Katie Martin / The Atlantic

The Fog of the Pandemic Is Returning

Millions of coronavirus tests may be happening without their results being made public.

President Donald Trump has never hidden his ambivalence about testing for the coronavirus. In June, when he told an arena of supporters in Tulsa, Oklahoma, that he had instructed “his people” to “‘slow the testing down, please,’” the disclosure prompted one of the more dire news cycles of the pandemic. The president said repeatedly that he wanted the United States to reduce its testing. But in the weeks that followed, testing increased.

Not so now. In the past month, the number of tests conducted in the United States has actually drifted down—and that may be partly because of Trump-administration policy.

The United States now reports about 100,000 fewer daily tests than it did in late July, according to the COVID Tracking Project at The Atlantic. Some of this decline is due to reduced demand: The surge of infections across the South and West has subsided, and when fewer people are sick, fewer people seek out tests. Yet this cannot explain all of it. In the Midwest, the number of confirmed cases is growing faster than the number of tests, which has been a sign of a growing outbreak throughout the pandemic.

The decline in reported tests has come just as other changes have hit the testing system. In recent weeks, the Trump administration has taken unprecedented steps to interfere with guidance from the Centers for Disease Control and Prevention. As a result of White House meddling, the CDC now recommends against testing asymptomatic people, the group that may spread the virus the most. At the same time, new antigen-testing technology is rolling out nationwide. While quicker tests in greater numbers should help curb the virus, they are also decentralizing data collection.

So far, the U.S. has reported only about 200,000 antigen-test results. But some evidence suggests that these tests are being used on a much wider scale than is understood: Thousands, if not tens of thousands, of antigen tests may already be happening every day without their results appearing in any public data. Just as dark matter can’t be observed directly, even though it makes up much of the universe, this “dark testing” does not show up in the data but may already account for a substantial chunk of the coronavirus testing done in the U.S.

The result of these changes is that some once-trustworthy numbers and measurements—such as the number of tests conducted in each state, and the percentage of tests that come back positive—now seem less reliable. Over the past months, as states have developed their testing systems, the picture of the pandemic clarified, but now it is blurring again.

In the immediate future, antigen tests could roll out nationwide, allowing health-care workers to catch outbreaks before they erupt. Or the national testing system could deteriorate further under White House pressure, meaning that states and cities might realize that an outbreak is growing only when hospitalizations bloom. As we stand at this crossroads, still confirming tens of thousands of cases a day, the shape of the pandemic is going to change again, and we may not have the tools we need to see how.

The first major change to beset the testing system is entirely because of Trump. Two weeks ago, the CDC changed its official guidance about when Americans should get a coronavirus test. The agency had once maintained that everyone who was exposed to the virus should get tested for it. Now it altered this advice: If someone was exposed to the virus but did not yet have symptoms of COVID-19, they did not necessarily need a test, the guidance said.

The edit was small but significant. It also made no sense. Scientists have known for months that people can spread the coronavirus before they develop symptoms of COVID-19, and some evidence suggests that truly asymptomatic people—that is, those who are infected but never develop symptoms—may be more efficient spreaders than anyone else. The only way to distinguish between a healthy person and an asymptomatic person who has COVID-19 is to test them. But this is exactly what the CDC now recommends against.

“To say you don’t have to test asymptomatic people—while knowing at least half of infections are driven by asymptomatic people—is idiotic,” Kristian Andersen, an immunology professor at Scripps Research, told us.

This change in guidance did not originate inside the CDC, according to CNNand The New York Times. Instead, the change was imposed on the agency by the White House, acting on the advice of Scott Atlas, a neuroradiologist and conservative policy wonk who has started to advise Trump on the pandemic. Atlas fought with Robert Redfield, the CDC’s director, over the new policy, according to the Times.

Atlas, who has no background in infectious disease, has advocated for a so-called herd-immunity approach, asserting that the federal government should protect only the elderly and the most vulnerable from COVID-19. This would lead to many more American deaths—Sweden, which pursued a similar policy, has a higher case-fatality rate than the United States, the European Union average, Iran, and more than 100 other countries, according to data from Johns Hopkins University—and it may not even work. But more important, herd immunity has nothing to do with testing. There is no reason that advocates of the herd-immunity approach should oppose testing, unless their goal is to let an outbreak spiral beyond control before anyone notices, Andersen said. “If you don’t test [asymptomatic people], you have a lower reported number of cases, but you end up with more cases overall. And you end up with more deaths and more hospitalizations, which you can’t hide, because you lose control of the virus.”

The change to the CDC guidance is not the only disruption of the testing landscape. In the past month, doctors and hospitals have started to use faster but less sensitive tests to look for the coronavirus. Unlike the gold-standard PCR tests, which detect genetic material from the virus, these tests look for the presence of chemicals, called antigens, that make up the virus. As we’ve written, these antigen tests will be a crucial tool in defeating the pandemic, because they will let offices, nursing homes, and other semipublic places identify contagious but asymptomatic people before they spread the virus.

We believe that dark testing is happening, because we see a hole where data about antigen testing should be. Millions of antigen tests are now being manufactured every month. Quidel, a $6 billion company that makes one of the most widely used antigen tests, says that it began producing at least 1 million tests a week earlier in the summer. In recent days it has upped that rate to nearly 2 million. “We don’t have any inventory,” Doug Bryant, its chief executive, told us. “We ship every day with what we have.” Becton Dickinson, which makes a competing antigen test,  has predicted that it would be manufacturing 2 million tests a week by the end of September.

Some federal agencies have made these tests central to their national strategy. In August, the Centers for Medicare and Medicaid Services announced that it would buy antigen tests from Quidel and Becton Dickinson, for nursing homes nationwide. Estimates calculated from agency data suggest that it will distribute 2 million to 4 million tests to more than 13,000 nursing homes by September 30.

Yet these millions of tests are missing from the public data. Only six states, representing 50 million people, make separate antigen-test data readily available. Those data show that a mere 215,000 antigen tests have been reported since early August, when they first appeared on state dashboards. Even if the data are taken as representative of the U.S. as a whole, and scaled accordingly, they imply that only 1.4 million antigen tests have ever been conducted—far fewer than the number of tests that companies have shipped since June, which is on the order of tens of millions.

Even though the Department of Health and Human Services has spent tens of millions of dollars distributing tests, it could not tell us how many of the tests have been used. The agency has said that it is aware of the reporting issue, and in late August, it threatened to fine nursing homes that do not report test results accurately. (The department did not respond to multiple requests for comment.)

In some ways, a small data gap is not surprising: Data about antigen tests are virtually guaranteed to be spottier than data about PCR tests. Antigen tests are conducted and analyzed in the same places where they’re used: nursing homes, doctors’ offices, and schools. PCR tests, meanwhile, must be analyzed at a central lab or hospital. Because labs and hospitals regularly report large amounts of data to public-health agencies, but schools and nursing homes do not, PCR data will almost always be more complete.

But this dark testing is missing from other places you might expect it to show up. For instance, under CDC rules that define who has a “case” of COVID-19, a person who tests positive on an antigen test is said to have a “probable case.” If antigen tests were flooding the market, states would report hundreds of thousands of probable cases. Yet again, there’s a gap: Most states do not report probable cases as a separate category. Of the more than 6.3 million COVID-19 cases reported in the United States, only 80,000 are “probable.” The antigen tests are missing here, too.

The dark-testing problem is certain to get worse. Quidel and Becton Dickinson say they will produce about 4 million tests a week, combined, by the end of September. Quidel is “building towards 5 million tests a week next year,” Bryant said. A third company, Abbott Laboratories, claims that in October, it will begin producing 50 million of its cheaper coronavirus tests a month. Abbott’s prospective volume alone is more than double the number of PCR tests ever conducted in a month nationwide; it means that dark testing would encompass as many as two in every three American coronavirus tests conducted by the end of the year.

All of this is to say: Antigen tests are being produced by the millions and showing up in our testing data in small numbers that require major and unreliable extrapolation.

What’s actually happening? There are multiple possibilities: First, perhaps only tens of thousands of antigen tests have actually been conducted, even though millions exist. For now, most antigen tests require a desktop machine, so the throughput for any individual location is limited. “You can run eight Quidel tests in the same amount of time you can run a full PCR plate of 384” tests, Andersen, the immunologist, said. “I would assume these tests would be dwarfed by PCR capacity—I would hope.”

Second, perhaps most test locations are simply not reporting test data back to state authorities; few point-of-care locations are set up to report these data electronically, so the hassle factor is high. Third, some states may be lumping antigen tests in with their PCR testing or case numbers. Researchers at the COVID Tracking Project at The Atlantic think at least 10—and perhaps as many as 30—states are lumping antigen tests in with PCR tests, which would hide them from our analyses. Finally, probable cases might not show up in the data, because nearly everyone who gets a positive antigen test is quickly retested by PCR. That’s what happened to Ohio Governor Mike DeWine, who tested positive for the virus on a rapid antigen test last month, then tested negative for it when retested by PCR later in the day.

Of course, some combination of factors could be at play simultaneously.

This is the current state of pandemic tracking: It is difficult to know whether millions of tests have been conducted at all. And if the data system is missing so many tests now, when maybe hundreds of thousands of antigen tests are being done each week, imagine what this will look like in a few months, when 1 million antigen tests might be completed each day. There is little doubt that we will lose the ability to track the large majority of tests completed in the country.

That could ultimately be a good thing, because it would signify that tests are so ubiquitous—and such a regular feature of everyday life—that they no longer need to be tracked as systematically. But because that change has come now, just as officials within the Trump administration are acting to suppress testing as never before, it has introduced uncertainty and disorder. If the number of tests completed each day continues to fall in the data, what will that reveal? It could mean that every American who wants a test can get one. Or it could mean that the president has finally succeeded in reducing testing.

This story was originally published by The Atlantic. Sign up for their newsletter.