How rapid can coronavirus tests get?

“The ideal [test] is something right at the bedside, or the person wheels up and in a few minutes you can tell them if they have [coronavirus],” said Byron Berenger, a medical microbiologist who works at Alberta’s public health lab in Calgary.

But that ideal is far from the norm. The most reliable tests for the coronavirus detect the presence of the virus’s RNA (its genetic material), a process that takes more time than tests that detect, say, the presence of a hormone, such as in a pregnancy test.

The speediest test on the market is Abbott Laboratories’ ID Now test at five minutes for a positive result, but it’s been dogged by claims of inaccuracy. Cepheid and DiaSorin, whose tests were sent to High Level, have shown promise in striking a balance between speed and accuracy, but have other hang-ups.

Private companies and public labs racing to develop faster tests face obstacles like a high rate of false negatives, manufacturing limitations, and inexperience, leaving medical workers and patients to wait hours or even days for results, a lag that can impede containment strategies.

A rapid test is “better for the person, it’s better for managing them as a patient, it’s better for outbreak detection,” Berenger said. But “the important thing is diagnosing accurately, and then second is getting a quicker result that people can act on.”

Speed vs. reliability

In March, U.S. President Donald Trump stood before reporters on the White House lawn and unboxed the ID Now coronavirus test kit from U.S.-based Abbott that, according to the firm’s website, delivers negative results in 13 minutes and positive results in five.

Michael Reynolds—EPA/Bloomberg/Getty Images

That turnaround time makes ID Now the fastest coronavirus test on the market, but its reliability is questionable. In mid-May, New York University researchers found the test produced false negatives almost half the time. Abbott defended its test and criticized the NYU study’s methodology. Still, the Food and Drug Administration later issued an alert about “possible accuracy concerns” with ID Now.

Some degree of tradeoff between speed and accuracy is to be expected, says Mike Catton, deputy director of the Doherty Institute for Infection and Immunity at the University of Melbourne. “Generally, the most rapid tests make some degree of sacrifice, mainly in the sensitivity of detection,” he notes. “It’s a balancing act that scientists are incredibly cautious of.”

Gold standard

The most reliable test is the reverse transcriptase polymerase chain reaction (RT-PCR), which Catton calls the “gold standard detection method” for the coronavirus. Many companies and most public labs use RT-PCR.

The tests have a 96% accuracy rate for negative samples and a 100% accuracy rate for positive samples, according to researchers at the Foundation for Innovative New Diagnostics, a global health nonprofit.

But it takes time.

Extracting virus RNA from the sample—a necessary step for RT-PCR—eats up 30 minutes, says Matthew Lalli, a postdoctoral research scholar at Washington University School of Medicine in St. Louis.

The extracted RNA is mixed with a cocktail of reagents and heated up, which takes 15 minutes. Then it goes through signal amplification: 45 one-minute cycles of PCR, which makes billions of copies of the original specimen and provides a large enough sample to detect the virus.

The high sensitivity of the test depends on the amplification process; cutting that step would mean potentially compromising the chances of detecting the virus, and, therefore, the test’s accuracy.

Alberta’s PCR test takes 90 minutes, start to finish, but the time it takes to collect the sample, transport it to the lab, and accumulate enough individual specimens to start one round of the machine, which runs in bulk batches, results in an 18-hour turnaround at Berenger’s Calgary lab, the scientist says.

Striking a balance

The Cepheid and DiaSorin tests sent to High Level are promising in that they automate or eliminate certain steps required for the standard test, saving time. Cepheid’s test takes anywhere from 30 minutes to 45 minutes, and DiaSorin’s test takes “a little more than an hour,” its website says; it forgoes the “traditional RNA extraction steps” in RT-PCR, which shaves off 30 minutes.

They’re accurate too: Studies have found the Cepheid and DiaSorin tests to be at least 95.8% and 89.3% accurate, respectively.

But two factors keep the tests from being a silver bullet; price and supply. One of Cepheid’s PCR machines costs $17,500, not including the price of the cartridges needed to run them, according to the Foundation for Innovative New Diagnostics. And they’re scarce; the manufacturers haven’t been able to keep up with the global spike in demand, Berenger says. Neither company responded to Fortune’s requests for comment.

Faster tests also tend to have a lower throughput capacity, meaning they can’t simultaneously process as many samples as the slower RT-PCR tests. Abbott’s ID Now processes one sample at a time, as does the 45-minute Cepheid test, while standard RT-PCR tests feature well plates that can process 96 or 384 samples at once.

Amped up

Lalli is one of several researchers working on another testing method called loop-mediated isothermal amplification (LAMP) that is similar to PCR but does not require the resources of a centralized lab.

Lalli and his colleagues are building a portable device that would let patients process a coronavirus test in a clinic or at home and would deliver results within an hour. The LAMP test is “most similar” to Abbott’s ID Now, Lalli says, but “we are taking great efforts to improve the sensitivity of viral detection.” Lalli’s research is still in the preliminary stage, a preprint of the paper notes, but the results are “promising.”

LAMP saves around 30 minutes compared with RT-PCR, Lalli says. The reverse-transcription and the amplification happen simultaneously. Plus, the reaction is isothermal, meaning it occurs at a constant temperature rather than cycling between heating and cooling, a process that adds time in the standard PCR procedure. The ID Now test is also isothermal.

Some companies are developing tests that look for the presence of antibodies (immune response proteins) or antigens (virus proteins) instead of nucleic acid, which is what the conventional PCR tests, LAMP, ID Now, and the rapid test kits all detect.

Antibody and antigen tests are typically run as lateral flow immunoassays, says Lalli, and—like pregnancy tests—come as a dipstick. Using a blood or spit sample, they give a visual indication of whether the person has the coronavirus. They’re easy to use and deliver results in five to 10 minutes.

But they have shown “really poor sensitivity” compared with PCR, Berenger says. “We’re talking under 80% [accurate]…So you’re missing 30% to 40% of cases sometimes.”

The lateral flow method skips the vital amplification process of the PCR, which saves time at the possible expense of sensitivity, Lalli says. There are antibody and antigen tests with signal amplification (a method called ELISA), but these take one to two hours.

Lax regulations

Regulations for diagnostic tests are relatively lax in most parts of the world, meaning there are “hundreds on the market” that haven’t gone through regulatory approval, Berenger says, and many manufacturers are not experienced in making tests for respiratory viruses.

The FDA barred 29 antibody test manufacturers from selling their products in the U.S. last week, a change of tack from April, when the FDA said it would allow manufacturers to market their antibody tests, including those it had not yet reviewed. Also in April, the U.K. paid $20 million to two Chinese companies for antibody tests that didn’t work.

There are more reliable antibody tests, but they measure if you’ve had the virus and recovered, so they aren’t helpful in detecting active carriers of the disease who could be passing it on to others.