Word came yesterday that Abbott received an Emergency Use Authorization for a new coronavirus test, one that is faster and cheaper than anything currently out there. The two types of tests that we see in use now are RT-PCR, the nasal-swab test that detects viral RNA, and various antibody tests, that tell you if you have raised an immune response due to past exposure to the virus. This one has features of both, but its main use is more like the RT-PCR test: it will tell you if you are actively infected. It does that by detecting a particular antigen, the nucleocapsid protein (Np) of the coronavirus. It’s a key part of its structure, and in an actively replicating infection you can be sure that there will be plenty of that one floating around.
The test itself is one of Abbott’s “BinaxNOW” assays, and they have a whole line of these already as tests for malaria, RSV, various bacterial infections, and so on. It’s a lateral flow assay, which will be familiar to anyone who’s seen a pregnancy test, and I explained the general principles of those (as antibody tests) in this post. This new test is a sort of flipped version of what I described there, though. In this case, a nasal swab is taken, and several drops of solvent are used to put that sample onto the beginning of the absorbing strip inside the card. As it soaks up along the length of the strip, the sample will encounter a zone of antibodies that recognize the Np antigen, and these antibodies are also attached to nanoparticles of gold. This gold-antibody-Np complex is carried along in solution further along the strip until it runs into another antibody zone, one that’s immobilized on the solid support and which will bind the gold-antibody-Np complex molecules tightly. That stops them in their tracks and allows the gold nanoparticles to pile up enough to be visible as a pink or purple line. Along the way, the sample has also crossed a zone containing another soluble gold conjugate species as a control, which gets carried along until it runs into another separate zone of immobilized antibodies specific to it. The presence of a pink control line means that the test has been performed correctly; absence of such a control line means that the whole test has been messed up somehow and needs to be run again.
I had described earlier a test that looks for antibodies to the coronavirus by running them past gold-conjugated antigens on the test strip, but this one looks for antigens by running past gold-conjugated antibodies. Developing a test like this involves a lot of work to find the right antibodies, to make sure that they’re attached to the gold nanoparticles in ways that don’t impair their function, to find the right second immobilized set of antibodies that will develop that test line, and to make sure that the control line system is compatible with the test itself. You’ll also need to work on the composition of the test strip and the solvent that’ll be used to take the patient’s sample into it: these need to allow as much of the antibody complex to flow down the strip in a controlled fashion as you can get, and to do so in the same way every time. And finally, you need to validate the assay with a lot of coronavirus patients and controls, to see what your false positive and false negative rates are.
For this assay, those come out to a sensitivity of 97.1% (positive results detected when there should have been a positive) and a specificity of 98.5% (negative results when there should indeed have been a negative). Flipping those around, you’ll see that about 1.5 to 3% of the time, you will tell someone who’s infected that they’re not, or tell someone who’s not infected that they are. That’s about what you can expect for a test that sells for $5 and takes 15 minutes to read out with no special equipment, but such tests (if used properly) can be very valuable. Flipping that around, you can also infer that if used improperly, they can be sources of great confusion.
What’s proper? The FDA’s EUA is for testing people that show up with symptoms to see if they really do have SARS-CoV2. I think that’s appropriate, because you’re more likely to have a higher percentage of those folks who are really infected. If you tried to deploy this test across a large asymptomatic population with a very low true infection rate – everybody in New Zealand, for example – you would create turmoil. New Zealand’s real infection rate is vanishingly small, but Abbott’s quick $5 test would read out a false positive You Are Coronavirused for 1.5% of the whole country, never lower, which would be a completely misleading picture that would cause all sorts of needless trouble.
On the other hand, if you’re testing symptomatic people in a community where the virus is already known to be spreading, you can do a huge amount of good. Let’s imagine you test 1000 such coughing, worried patients under conditions where you expect that 10% of them really do have the coronavirus. In the course of testing all thousand, you’ll run those 100 positive folks through, and you’ll correctly tell 97 of them they they need to go isolate themselves immediately, which is a huge win for public health. Three of them, unfortunately, will be told that they’re negative and will go out and do what they do, but that’s surely far fewer than would be out and around without the test. You’ll also run the 900 other people through who actually have a cold or flu or something and not corona, and you’ll tell maybe 13 of them (900 x 0.015) that they’re positive for coronavirus and that they should isolate as well. That’s not great, either, but it’s worth it to get the 97 out of 100 real infectious coronavirus patients off the streets. And meanwhile you’ve correctly told the other 890 people in your original cohort that they do not have coronavirus, which is also a good outcome. But remember, with that 98.5% specificity you’re going to send 15 people out of every thousand you test home to quarantine even if no one really has it at all. If 1% of your sample of 1000 people is truly infected, you’ll probably catch all ten people who are really positive. . .but you’ll also tell 14 or 15 people who don’t have it that they do, crossing over to finding more false positives than there are real ones.
And let’s not forget the other really good aspects of this test: it’s cheaper than anything else out there but best of all, it’s fast. The delays in the RT-PCR testing have been killing its usefulness in too many cases – what good is knowing that you tested negative sometime last week, really? Far worse, what good is knowing that you tested positive last week if you didn’t isolate yourself because you weren’t sure if it was the coronavirus or not? But an answer in fifteen minutes, that’s actionable. As long as this test is deployed correctly, it can be very useful.
Addendum: I’m well aware that the CDC seems (controversially) to be changing its testing recommendations in general. This “only test if there are symptoms” guidance seems to apply to RT-PCR testing as well – and turnaround problems aside, that test still has far higher selectivity and specificity than this new 15-minute one and is far more appropriate for use in a broader, largely asymptomatic population. We need to be addressing the delay problems in RT-PCR, because we need to be doing a lot of those tests – not closing our eyes and whistling a happy tune instead. This appears to me, and to many others, to be political interference from above. What else is one to think when administration officials have suddenly started referring to the pandemic in the past tense? So here’s something I never pictured myself saying: it is my hope that this CDC guidance will be ignored. It’s a hell of a situation to get to, isn’t it? Update: I am very happy to report that the CDC appears this morning to be walking this one back. Good.