Editor’s note: Find the latest COVID-19 news and guidance in Medscape’s Coronavirus Resource Center.
What’s claimed to be the largest symptom survey database ever compiled during a public health emergency is being enlisted to help public health experts and the public identify emerging COVID-19 trends in real time.
The three universities backing this effort — Carnegie Mellon University, the University of Maryland, and Duke University — and a marketing/conference firm called [email protected] 2.0 have launched a cash contest to elicit “novel analytic proposals” to use the Facebook-based survey data for earlier COVID-19 detection and improved situational awareness of the outbreak.
Open to academic researchers and others, the COVID-19 Symptom Data Challenge will award $50,000 to the grand prize winner, $25,000 to the runner-up, and $5000 to the semifinalists.
The publicly available symptom database includes the results of about 30 million consumer responses worldwide, according to the challenge’s website. Facebook has placed a button for the symptom survey on its news feed as part of its Facebook Data for Good initiative.
Consumers click on the button and are taken to the platform of the continuing survey, which is conducted by the Delphi Group at Carnegie Mellon University and the University of Maryland. The survey has been ongoing since April 6 in the United States and since May 1 globally.
In a September 1 post on The Health Care Blog announcing the contest, Farzad Mostashari, MD, a former national coordinator of health IT at the US Department of Health and Human Services who has conducted research on syndromic surveillance, explained the purpose of the competition.
The main types of data now being used to identify COVID-19 outbreaks, he noted, all have serious flaws: Case numbers are dependent on lab tests, which are constrained by limited lab capacity, and positivity rates vary depending on the population tested. Studies of excess mortality suggest numbers of deaths from COVID-19 are unreliable because the disease is underdiagnosed. And traditional syndromic surveillance depends on people presenting to emergency departments and hospitals. If they avoid these settings because of the pandemic, “it obscures that lens,” he said.
“So [compared] with all three of these [data sources], the symptom survey data presents unique advantages. Compared to deaths, it’s much more timely. In fact, compared to any of the other data sources, you would expect it to be the first indicator. It’s completely unrelated to health-seeking behavior or testing availability. And, particularly in the global context, there are many countries where the lab capacity is really challenged, and even mortality surveillance is really challenged. This could be a major tool,” Mostashari said.
Preliminary evidence suggests that the symptom survey data are valid, he noted. Although the universities have made the data available online, little has been done with them so far. The goal of the competition is to find ways to validate the data and make them useful to policymakers and the public in guiding their decisions.
Mostashari emphasized that the respondent data are anonymized and that Facebook has no access to them. The social media giant is simply trying to encourage its 2 billion or so users to contribute the information.
The 30 million responses to the survey have come from 200-plus countries and territories in more than 55 languages. The US and the global survey instruments contain questions about COVID-like symptoms (eg, fever, cough, anosmia), demographics (eg, age, sex, ZIP code, size of household), and behaviors (eg, mask wearing and the number of direct contacts outside the household in the previous 24 hours).
There have been more than 1.5 million unique respondents per week, and the survey could be expanded if its utility for outbreak monitoring and response were established, the contest sponsors note.
“Early research shows that US state-level rates of COVID-like symptoms correlate with rates of diagnosed disease burden, and state-level analyses tend to show that symptoms are roughly coincident in time with cases and about 14–21 days earlier than deaths,” the contest website states.
A 2018 study of syndromic surveillance systems in six state and local health departments found that this method was helpful in detecting surges in seasonal flu cases and in identifying an outbreak of foodborne illness prior to traditional, disease-specific systems.
The study interviewees agreed that syndromic surveillance may be useful for monitoring the spread and intensity of large disease outbreaks; improving public health awareness of mass gatherings and natural disasters; and assessing new conditions when real-time alternatives are unavailable.
Preeti Malani, MD, chief health officer and a professor of medicine in the Division of Infectious Diseases at the University of Michigan, told Medscape Medical News that syndromic surveillance has proved helpful in some situations, especially with seasonal influenza.
However, she was skeptical about how useful the Facebook-based symptom survey data might be.
To begin with, she noted, “You’re missing all of the asymptomatic people, so you’re getting only a partial picture.” Also, from the University of Michigan’s experience with its own symptom checker app, a lot of people don’t want to complete surveys or they lie about their symptoms.
“You have to make sure enough people are filling surveys out, and people don’t want to do things they don’t have to do. To be really helpful, you have to make sure people are doing this consistently,” she said.
Malani also pointed out that the Facebook-based survey would have to be scaled up enormously to have any impact at a local level. Although 30 million total responses sounds like a lot, she said, the number of responses in the past week in a particular US ZIP code would be small and could be misleading if only a few individuals filled out the survey.
“Maybe you’re responding, and your risk is very low. But you could live in a place where there’s a big outbreak, and we’d never pick it up [from the survey data],” she said.
That said, Malani added, a symptom survey is one of the pieces that might help improve COVID-19 surveillance. “COVID really requires this multilayered approach,” she said, “and symptom monitoring is one of the things we could use.”