The Elephant in the Room (2/2)
Open Question: How Many SARS-Cov-2 infections go undetected?
A friend took me to task — tusk? — over the elephant.
“The elephant in the room,” she pointed out, “is the thing everybody knows, but is afraid to talk about.”
“Whereas Covid’s asymptomatic rate,” she continued, “is something everybody is talking about, and nobody knows.”
She can be annoying like that.
In my defense I opened the current issue of The New Yorker, to an interview by Issac Choiter of Justin Lessler, professor of epidemiology at Johns Hopkins Bloomberg School of Public Health.
In his very first answer, Lessler said:
I had been hoping that we would have more of a sense of exactly how many infections there have been out there. I think that still remains the biggest unknown of the whole thing.
“See?” I said. “Biggest unknown? Whole thing? Like an elephant.”
But since the elephant came in the door for Part I, she’s back for Part II. Elephants in the room are like raccoons in the attic: hard to get rid of.
I promise I will get to the blood serology results that came in from Telluride last night and Stanford today. Very soon.
But I got distracted by some recent twists and turns in the testing story. This, against my will, turned into an essay.
When I turn on the news, suddenly even the politicians want more testing.
They think it will help reopen the economy. Get us back to normal.
It won’t certainly won’t hurt.
But I think they have no idea of the scale that’s needed.
The slow start with testing in January-February was a blunder that by now, April, has already — past tense — cost lives. Six weeks with the disease. Those people are dead by now.
But there is finally a lot more testing going on in the U.S. It is an accomplishment.
Americans can be counted on to do the right thing, after exhausting the alternatives.
“Total Tested in the US” has made it onto the Johns Hopkins dashboard, replacing the distressingly slow-moving “Recovered.” As of today, 17 April: the cumulative number of tests administered in the U.S. stands at 3,423,034. (That’s since the COVID Tracking Project started keeping count, March 3rd.)
There are the usual caveats about this number: does it include multiple tests given to the same individual? (a required thing to clear patient for discharge from a hospital). And we stop fixating on the cumulative total of tests, many of which are now weeks old and worthless, and display the number given per day?
Still, that’s a lot of tests, over 1% of the U.S. population.
Axios gives the number administered on 15 April as 146,529. (The U.K., by contrast, a smaller country to be sure, is struggling to build capacity to get to 100,000 tests per day by the end of the April.)
Before congratulating ourselves that our politicians finally seem to comprehend the importance of testing, we need to look at two odd bits of testing news that came up this week.
First, Axios reported the number of tests given per day in the U.S. has plateaued over the last week, and offers this chart:
(The top of the chart is a target of 500,000 per day.)
Axios blames lack of supplies, poor coordination, and restrictive rules about who get tested.
To which I would add “logistics” and a simple-minded observation that the tests are scary and to try to get one a pain in the… nose.
And perversely, lower testing may be an unintended consequence of lock-downs and social distancing. Like hospitals, testing sites are known to attract sick people. Who wants to go there?
Second, and more intriguing, is a rumination posted to The Atlantic’s web site on 16 April by Robinson Meyer and Alexis C. Madrigal, headlined “A New Statistic Reveals Why America’s COVID-19 Numbers are Flat.”
First, Meyer and Madrigal are correct, as always, in being criticizing our wrong-headed policy of prioritizing tests for incoming clinical patients, and not reserving a sizeable chunk of them for general population random sampling.
The mysterious 20 percent
The new statistic mentioned in the headline is the “test-positivity rate,” which hovers just under 20 percent.
(Modest guy that I am, I want to remind my readers that I wrote something about this two weeks ago, at the bottom of this post.)
The weird anomaly to explain is that this 20% positive held near-rock steady even as testing greatly increased.
It breaks your head to think about it in a cause-and-effect way. How could the incidence of new infections magically match the increase in the number of tests given?
If there were a finite pool of the afflicted , one would expect the ratio of positive tests to negative tests to eventually go down. I like Meyer’s and Madrigal’s “Jelly Bean” analogy:
If the U.S. were a jar of 330 million jelly beans, then over the course of the outbreak, the health-care system has reached in with a bigger and bigger scoop. But every day, 20 percent of the beans it pulls out are positive for COVID-19. If the outbreak were indeed under control, then we would expect more testing — that is, a larger scoop — to yield a smaller and smaller proportion of positives. So far, that hasn’t happened.
That said, they offer the “test-positivity rate” as a useful alternative for comparisons between areas, and it is. We should use it. It certainly came as a shock to me that 1 out of 2 people tested in New Jersey are coming back positive.
Also disturbing, Meyer and Madrigal suggest that the apparent “apex” or plateau (my cautionary post, BTW, was subtitled “If you look hard enough for an apex, you’ll see one”) in new positive cases per day may be an artifact of testing hitting a capacity limit, and so does not convey any useful information about the state of the disease in the wild. That could well be.
Yogi’s 20 percent
I can come up with two explanations for the Jelly Bean phenomenon. One makes some sense. The other I call a “Yogi Berra” explanation, because it sounds really dumb at first, but may have a hard kernel of common sense. (Do we have to worry about future generations getting his dismissive comment about a popular restaurant, “No one goes there anymore. It’s too crowded.”?)
First, don’t forget that the 80% of the Jelly Beans coming out negative (a good thing) are being tested for a reason (some bad thing). My first explanation is that what we’re seeing is the split in the general population between ILI (flu-like) symptoms and Covid-19 symptoms. These symptoms look alike — alike enough to fool doctors, which is why they order the test. So we a just drawing a dividing line inside a pool of people with “symptoms.”
This would also explain why, as flu season has waned, the test-positivity rate of Covid has creept up slightly. Covid is not up; flu is down.
The Yogi Berra explanation is more embarrassing, but hear it out.
We want to know the percentage of the general population that is infected with SARS-Cov-2. We test people and we test people but gosh, it always comes back around that blasted 20%…
Stop right there. Take Yes for an answer.
Twenty percent, 20%, as our true infection rate, the prevalence.
How do we know? We took a sample. A biased sample to be sure, but a hell of a big one. Any pollster would kill to have a sample size of 1 million.
If that’s our number, then other ones just fall out of the spreadsheet: The number infected in the United States: 66 million.
For scale, compare with the population of California, 40 million.
That’s a lot of Jelly Beans.
Using a recent number for confirmed cases, 640,291 on 15 April, the undetected rate is— ouch — 99%.
Meaning despite all the testing we’re doing now, we’re missing 99% of cases.
That may sound horrible. Icebergs are mostly below water.
Iceland, a small country with good health care and abundant free testing (3% of population), had its testing results studied by Harvard and MIT economists. They concluded:
Our primary estimates for the fraction of infections that are undetected range from 88.7% to 93.6%
So Yogi Berra may be in the ballpark.
Enough for now.
For new posts, follow me on Twitter.