The Covid-19 epidemic that raged through New York City in late winter and early spring is starting to feel like ancient history. The disease has mostly moved on to other locales. Life in the city is far from normal, but tons livelier than it was in March and April.
One of the many advantages to having the coronavirus in the city’s rearview mirror (for now at least) is that it’s getting easier to see how the biggest urban explosion of the disease in the US (so far at least) actually played out. Thanks to antibody surveys conducted by the state of New York and a study released in preliminary form last month by a large group of researchers at New York City’s Icahn School of Medicine at Mount Sinai, it is now possible to craft a rough estimate of how many people were infected with the disease in the city and when. Turns out it’s a lot different from the standard picture provided by confirmed-case counts!
Note the asterisk and footnote. The blue line is my estimate, and I’m a journalist who usually writes about business and economics. No epidemiologist or public health official would dispute that there were a lot more actual cases of Covid-19 in New York City than confirmed ones, or that the peak in new infections occurred before the peak in confirmed cases, but the exact ups and downs of the disease depicted above are the product of a bunch of assumptions that may not be entirely correct. I will describe those assumptions and their possible flaws below, but I’m convinced that this is nonetheless a big improvement over the confirmed-cases chart, which misses most early cases of the disease because there simply wasn’t enough testing capacity in March and early April. The very different picture of the New York City Covid-19 epidemic that my estimate paints can thus better inform our understanding of how the disease spread and what may have stopped it.
What enables such estimates is that the virus that causes Covid-19 leaves traces in the blood of those who recover from it. In one study of New York area Covid patients by Mount Sinai researchers, antibodies to the new coronavirus appeared in blood samples from 621 of 624 of those who had previously tested positive for the disease via nasal-swab tests, with a median time from symptom onset to antibody appearance of 20 days.
Some of those same researchers then tested blood samples from patients at Mount Sinai Health System hospitals in the city from mid-February through mid-April to compile weekly estimates of antibody prevalence among patients who had been admitted to the hospital from the emergency room and those there for other reasons. Even the latter group is not necessarily representative of the overall populace, of course, but the 19.6% of non-emergency Mount Sinai patients testing positive for antibodies during the week ending April 19 was close enough to the 22.7% prevalence estimated for New York City from a large state survey conducted April 19-28 that the Mount Sinai researchers concluded that their weekly measures did convey useful information about Covid-19’s presence in the city. A Centers for Disease Control and Prevention survey based on blood samples taken in late March in the New York metropolitan area also delivered results compatible with Mount Sinai’s.
Given the aforementioned 20-day median delay from the onset of symptoms to the production of antibodies and the 6-day median lag from infection to symptom onset estimated by the CDC, this implies that a whole lot of New Yorkers got Covid-19 in March. With a city population estimated by the Census Bureau at 8,336,817, the 22.7% prevalence from the late-April state survey would mean that nearly 1.9 million New Yorkers had already contracted the disease as of early April. The 10.1% prevalence estimated from the Mount Sinai blood samples for the week ending April 5 implies that 842,000 already had it as of March 10.
Then again, the 2% prevalence estimated for the week ending March 1 implies that 167,000 New Yorkers had already contracted Covid as of Feb. 4. This is incompatible with pretty much all other research on the early spread of the disease here, which mostly depicts it first arriving in late January or early February and taking off thereafter. So there is clearly error in these estimates, and the numbers derived from the early Mount Sinai blood samples in particular shouldn’t be taken too seriously because they’re within that margin of error.
Even the numbers from the bigger state surveys aren’t perfectly reliable. For one thing, the samples were collected at grocery stores, meaning that cautious sorts who had all their groceries delivered were excluded. The estimates have also shifted over time. New York Governor Andrew Cuomo initially reported the late-April New York City percentage as 21.2%, then revised it down to 19.9% after the survey was completed. The researchers at the state Department of Health and University at Albany who designed the survey subsequently revised that up to 22.7% for a paper published in the Annals of Epidemiology, which reflected their estimate of cases that were missed because the test they used turns up false negatives from time to time.
Cuomo has reported that a survey conducted from May 1 through June 13 found 21.6% of New York City residents tested to have the antibodies. There’s no scientific-paper version available yet, so I took the liberty of adjusting that up to an estimated prevalence of 24.3% using the same formula as the Annals of Epidemiology paper, in part to avoid describing a situation in which the number of people of who have had Covid-19 in the city went down over time. I wouldn’t count on that being exactly right, though, and it’s even possible that antibody prevalence is declining in New York, warned Eli Rosenberg, the University at Albany epidemiologist who was lead author of the Annals of Epidemiology paper. New research emerging from China, Spain, the UK and the US. indicates that antibody levels can fall quickly in people with mild and asymptomatic cases of Covid-19. “This news is going to (and is starting to) rock the world of us folks who are conducting COVID-19 population surveys,” he emailed.
So yes, these antibody-survey-based estimates are provisional and imperfect! So is the rest of our knowledge about Covid-19, though, and the disappearing-antibodies effect shouldn’t have too big an impact on the numbers from this winter and spring.
The simplest way to use these numbers to estimate exactly when people became infected with Covid-19 is to calculate backward from the data on deaths. I estimated a fatality rate of 0.98% from the antibody prevalence estimated in the Annals of Epidemiology article and the city’s estimate of the Covid-19 death toll (both confirmed and suspected cases) at the end of April, then assumed that infections occurred an average of 19 days before deaths did.
The CDC estimates that the median time from infection with the coronavirus to death is 18 days for those 65 and older and 21 days for everyone else. Three-quarters of Covid-19 deaths in New York City have been among those 65 and older, which is why I went with 19. This may understate the early spread of the disease, though, because it was probably concentrated among younger people who had just been to Europe or hung out with those who had, and were more likely than senior citizens to be spending lots of time in crowded bars, concerts, dance clubs, restaurants, offices and subways. So I also charted the disease’s possible course through the city’s population using the Covid-19 Scenarios tool put together by researchers in the laboratory of physicist-turned-epidemiologist Richard Neher of the University of Basel in Switzerland (and some other people).
Covid-19 Scenarios consists of a Susceptible-Exposed-Infected-Recovered model of disease spread that allows you to tweak parameters such as the reproduction number and the period of infectiousness, as well as impose interventions that slow transmission. These interventions are not specific actions such as school closures; you simply put in an intervention for a period of time and estimate for yourself how much it will reduce transmission. My goal was to construct a scenario in which the epidemic began in early February and cumulative infections at various dates were near those estimated from Mount Sinai’s April samples and the state surveys. After some experimenting I decided to construct multiple scenarios, because designing an epidemic that hit all of the weekly Mount Sinai targets proved too difficult.
These are admittedly some pretty crazy-looking epidemics.
“It looks to me that in your scenario the outbreak is coming down a little too fast,” Neher emailed in critique of one of my early efforts, and I was never able to fix that entirely. Average the four scenarios together, though, and average that with the deaths-based estimate, and the result is the plausible-ish epidemic trajectory seen in the first chart and reproduced here in bar-chart form.
As a New York City resident, I can testify that the Covid-19 epidemic was at its most awful and frightening here in late March and early April. If my estimates are even close to right, though, the epidemic peaked sometime in mid-March and was rapidly declining as all New Yorkers who could holed up at home and freaked out. In other words, the holing-up and freaking-out seems like it actually worked — and it’s too bad we didn’t do it a little sooner.
Bloomberg
