Needed: Different Strategy for COVID-19 Testing

The several different plans to reopen the economy share a common element — more testing to identify who has or is at risk for infection, and to help determine who can safely return to work. But, we still lack a coherent national testing strategy. Continuation of what we are currently doing will fail to identify some of the most crucial information: what proportion of the population is currently infected and whether the tests offer meaningful information about safely reintegrating previously infected workers. We can do it better. Here’s how.

Different Tests, Different Limitations
There are two types of tests. PCR-based tests use samples collected using swabs of the nose or throat. Some patients can have positive tests from the nose but negative tests from the throat (and vice versa), and some patients with active infection can have negative tests at both sites. [Ed.: After this article was submitted and accepted, the FDA authorized a viral protein antigen test.] A different type of test is used to identify the presence of antibodies in the bloodstream. IgM antibodies suggest a current infection. They rise after a new infection and then decline. IgG antibodies imply a past infection. They increase a couple of days to a few weeks after IgM and remain elevated, but it is unknown for how long. Nor do we know if such antibodies confer protection against reinfection. Not all patients infected show positive antibody tests. Further, some positive antibody tests can be falsely positive due to exposure to other viruses.

Who Should We Test?
Paul Romer, a Nobel laureate economist, proposed testing every person in the population an average of 7 times over 500 days. But, is testing everyone feasible or desirable? There are several reasons why the answer is “no.” First, it simply is not feasible. The current testing capacity has expanded greatly to about 200,000 tests per day. An ambitious goal is testing three-quarters of the population. If we could double the current capacity to 400,000 tests per day, it would take more than 21 months, or 625 days to test 250 million people a single time. A 21-month long program does not tell us what to do today, nor will it necessarily be instructive 21 months from now.
A second challenge is that clinical doctors and public health agencies have different testing goals. Doctors test patients to make an individual diagnosis; public health agencies need to understand population prevalence and the dynamics of disease transmission in communities. They do this not by testing everyone, but rather by testing a scientifically determined reference sample of people from the population, whether or not they have symptoms.

There are several reasons why trying to expand on our current strategy will fail. First, swabs and reagents are often unavailable. Second, clinical testing focuses on people with suspicious symptoms. They were tested for individual diagnostic reasons and are not a representative sample from the population. The fraction of people who test positive in clinical settings does not provide an accurate picture of the percentage of people who are infected now. So far, less than 1% of the U.S. population have tested positive for the virus. The real number of positives is undoubtedly higher. How much higher? We can’t say with any confidence because testing only those with symptoms will never offer unbiased estimates of how many people in the community at large are infected.
Monitoring the Epidemic
From a public health perspective, trying to test everyone would be a foolish waste of resources. There is a simpler, less costly, and entirely feasible alternative. Public health practitioners use sample surveys to estimate population rates. Similar to political polls, we can test randomly selected people stratified by a few variables in order to learn how many have been infected, or are currently infected, and among them, how many have symptoms, or not.

The value of diagnostic tests is typically described in terms of sensitivity (false negatives) and specificity (false positives). We have surprisingly little information about the sensitivity and specificity of our current and emerging tests. The one-peer reviewed study that has received the most attention is based on only 205 patients in China. In that evaluation, nearly 40% of people known to be actively infected with the virus tested negative based on nasal swabs. To be fair, our colleagues argue that the new tests used in America are much more accurate. Are they? One evaluation suggested that a widely touted new test failed to detect infection in about 15% of the cases. Similar reports have emerged from clinical experience. Because of the urgency associated with the COVID emergency, the FDA has waived the requirement for substantial evidence of test accuracy before the tests could be marketed to the public. Emergency authorization was given to 85 different companies or laboratories. In response to concerns in Congress, FDA has now reversed course and is asking companies to provide more quality evidence. But the cat is already out of the bag. More than 85 different tests are out there, assuring that results from different communities are not directly comparable.

Another reason we need smaller random-sample studies is to evaluate the accuracy of the tests. Information on the accuracy of the tests and the prevalence of the disease are linked. When the prevalence of a disease is low, even small imperfections in test accuracy lead to substantial numbers of misdiagnoses.
For example, if 1% of the population has the infection, and we use the specificity estimates from the Chinese study, about five well people will be incorrectly told they have the disease for each one person who is accurately detected as being ill. In other words, for each person that we justifiably quarantine, five uninfected people might be restricted. Even if the sensitivity of the test is 95%, and the specificity is 97%, about three people will be falsely told they are infected for each one who is correctly detected. The reason – with a prevalence of 1%, 99% are uninfected. A very low error rate is applied to a very large number of uninfected people.

As the background prevalence increases in the population, the chance that a positive test result is actually a true positive becomes more likely. It is not until the prevalence rate reaches 5%, or about quintuple the current estimate, that the number of false positives reaches the level of true positives.
Still, a balance between false and true positives is miles from a reasonable goal. For a test to be valuable, we would hope to have at least 20 true positives for each false positive. More disturbing is that our current testing strategy is not designed to tell us how many people are infected. Without knowing that number, accurate estimates of test quality are impossible.
Move Both Fast and Not So Fast
We are in an urgent situation and we need to take action. Given what we know now, attempting to test everyone may cause more problems than it solves. A feasible national strategy might have three components.

First, to determine which people with symptoms should be treated and quarantined, diagnostic tests should continue to be used by clinical facilities.
Second, it is necessary to divert an adequate portion of tests for studies of the general population. We don’t need to test everyone, just the minimal number to estimate how often the virus is present in people without symptoms (diagnostic test) and how often people may have previously been infected (antibody test). To assure enough tests are available, asymptomatic people or those with minor symptoms might be asked to defer testing and to self-quarantine.
The third component of the strategy is a nationally coordinated research effort. Currently, we don’t know how many people are infected, what fraction of infected people will die, or whether someone who has been infected and has antibodies is immune to getting the disease if they encounter the virus again. We need a strategy to figure it out so that we can prevent more illness and death and get Americans back to work.

Time is slipping. Let’s not let more weeks pass before we will develop testing capacity to meet both clinical and public health needs.
Robert M. Kaplan, PhD, is a distinguished research professor of Public Health and Medicine at UCLA and a faculty member at the Stanford School of Medicine Clinical Excellence Research Center. He is a former associate director of the NIH and a former chief science officer at the Agency for Healthcare Research and Quality. Kevin Winthrop, MD, is principal investigator and director of the Center for Infectious Disease Studies at the Oregon Health & Science University. He holds positions as professor of public health at the School of Public Health and professor of infectious diseases and ophthalmology at the School of Medicine and is a former infectious disease epidemiologist at CDC. Michael H. Weisman, MD, is distinguished professor of medicine emeritus at David Geffen School of Medicine at UCLA and professor of medicine emeritus from Cedars-Sinai Medical Center. His research interests involve genetic and environmental triggers for chronic rheumatic diseases and as such he has developed a keen interest in epidemiologic testing and assessment methodologies.
https://www.medpagetoday.com/infectiousdisease/covid19/86450