Wearables Could Improve Flu Forecasting, but They Miss the Most Vulnerable

Flu cases can be predicted from heart rate and sleep information

Photo: PeopleImages/E+/Getty Images

In 2017, a group of Stanford researchers found that wearable devices for tracking exercise could also tell when you’re getting sick based on abnormal measures of heart rate, skin temperature, and other biometrics.

The results intrigued Jennifer Radin, an epidemiologist and digital medicine expert at the Scripps Research Translational Institute in La Jolla, California. She wondered whether it was possible to pull together data from many wearable users to predict cases of flu. “Across my own wearables, I noticed my heart rate rose when I got sick, so I was really interested to see if I could apply this to a population level,” Radin told OneZero.

Using de-identified data from more than 47,000 Fitbit users, Radin and her colleagues improved flu forecasting in five states compared to current surveillance methods. The findings were published Thursday in the Lancet Digital Health. But crucially missing from this forecast were people who don’t own wearables, which are most likely to include children, the elderly, and lower-income people. These groups are also more susceptible to serious complications that can arise from the flu.

In the United States, about 7% of working adults and 20% of children younger than five get the flu annually. While most cases are mild, between 12,000 to 61,000 people have died annually from the flu since 2010, according to the U.S. Centers for Disease Control and Prevention (CDC). Public health experts also worry about another pandemic flu, like the H1N1 outbreak in 2009, which hospitalized and killed more people than seasonal flu during other years.

The CDC predicts flu cases using information about new flu-like illnesses from hospitals and health care providers across the country that share data with the government. But this method has a lag time of one to three weeks, the estimates are often later revised, and reporting delays can allow outbreaks to go unnoticed. In the meantime, the flu can spread.

Just a fraction of the health care providers and hospitals across the country report cases of flu-like illness to the CDC, and only a quarter to one-third of people actually see a doctor when they have flu symptoms. Individuals with lower incomes may be less likely to visit a health care provider when they’re sick because of cost and access issues. So the CDC’s current flu surveillance system misses some cases.

Acknowledging these gaps, Radin and her colleagues investigated whether Fitbit data could be used to make accurate flu predictions in real time. From a data pool of de-identified users, they identified more than 47,000 people from California, Texas, New York, Illinois, and Pennsylvania who wore their Fitbits consistently. They then calculated those users’ average resting heart rate and sleep duration, as well as deviations from the average, to determine which of the users were potentially ill. A person’s resting heart rate tends to rise when they have an infection, especially when they have a fever. Sleep and activity patterns also differ when a person is sick. By comparing the proportion of sick users each week by state with the CDC’s weekly estimates for flu-like illness, the researchers found they were able to make more accurate flu forecasts.

Wearable users are generally more likely to have higher incomes and thus better access to health care.

“This has the potential to cover a large fraction of the population and be able to monitor trends in this population over time, not just in flu but other changes in population health,” Cecile Viboud, a senior research scientist at the Fogarty International Center at the National Institutes of Health, told OneZero. Vibound wrote a commentary accompanying the new study.

With access to 24/7 real-time Fitbit data, scientists could potentially identify flu rates and outbreaks on a daily instead of weekly basis, says Radin. But she and her colleagues recognize that some people won’t be captured by this model.

Owning a Fitbit selects for a very specific subset of the population. For one, wearable users are generally more likely to have higher incomes. A recent survey by the Pew Research Center found that about 31% of Americans living in households earning $75,000 wear a smartwatch or fitness tracker on a regular basis while 12% of people with an annual household income below $30,000 do. Children and the elderly are also much less likely to use wearables.

Despite these gaps in ownership, health care providers and public health experts are keen on the potential for wearables to predict patient health, track diseases, and improve the management of certain health conditions. As wearables become cheaper and more ubiquitous, their use as a source of public health data will become more feasible. But for now, it’s not clear yet whether doctors and researchers can rely on wearables for long-term data collection. By one estimate, a third of U.S. customers who own a wearable eventually abandon them.

And with Google’s recent purchase of Fitbit, concern for data privacy may also put off some users. All Fitbit users, including those whose data were included in the Lancet study, are notified that their de-identified data could potentially be used for research when they purchase a Fitbit device, according to Fitbit’s privacy policy.

Viboud says other data collection methods, like smartphone apps, could also be used to close the gaps in disease surveillance. One app, called Flu Near You, asks users each week whether they are experiencing flu-like symptoms. “The idea,” she says, “would be to combine different streams of information that might target or capture different populations.”

Google Flu Trends attempted to do real-time tracking, but it missed early waves of the 2009 H1N1 pandemic flu and overestimated other cases. Researchers have also turned to Twitter to predict flu cases, but those predictions only include a small segment of adults who use the platform.

For now, data from wearables will most likely be used to fill in gaps in existing data. “Fitbit data and Google searches might be able to accurately predict flu for a fraction of the population that uses this technology,” says Viboud, “but we could still rely on some traditional surveillance data from general practitioners and hospitals to get other age groups or populations that don’t use these modern technologies.”

Former staff writer at Medium, where I covered biotech, genetics, and Covid-19 for OneZero, Future Human, Elemental, and the Coronavirus Blog.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store