peers.
But there’s much more to COVID-19 testing than capacity. The two main categories of COVID-19 testing—screening testing and diagnostic testing—play very different roles in an effective public-health response to the pandemic. We have different needs and different technologies for each: antigen testing for screening and RT-PCR for diagnostic testing. (For simplicity’s sake, we’re setting aside antibody testing and rapid NAAT.)
Ideally, screening testing is for everyone, or at least everyone who has to go into work or school. It casts a wide net and is meant to be in constant use throughout the population. It will miss some cases but will hopefully catch the most infectious patients, making up for its flaws with its ease and omnipresence. Cheap, fast antigen tests aren’t terribly accurate, but consistent testing, some math, and a clear sense of what they are capable of could compensate for their comparatively low sensitivity.
In many scenarios, antigen tests are more appropriate than highly sensitive diagnostic tests, whose role is suggested by their name: If there’s good reason to think that a person has been infected with COVID-19, such as returning a positive surveillance test, showing symptoms, or having been exposed to a known case, these more powerful tools can provide a clinical diagnosis.
How screening tests can provide the first line of defense against COVID-19
When you go to the doctor, a quick listen to your heart and lungs through a stethoscope can help the doctor decide if they need to order more timely and costly procedures. That’s the role of screening tests in the COVID-19 public-health response. They’re meant to be ubiquitous and fast, and to stop obvious problems before they get completely out of hand. The World Health Organization’s September interim guidance for antigen-based tests, for instance, suggests these tests should have minimum sensitivity of 80 percent. In other words, out of 10 people with COVID-19, the test returns positive results for eight of them.
At first glance, that might seem dangerously low, especially compared to PCR. But if that test’s cost and ease mean that it can be used on a consistent basis by millions of people, it could be more useful than a highly sensitive but expensive and slow test that requires an expert to run it. To begin with, we can use basic logic to make screening-test results more useful. If a health care worker in a community with high prevalence gets a negative test result, it’s more likely to be a false negative than the same result for someone who works from home in a low-prevalence community. This is what’s called high clinical suspicion or pre-test probability.
Physicians Jeffrey L. Schnipper and Paul E. Sax suggest using Bayes’ Theorem to put numbers to this logic. For example, they estimate, based on testing data, that the pre-test probability of someone in Massachusetts without COVID-19 symptoms is 1 percent, so 10 out of every 1,000 people are infected. The sensitivity of the test is 80 percent, so out of those 10 people, 8 test positive and 2 test negative. We can then estimate that the test misses 2 cases of COVID-19 out of every 1,000 tests. Thus the pre-test probability and the test sensitivity, used together, give a negative predictive value of 99.8 percent.
If the 1,000 patients being tested are all symptomatic in a high-prevalence community, they estimate the pre-test probability to be 30 percent. Using the same method produces a negative predictive value of 92 percent, and testing that cohort will result in 60 false-negative results out of 300 total infected people in a population of 1,000. This, as Dr. Schnipper and Dr. Sax wrote in Stat News, is a problem: “Now it’s time to worry about low sensitivity: 8 percent of those with negative tests actually have Covid-19. That is too high, and it’s why symptomatic patients at our hospital are tested twice—using the most accurate tests available—24 hours apart, with additional testing if we’re really suspicious.” It’s the same exact test in both of these scenarios, but the pre-test probability changes everything.
Accessible, lower-sensitivity tests could be better for screening because they’re more likely to deliver information when it’s useful. PCR tests can take hours to run, and when we take sample transport and processing capacity into account—one machine can run up to 1,000 tests a day—a good turnaround time can be 24 to 48 hours. Many Americans have, of course, experienced turnaround times of days, or even weeks. September’s average in the US was 2.7 days, down from four days in April, but the average turnaround time for Blacks and Latinos was still over four days in September. As of November 13, according to the most recent data from the US Health Resources and Services Administration, 57 percent of tests were turned around in two to three days; another 18 percent took four to five days.
The researchers who calculated that figure also found that the time between when a person decides to get tested and takes the test is 2.5 days. That delay may be thanks to the expense of tests, the travel required to get one, or the lack of clarity about where to get a test—all problems that cheap, ubiquitous tests would address. The window for peak COVID-19 infectiousness begins a day or two before symptoms appear and closes within a week thereafter. Delays mean that a test could miss a large portion, or even all, of the most infectious period of the illness. For example, if someone thinks they may have been exposed and then waits three days to get a test, and that test takes four days to return results, that’s a total of seven days from exposure—and by the eighth day, infectiousness may be in “steady decline.” If the delay is too long, it makes the test useless for screening purposes. And, of course, a lack of testing during the peak infectious period could result in a person unknowingly exposing others to the virus.
Antigen tests are also most effective when patients are most infectious to others, which typically begins a couple days before symptoms present and continues for a few days. Antigen tests have low sensitivity compared to PCR, meaning that they will miss the presence of the virus in a higher percentage of people in which the virus is present. The lower sensitivity of antigen tests means that most of the false negatives they generate mostly come in the “long tail” of COVID-19, when the patient still has the disease but may no longer be infectious. But that’s not as worrisome as it sounds. “In the context of outbreak containment, the antigen test’s limited window of sensitivity is a major asset,” A. David Paltiel and Rochelle P. Walensky recently wrote in Health Affairs. “The antigen test is ideally suited to yield positive results precisely when the infected individual is maximally infectious.”
The range of reported antigen test sensitivity is much wider than that of PCR, but at the peak period of infectiousness, it can be comparable to PCR. To be clear, a rapid antigen test does not indicate infectiousness, nor does a negative antigen test and a positive PCR indicate that a patient has COVID-19 but is not infectious. Antigen tests are most likely to come back positive if performed when a patient is in the period of maximum infectiousness.
There is a short period where the patient is infectious but antigen tests are not sensitive enough to detect it, meaning that there are risks to a false negative. But that risk has to be balanced against the risks introduced by the long processing time of PCR tests. Michael Mina, a Harvard epidemiologist, laid out a plan to The Atlantic that would involve Americans taking tests multiple times a week, as one might take a pregnancy test, although Americans use a mere 20 million pregnancy tests a year while Mina’s plan would require more than that every day. (In a recent study co-authored by Mina, it’s suggested that antigen tests would be essentially as effective as PCR for surveillance testing if given every three days.)
As a screening tool for COVID, the CDC’s guidance supports the effectiveness of antigen testing, even suggesting that it could be useful in high-risk congregate settings, though it cautions that there is “limited data to guide the use of rapid antigen tests as screening tests on asymptomatic persons to detect or exclude COVID-19, or to determine whether a previously confirmed case is still infectious.” In a high-risk setting with high pre-test probability, the CDC suggests that confirming a positive rapid antigen test with PCR may be unnecessary.
The recent White House outbreaks are an example of the uses and limitations of antigen testing, with the caveat that there is still a lot we don’t know about them, such as what screening test the White House has been using, how often tests were given and to whom, what types of tests returned positive results and when, and what actions were taken after positive test results. The White House has reportedly relied heavily on screening tests without a similar dedication to mask use and social distancing. Because of antigen tests’ lower sensitivity, face masks and social distancing measures—which don’t always appear to be standard practice in the White House—are critical to preventing the spread of the virus. One concern about less-sensitive point-of-care antigen tests has been that they will be treated as conclusive rather than presumptive by people who take them, leading them to forego basic precautions—something public health and science communications have to take into account.
But experts also told Stat News that without antigen testing, the problem at the White House could have been far worse; those who tested positive could have continued spreading the illness for even longer. Antigen tests will miss some infections, at times when the test subject is still infectious. Constant testing means that an infection that isn’t caught today can be caught tomorrow. The White House outbreaks prove that while widespread antigen testing is a powerful tool, it must be paired with PCR testing and basic precautionary measures.
Why PCR tests are the gold standard for detecting COVID-19
That’s why PCR, despite its expense and delay, remains an important part of the process. If antigen tests are like stethoscopes, the high sensitivity of PCR tests make them like chest CT scans: They can confirm the results of antigen tests and ensure more accurate results in especially sensitive scenarios, like clinical settings, when the potential for false negatives presented by antigen tests is too great. Wider use of antigen testing would also free up capacity for PCR tests for their more appropriate uses.
In real-world clinical settings, there are confounding factors that can reduce the sensitivity, particularly how the sample is obtained and when it is obtained in the course of an infection, which is why the sensitivity is sometimes described as a range. If that sounds confusing, that’s because it is, even for experts. As Edward C. Stites and Craig B. Wilen wrote in “The Interpretation of SARS-CoV-2 Diagnostic Tests” in the November 2020 issue of Med, “there are several nuances to laboratory testing that are not well appreciated by the general healthcare provider.” The virus, for example, may be present in one part of the respiratory tract but not another, so the sensitivity may be improved by taking samples from more than one area.
The sensitivity of PCR tests improves over time, because the amount of virus an infected person carries increases until approximately a week after infection; then the accuracy diminishes because the amount of virus an infected person carries declines after its peak. An August 2020 analysis in The Lancet estimated a peak sensitivity of 95 percent six days after infection.
Controlling the pandemic will require both screening and diagnostic tests
The Harvard Global Health Institute has developed targets for how many COVID-19 tests are needed every day for an effective response, which puts the two types of testing in the context of their uses in public health. Under the Institute’s base targets, the US would require more than 4 million tests per day. A little more than half would consist of screening tests (rapid antigen tests) for people in schools, nursing homes, and jails and prisons, as well as first responders. A little less than half would consist of diagnostic tests (PCR tests) for anyone with symptoms and their contacts.
Under the Harvard Global Health Institute’s ideal target, the number of diagnostic tests would remain the same, a little more than 2 million. But screening testing would expand to cover health care, food service, and sales workers, and the testing frequency for K-12 teachers and students would increase. That would add nearly 10 million daily surveillance tests on top of the base target, totalling more than 14 million tests per day as an ideal.
While the US has been hovering around 1.5 million tests per day, the country can achieve the goal of conducting 14 million tests daily with the inexpensive antigen tests that are coming to the market. On their own, antigen tests aren’t a failsafe; because of their lower sensitivity, they can return false positives during the infectious period, so they have to be combined with precautions like distancing and masking to be effective. Their promise is to catch cases near peak infection, so that those testing positive can take even greater precautions for themselves and others.
More “Testing Data” posts
An understanding of state reporting schedules and day-of-week effects can help explain the reasons that numbers fluctuate from day to day, and what those changes mean.
Federal testing data is already more standardized and more detailed than the data we compile from states can ever be. But in some jurisdictions, state-provided data still provides a more comprehensive picture of testing volume. The federal government should address lingering quality problems so its data can become the national standard.
Although we’d prefer to have precise, real-time, comprehensive COVID-19 data for every US state and territory, the reality is that every metric that each jurisdiction publishes comes with quirks of timing and content that can make precise calculations impossible.