HIV and ID Observations

As the days grow shorter and we celebrate Labor Day here in the United States, the end of summer looms awfully near. With that will soon come colder temperatures, more time spent indoors, kids back in school, and the inevitable respiratory virus season.

How we address these “viral URIs” in the midst of the COVID-19 pandemic presents a major challenge for clinicians, patients, healthcare systems, and — the focus of this piece — diagnostic laboratories.

Because while certain symptoms might suggest rhinovirus or influenza or RSV or adenovirus or metapneumovirus more than SARS-CoV-2, there is no single clinical sign or symptom that would reliably separate one from the other.

Which means that essentially everyone with a viral respiratory tract infection is eligible for — and arguably needs — COVID-19 testing. This adds considerable volume to an already overburdened testing system that, in this well-argued piece by Atul Gawande in The New Yorker, is “as messed up as a pile of coat hangers.”

What I would propose is that we start by separating out symptomatic from asymptomatic testing. Let’s use a quick and less expensive antigen test for asymptomatic testing, and save the PCRs only for people with symptoms.

We have to do something. The demand for asymptomatic COVID-19 testing is already off the charts, increasing all the time.

One of my colleagues mentioned to me that testing people without symptoms is at least 40% of our hospital’s testing volume. Most of these tests are for admissions (all get tested) and for pre-procedure tests, but increasingly also for travelers re-entering Massachusetts or visiting Maine, people who want to see elderly family or friends, as part of school entry requirements, or just someone worried about a recent potential exposure.

But, you note, aren’t these antigen tests notoriously inaccurate? Won’t they miss cases because they aren’t as sensitive as PCR? Indeed, something of a backlash on rapid home testing appeared recently in a New York Times piece (one I happen to disagree with, for the record), and this lack of sensitivity was highlighted:

Experts also noted that antigen tests aren’t great at sussing out small amounts of the coronavirus, which means they’re far more likely to miss a case that a technique like PCR would catch.

For testing of asymptomatic individuals, it’s worth addressing this concern head-on. Because over the last week, I’ve been asked several times to explain why tests for COVID-19 could be different in symptomatic versus asymptomatic people.

Once it was to a group of research scientists.

Once it was to a local news reporter.

Once it was to a gastroenterologist.

Once it was to a bunch of friends in our backyard, during a socially distanced gathering.

All were worried about lower sensitivity of non-PCR testing.

From this diverse group of people, we can conclude that the concept of accepting a less-sensitive test for testing people without symptoms deserves further clarification — it’s a tough one to master, employing the scary-sounding concepts of Bayes’ Theorem.

So let’s get out our #2 pencils, or our handy Bowmar Brain, and do the math. And, following up on a presentation I made this past week on the topic, and a piece co-authored last month with my colleague Dr. Jeffrey Schnipper, we’re going to consider a very simple case.

(The nice editors at STAT wouldn’t let us publish the full math, except in a linked appendix. Since this is my blog, I can write what I want, so am including it here.)

Here’s the case:

August 2020, Boston.
41-year-old woman planning her summer vacation.
Business consultant; has been working remotely since mid-March.
Lives with husband and 2 children, ages 13 and 9.
Completely asymptomatic; everyone in the household well.
Needs testing for COVID-19 before entering the state of Maine.

What are the chances this woman has a contagious infection with SARS-CoV-2? (She has no symptoms, so that’s the “disease” we are testing for.) We’ll call this estimate our pre-test probability — it’s the likelihood a disease is present even before we do any testing.

In asymptomatic people in Boston currently, the pre-test probability is at most 1%. This is the positive test rate at our hospital now among people without symptoms. Now that we have that estimate, it’s math time!

• For 1000 people like this currently, 10 (1% of 1000) will have COVID-19, and 990 would have nothing.
• A test with 80% sensitivity will be positive in 80% of these 10 — or 8, missing 2 cases.
• The test will be negative in 992 people, which includes the 990 without COVID-19, plus the 2 with the infection we missed.
• The negative predictive value — which is how often the test correctly calls someone negative — is 990/992, or 99.8%.

The chance of missing an infectious case with antigen testing is only 2/1000 — and potentially lower since those who are most infectious have the highest amounts of virus, making false-negative results less likely. This 99.8% negative predictive value is plenty high enough for routine use in asymptomatic people, where the goal is detecting people who might be contagious without knowing it.

In fact, the big worry for testing asymptomatic people is the opposite — a false-positive result. Since false positives are so much more likely in a low prevalence population, all positive results will need confirmation by PCR.

But the take-home message from going through this exercise is that we should not hesitate to deploy “good enough” testing for screening low-risk people without symptoms. The pre-test probability is key to defining the trustworthiness of a test result.

And now you can cue the Jim Gaffigan-esque self-criticizing, high-voice stage whisper: Did he just to write about COVID-19 testing again? Can’t he write about anything else?

Hey, last week I wrote about reinfection!

Take it away, Jim — we need you now more than ever!