Michael T Osterholm and Mark Olshaker have one of the rare illuminating and wise opeds in NYT which pierces through the noise surrounding SARS-CoV-2 testing and gets to the heart of the issue. The headline summary is very blunt,
Let’s Get Real About Coronavirus Tests. There aren’t enough. Many are shoddy. Most aren’t even designed to tell us what we really want to know.
He outlines the problems with the virus test,
The reverse transcription polymerase chain reaction (RT-PCR) test, diagnoses SARS-CoV-2 infections by analyzing cells collected from the nose or back of the throat. It converts the cells’ RNA into DNA and then, using polymerase enzymes, duplicates the DNA again and again, so that there’s enough of the virus that it can be detected, if it is present at all. This process is known as “amplification.”... The accuracy of RT-PCR tests is inherently limited. The U.S. Food and Drug Administration recommends 40 cycles of amplification, but even after those, too little of the virus’s genetic material might be present to be detectable. One consequence is that even when diagnostic tests aren’t faulty and they are performed properly, some people who test negative for SARS-CoV-2 actually are infected — a reading known as a “false negative.” In a recent study by researchers at the Cleveland Clinic of five commonly used diagnostic tests, nearly 15 percent of the results were false negatives. Chinese scientists published a study in February that found the false negative rate of some tests conducted at the Third People’s Hospital in Shenzhen, southern China, between Jan. 11 and Feb. 3 was as high as 40 percent. An article published earlier this month in Mayo Clinic Proceedings cautioned that “even with test sensitivity values as high as 90 percent” (really rather precise), the danger posed by false negative results — that is, the health risk created by infected people mistakenly being told they are infection-free — was significant and that it would only increase as testing increases overall.
And with the antibody test,
The second kind of test is serology, which detects the presence of antibodies to the virus in the bloodstream. Antibodies are evidence of the body’s reaction to an infection, of the fact that a person was previously infected; their presence might also suggest that the person is now immune to the virus. We say “might” and “suggest,” not “prove,” because the notion that immunity to SARS-CoV-2 can be acquired through infection is only, for now, an assumption based on past experience with other viruses. No scientific studies have confirmed this hypothesis yet. Scientists worldwide are working to determine if in the case of SARS-CoV-2, too, infection confers immunity, and if so, how effectively and for how long. But the first serological studies made public to date have been flawed or too easy to misinterpret...
As for the blood work itself, serological tests, like RT-PCR tests, have inherent limitations to do with accuracy. Even the most precise antibody tests don’t produce neat, binary results. Measuring antibodies isn’t like determining if a light has been switched on or off; it’s more like gauging the intensity of a bulb controlled by a rheostat. One example: In the early days of an infection, while a patient’s immune system is still revving up, their antibody levels might be too low to detect. Serological tests also suffer from an internal contradiction, a structural tension. A very precise test is able to correctly identify both the presence of any antibodies if they are present (this is known as “sensitivity”) and the absence of antibodies when they are not there (this is “specificity”). But sensitivity and specificity are somewhat at odds with each other, and they compete. For instance, the characteristics that make a test more sensitive, or better at turning out true positives, also make it more likely to yield false positives instead of what should be true negatives.
At the same time, it is also a principle of epidemiology that the lower the prevalence of an infection in a studied population, the greater the chance that testing for antibodies will yield false positive results. (That’s because when testing in a population with few total cases of infection, the number of false positives will make up a larger share of all positive results.) And the consensus among the leading epidemiologists and clinical-lab experts we talk to regularly is that, to date, only between 5 and 15 percent of the population of the United States has been infected with SARS-CoV-2. These features are one reason an April 17 advisory from the F.D.A. recommending the use of serological tests simultaneously warned that the agency “does not expect that an antibody test can be shown to definitively diagnose or exclude SARS-CoV-2 infection.” Last Friday, the World Health Organization released a scientific brief that said, “There is currently no evidence that people who have recovered from Covid-19 and have antibodies are protected from a second infection.”
Instead he suggests good old clinical screening,
... for as long as testing for SARS-CoV-2 is too limited or unreliable, the United States must ramp up what public health professionals call “syndromic surveillance”: the practice by medical personnel of observing, recording and reporting telltale patterns of symptoms in patients so that local health authorities, mayors and governors can anticipate and plan for the likely spread of a disease. This system, supported by funding and technical assistance from the Centers for Disease Control and Prevention, has been put in place for seasonal infectious diseases like influenza and are currently being used to track Covid-19 symptoms. It should be expanded to include even more reporting locations.
His cautionary note is well worth repeating for models, testing, drugs, vaccines, and the like.
Informing the public involves clearly acknowledging what is still not known about this virus, and it involves stating what tests simply cannot do. It also means accepting this painful paradox: We turn to testing in the hope of managing the pandemic, but testing won’t get better until the pandemic gets worse.
The idea of symptomatic screening is even more relevant for developing countries, where the idea of large-scale testing which is meaningful enough (beyond the limited group of foreigners, contacts, and suspected cases) is simply impractical. Using local data and layering it with AI to figure out robust symptomatic screening is an area worth investing in.
As an illustration of the noise, see this latest misleading story in CNN on the three types of tests.
Update 1
This is a good summary of the problems with false positives and false negatives.
Recent data suggests that approximately 15% of all tests conducted in the United States are returning false negatives, meaning that of every 100 individuals infected with COVID-19, 15 of them are told they don't have it. If we were to have 1 million infected individuals in the country, and every one of them got tested, 150,000 possibly contagious individuals would test negative and be given the "all-clear" to neither quarantine nor self-isolate. Even if testing were perfect — if everyone in the country were tested on even a biweekly or weekly basis — a test with such high levels of false negatives would be unacceptable. Although there are many epidemiological reasons why this is true, it's easy to illustrate with one simple example: the idea of a superspreader event.
No comments:
Post a Comment