Nobody knows how long this coronavirus pandemic will go on: When can schools safely open again? When will we be able to see a concert? Dine in at a restaurant? When will life return to normal?
It’s tough to say. COVID-19 is still very new. It spreads quickly, and despite all our technology and scientific know-how, there’s still a lot we don’t know about it.
That includes how many people have it. A lack of testing capacity in the U.S., including Minnesota, means a fraction of the people with COVID-19 symptoms have been tested for the virus. While new tests that can detect if a patient has had COVID-19 are expected soon, they’re far from widely available.
But if this virus sticks around for a while — and experts say that’s a real possibility, especially if it takes a long time to develop vaccines and therapies – it would help decision makers greatly if they had a sense of how prevalent COVID-19 is at any given time.
A researcher at the Federal Reserve Bank of Minneapolis has an idea that could help with that.
Abigail Wozniak, the director of the Opportunity and Inclusive Growth Institute at the Minneapolis Fed, is not an epidemiologist, the type of researchers who typically model disease spread. But the absence of that critical data that could help policymakers make better decisions about how to handle COVID-19 sounded a lot like some of the problems she’s seen in her field of economics.
In this case: “How can we potentially get a better idea of the scope of infection if we’re not able to ever test the population in a widespread way?” Specifically, she said, we’re missing timely information on where people show symptoms and how are people doing under social distancing restrictions in given geographies.
She put together a research proposal designed to fill in those gaps.
Wozniak proposes surveying 2 million people living in the U.S. every day — enough to drill down geographically to see whether COVID-19 is flaring up in some communities or dying down in others.
“So what you want to do is survey tons and tons of people about their own physical symptoms and basically use big data techniques to try to pull out information about areas where the illness is emerging or retreating.” she said.
In addition to asking respondents questions about things like fever, chills, cough and underlying health conditions in a physical health segment, Wozniak proposes asking about social and mental health, as well as economic health. This type of data could help policymakers better understand how things like stay-at-home orders are affecting different populations.
At 2 million people per day, Wozniak said the proposal is a bit of a moonshot. But she feels it’s attainable: The U.S. Census Bureau surveys 2 million people every year in the American Community Survey, the biggest survey it conducts.
“The ACS is representative of statistics at really fine geographies, so you can get county level statistics. You can get small populations like Native American populations, so you can really figure out reliably what’s going on at a really fine level. So that’s why I picked that number,” she said.
She envisions the survey as a partnership between government agencies with the statistical chops and tech companies with access to millions of users.
“[This is], well within the scope of that industry to collect this volume of data and process it quickly,” she said. At its peak popularity, she notes in her proposal, Pokémon Go handled the daily interactions of more than 100 million users.
“This is just something that a tech company can roll out [that] people can take it on their phones,” she said. “As long as you’re getting a reliable sample, you’ll be able to figure out what the distribution of these symptoms are, and then you actually have real data on where cases have emerged.”
Techniques like Wozniak proposes, which rely on the reporting of symptoms in lieu of lab-verified tests to track disease, are called syndromic surveillance.
State health officials have long tracked flu symptoms across the state this way through reports from providers on the share of people who show up with flu-like symptoms.
Because of a lack of lab testing capability, the Minnesota Department of Health has said it is working to ramp up syndromic surveillance efforts to track COVID-19.
“When we talk about syndromic surveillance, what we’re acknowledging is that it’s true that we’re not going to be able to identify every single case of COVID-19 in the state,” MDH Infectious Disease Director Kris Ehresmann said in a call with reporters Saturday.
The Minnesota Department of Health has been promoting the website of COVID Near You, a collaboration between Boston Children’s Hospital, Harvard Medical School and computer programmers. COVID Near You is a quick online health survey designed to pinpoint the prevalence of COVID-19 symptoms geographically.
Kumi Smith, assistant professor in the University of Minnesota’s School of Public Health’s division of epidemiology and community health, said she hadn’t reviewed COVID Near You, but took a look at Wozniak’s proposal.
While efforts to track symptoms of COVID-19 may be useful tools, they are likely to have limitations, she said.
Syndromic screening is typically done by clinicians — people who are well-versed in the way disease presents itself — in the absence of diagnostic tests. Asking the general public to self-report their symptoms may yield less reliable information. The general public may also get bored, especially if a survey is long, and not answer questions correctly.
“You can get back these nonsensical responses because especially if people are doing this online, they might just be clicking whatever,” she said. Then there’s the question of distinguishing COVID-19 symptoms from flu ones, which are fairly similar.
Smith said a tool like this might be useful to trigger a more targeted screening approach in some areas believed to have outbreaks.
Wozniak acknowledged that the large-scale symptom screening is a bit untested, but said she’s gotten positive reactions from the epidemiology community.
Plus, she said, it’s a “do-no-harm” approach — one that doesn’t rely on administering tests or processing them, which could take away from front-line operations to fight the disease.
“As long as you’re getting a reliable sample, you’ll be able to figure out what the distribution of these symptoms are, and then you actually have real data on where cases have emerged,” Wozniak said. “So you could really very quickly figure out whether this works and whether it’s predictive of disease spread.”