Study Identifies Abnormal Surge of Flu-like Illnesses in March
Study Identifies Abnormal Surge of Flu-like Illnesses in March

Study Identifies Abnormal Surge of Flu-like Illnesses in March

Modelers try a new approach to gauge the true number of COVID-19 cases in the US by using surveillance data for flu-like illnesses.

Katarina Zimmer
Katarina Zimmer
Jun 30, 2020

ABOVE: © ISTOCK.COM, DESKCUBE

It’s widely accepted among experts that reported COVID-19 cases—currently around 2.6 million for the US—is a vast underestimate of the actual number of infections by the SARS-CoV-2 coronavirus due to limited testing capacity. But figuring out just how much of an underestimate it is has remained elusive.

A three-person team of researchers estimates that 8.7 million Americans could have been infected with the novel coronavirus between March 8 and 28, a period during which fewer than 120,000 new cases were documented. The findings, published last week (June 22) in Science Translational Medicineare based on an abnormal surge of flu-like illnesses reported to a federal disease surveillance network in March, which the authors attribute largely to SARS-CoV-2.

“[The study’s] estimate is plausible but on the very high end of what other models and experts say was happening at the time,” says Nicholas Reich, a biostatistician at the University of Massachusetts Amherst, who wasn’t involved in the research. While using surveillance data on flu-like illnesses to detect emerging epidemics is a “really good idea,” he adds, “in practice it’s really hard.”

In February, recognizing that testing capacity was severely limited in the US, mathematical biologist Alex Washburne of Montana State University and biostatistician Justin Silverman of Penn State University began to investigate whether they could detect an increase in respiratory illnesses through data from ILINet, a network of 2,600 outpatient clinicians around the US who report weekly data on patients with influenza-like illnesses to the US Centers for Disease Control and Prevention (CDC).

Along with a colleague at Cornell University, the two observed an abnormal surge in flu-like illnesses between March 8 and 28. They then subtracted out diagnosed influenza cases as well as the typical numbers of flu-like infections at that time of year. Taking the remainder and assuming that physicians across the country who don’t report to the CDC’s flu surveillance network were seeing a similar influx of patients compared to physicians who participate in the network, they estimated that there were around 2.8 million cases of flu-like infections across the US during the study period that led people to see a doctor. They say they believe those cases are most likely due to COVID-19, the disease caused by SARS-CoV-2.

This spike of flu-like illness also correlated with the rise of confirmed coronavirus infections across different states and was also consistent with the predictions of an independent epidemic model they ran. While not proof that the surge can be explained by COVID-19 alone, “it’s some evidence [that supports] the hypothesis that these could be attributable to COVID,” Washburne says.

To extrapolate how many people nationwide contracted SARS-CoV-2 during this time, the team reasoned that one in three people infected with the coronavirus would visit the doctor. That assumption was based on one study of the Diamond Princess cruise ship, which became a coronavirus hotbed in February, that found that 18 percent of passengers tested positive for the virus but never developed symptoms, and a large 2011 study estimating that around 40 percent of adults with the flu will go and visit the doctor.

This led the team to its total estimation of at least 8.7 million infections across the US in the three weeks prior to March 28— a surprisingly high figure to Silverman. “For me, these results were sort of earth-shattering,” he recalls.

He and his colleagues estimated that around 8.3 percent of New Yorkers would have been infected by March 28, which they argue is roughly in line with a serological survey that estimated around 14 percent of residents had been infected with the virus by late April—a time when official infection figures counted only 0.3 percent of New Yorkers.

Columbia University epidemiologist Jeffrey Shaman finds the team’s estimates of true infections too high. If there were 8.7 million actual cases when only 120,000 were recorded, then only one in 72 cases was being reported in March, he says. “We get something more like one in twelve, which is one-sixth of that, which would suggest 1.5 million infections at that time,” he adds, referring to his own models. His skepticism of the study’s conclusions stem in part from early preprints of the report on medRxiv that presented even higher figures, one version predicting there were “at least 28 million presumed symptomatic SARS-CoV-2 patients” in March. (Washburne and Silverman stress they wouldn’t expect the one-in-72 ratio to persist much longer than March, partially due to improvements in testing).

Lower numbers would be more closely in line with other estimates, Reich says. In a call with reporters last week, CDC director Robert Redfield said the agency estimates that for every reported case, there are 10 more undetected ones—amounting to 20 million infections in the US to date—which is based on a number of recent serological surveys, TIME reports. In an April poll of various modeling groups published by FiveThirtyEight, Reich found that although estimates ranged from 289,000 to 12.8 million infections by the end of March, 1.1 million was the consensus estimate.

“Many other methods have shown that [the ratio of reported to unreported infections] closer to one in ten is more in line with the data,” Reich says. Although serological surveys have provided some “glimpses” into the possible scale of the outbreak, there’s still a lot of uncertainty around the true number of infections, he adds.

Using surveillance data for flu-like diseases to pick out abnormal fluctuations is a well-known technique. However, that kind of data is highly sensitive to changes in healthcare-seeking behavior, which likely shifted in unpredictable ways in March as the pandemic was taking off in the US. People with mild symptoms may have been more likely to visit the doctor because they were worried they had the coronavirus, or less likely to do so because they were afraid of catching it in a clinic. “It’s not clear to me that [the authors] account for this change in care-seeking behavior adequately,” Reich says.

It’s possible that some of the observed surge could be explained by other cold-causing viruses, notes Shaman, adding that he’d like to see more work untangling which other viruses were circulating in March. Although “it’s possible that there could be the perfect confound where there was some respiratory syncytial virus that blew up in New York,” Washburne says, he and Silverman think the surge is most likely due to COVID-19, he adds.

Shaman notes that the team’s estimates are also highly sensitive to certain assumptions, such as reasoning that one-third of SARS-CoV-2–infected people will visit the doctor, which may not reflect reality. Arthur Reingold, an epidemiologist at the University of California, Berkeley, tells Science News that assuming clinics that participate in the ILI reporting network see a similar cohort of patients as other clinics do may also not be necessarily true.

To the authors, their results demonstrate the potential utility of using symptom-based surveillance data as an early warning system for emerging epidemics, particularly for low- and middle-income countries with poor testing capacity, Silverman says. “Just asking people what their symptoms are, even though it’s not perfect . . . is really useful, even when it’s not designed for the disease that you originally thought.”

J.D. Silverman et al., “Using influenza surveillance networks to estimate state-specific prevalence of SARS-CoV-2 in the United States,” Science Translational Medicine, doi:10.1126/scitranslmed.abc1126, 2020.