During the Covid-19 pandemic, the United States is flying blind, unable to track its spread or accurately predict when the disease will peak in different parts of the country because of a lack of diagnostic tests.
To try to fix that, some researchers are turning to tools that first debuted after the 9/11 World Trade Center and anthrax letter attacks of 2001, asking whether data that is collected under other auspices—or volunteered by people who think they might be infected—could buttress the official detection system and offer early warnings of hot spots. Some think it could be one of the pillars for a national strategy that carries the United States into a safe zone beyond the peak—provided key issues of reliability, privacy, and equity can be worked out in time.
The term for collecting that kind of data is “syndromic surveillance.” In medicine, a syndrome is a cluster of signs (things that can be measured, such as temperature) and symptoms (things that are subjective, like a headache) that might be associated with more than one diagnosis, or hasn’t received a diagnosis yet. Noticing signs and symptoms is what propels people to seek a doctor’s visit if they can, and get tested if a test is available, so identifying syndromes can provide useful data days or weeks before test results arrive. But because syndromes are nonspecific—fever, body aches and cough, for instance, could indicate either Covid-19 or the flu—lifting the signal out of the background noise is a challenge to solve.
The first project to step up to the challenge is Covid Near You, a crowdsourced symptom tracker built by a team of epidemiologists and bioinformaticists at Boston Children’s Hospital who run the global early detection site HealthMap and created a similar project, Flu Near You, in 2011. Covid Near You launched in late March after a week of intense donated effort by Brownstein and volunteers from Apple, Amazon, Google corporate parent Alphabet, and a handful of other companies. As of Friday night, more than 444,00 people in the US had fed their symptoms into the tracker, along with more than 397,000 in Canada and 2,000 in Mexico.
The site doesn’t ask people to identify themselves, beyond their age, gender, and zip code. It also doesn’t restrict data entry to people who have been to a doctor or have a lab-confirmed test result. It asks a simple introductory question: How are you feeling? If you say Great, Thanks! it routes you to registering whether or not you had a flu shot and then optionally asks for a mobile number for texts. If you choose Not Feeling Well, it routes you instead to an interrogation of 21 signs and symptoms, as well as your flu shot history, travel history, whether you’ve had contact with a Covid-19 patient or sought a Covid test, and whether you’ve been quarantining.
The site’s back end sorts out the well from the potentially sick, stashes the information about the not-sick for future reference, and reports in map form only those whose symptoms match Covid-19; it also shows where people have tested positive. Recently, it’s been displaying the United States’ known local outbreaks—Boston, New York, Florida, Detroit, San Francisco, Seattle—but also intriguing upticks in Dayton, Ohio; Denver; and in California’s Central Valley.
“Most people who are ill don’t ever see a health care provider,” says John Brownstein, Covid Near You’s lead developer, who cofounded HealthMap and is an epidemiologist and a professor at Harvard Medical School. “That’s what we’re looking for, the bottom of the iceberg, the first wave of infection that doesn’t get detected until some of it turns into hospitalizations and deaths.”
Self-reports have a signal-to-noise problem, of course: The worried well, and people who might share a syndrome but be infected with a different organism, can clog up the data. But after a near-decade operating Flu Near You and validating its results against data published each flu season by the Centers for Disease Control and Prevention, Brownstein feels confident the Covid-19 data will have predictive worth as well.
“What we are seeing validates the hot spots reported in urban areas,” he said, “but we are starting to see increases outside of New York City, in places like Nassau County and on Cape Cod. Those are areas people are not yet talking about.”
Covid Near You doesn’t deliver any diagnoses, of course. But its flagging of hot spots could indicate to public health officials where the disease might surge next. Seeing the results might also help individual citizens be more engaged, Brownstein said. That could reinforce their decision to keep up social distancing while pressure is building to relax precautions, or encourage people to participate in the serosurveys that will demonstrate an individual’s potential immunity—and a community’s relative safety—once those tests roll out.
Syndromic surveillance has had a mixed reputation. If people know about it at all, that might be because of Google Flu Trends, a project that attempted to derive early warnings of flu season by analyzing locations where people were typing possibly flu-related search terms into the Google index page: “flu symptoms,” for instance, or “orange juice” or “chicken soup.” Google Flu Trends debuted to big hype in 2008 and failed spectacularly in the flu season of 2013. Among the general public, it left behind a vague sense that big data wouldn’t be predictively useful for public health.
But at that point big data, from individuals and institutions, was already being put to public health’s service. Syndromic surveillance arose out of post-9/11 worries that the US might have been targeted using biological weapons, the first symptoms of which might show up in emergency rooms as common syndromes—fever, chills, breathing trouble—indistinguishable from colds and the flu. The anthrax letter attacks a month later reinforced that fear; the first case, in Florida, was detected by an alert physician who realized a patient’s flulike illness was caused by something much more dire.
Farzad Mostashari, a physician, epidemiologist, and former CDC epidemic intelligence officer, was working in New York City’s health department at the time. He organized an effort to collect data feeds from city hospitals and emergency rooms into detection systems that would alert the department to spikes in flu symptom reports and clusters of unusual illness—a system that is still running and might deliver early signals of Covid-19, as he pointed out on Twitter on March 7.
Mostashari went on to become the National Coordinator for Health IT in the Obama Administration, and took advantage of that White House’s national economic stimulus legislation to help engineer American hospitals’ conversion to electronic health records. Built into that wiring-up was a commitment to harvest and send de-identified syndromic data into a detection system. Those became what is now the federal National Syndromic Surveillance System, hosted at the CDC, which collects data from approximately 70 percent of US emergency rooms.
“Syndromic surveillance became part of public health practice,” says Mostashari, who is now CEO of Aledade, a startup that applies data analysis to managing primary-care practices. “We now have three types of public health surveillance in the US: lab surveillance, clinical case reporting, and then syndromic surveillance added.”
That kind of syndromic surveillance—reporting signs and symptoms, recorded in a health care encounter—might help the country figure out how to open up once the Covid-19 peak passes. Mostashari is one of several authors of a plan for a national Covid-19 surveillance system, published April 7 by the Margolis Center for Health Policy at Duke University. The plan has four pillars; syndromic surveillance is the second. The authors argue that expanding what the existing national system looks for, and pulling in additional data streams, could help identify new Covid-19 outbreaks in time to test potentially infected people and get them isolated before the disease spreads. “Visit-based syndromic surveillance is going to be an absolutely key part of the arsenal,” Mostashari says.
But that leaves out, for now, the hospitals and other care centers that aren’t hooked into the system. Self-reports contained in tools such as Covid Near You might allow detection to happen earlier—what epidemiologists call “moving the curve to the left,” back down the time axis of an X-Y graph. To make that useful, designers are going to have to think through whether people who might be affected by a resurging disease possess the devices and access that let their illness be detected.
The promise of devices for detecting the spread of Covid-19 is attracting a lot of enthusiasm— and scrutiny—right now. For instance, there’s the joint Apple-Google effort to detect, via anonymous codes exchanged over Bluetooth, whether people have unknowingly been in touch with someone who is infected. There’s also the expectation that wearables, already tracking key signs such as respiration and heart rate, could gather enough data to predict a disease’s emergence—which is similar to the the hope that smart-reporting devices such as the Kinsa thermometer could flag the start of an outbreak before a human notices something wrong.
All of those envisioned tech-optimist benefits assume that someone can afford a smart thermometer, or a wearable device, or even a decent-bandwidth phone. That’s not guaranteed—which means that we can’t assume people will have access to the apps that collect epidemiological data either.
“Pew Charitable Trusts research shows that US adults in households making less than $30,000 a year, as late as 2019, had only 71 percent ownership of smartphones,” says Dina Mistry, a postdoctoral research scientist in the independent Institute for Disease Modeling outside Seattle. “The same survey showed that only 53 percent of those over 65 owned a smartphone. So if we’re relying on these kinds of tools, it becomes a question of equity, that we might be missing those kinds of people more than others.”
Those data gaps could lead to outbreaks not being detected as fast as they should, or the sources not being sent to communities who need them. As Covid-19 moves across the country, it’s going to take a mix of tools and detection systems to get help as fast as possible to those who need it most.
- Why are some people getting so sick? Ask their DNA
- New Yorkers, once again at ground zero, in their own words
- Un-miracle drugs could help tame the pandemic
- WIRED Q&A: We are in the midst of the outbreak. Now what?
- What to do if you (or a loved one) might have Covid-19
- Read all of our coronavirus coverage here
