Traditionally, estimating influenza trends relies upon the documentation of virologic and syndromic influenza-like illness (ILI). This surveillance typically has a 1- to 3-week reporting lag, though, and is usually preliminary and revised as more information becomes available.
As such, there is a clear need to enhance flu surveillance and employ objective data streams that are capable of providing real-time information.
A team of investigators from the Transitional Institute at Scripps Research set out to assess whether population trends of seasonal respiratory infections like the flu could be identified through heart rate and sleep data collected by Fitbits. Their research is published in The Lancet Digital Health.
“Activity and physiological trackers are increasingly used in the United States and globally to monitor individual health,” the investigators wrote. “By assessing these data, it could be possible to improve real-time and geographically refined influenza surveillance.
An acute infection can lead to elevated resting heart rates and individuals may also alter routine activities as a result. To assess whether resting heart rate and sleep data could bolster influenza estimations, the investigators looked at de-identified data from 200,000 individuals in the United States who used a Fitbit between March 1, 2016, through March 1, 2018. In particular, the team focused on the 5 states with the most Fitbit users: California, Texas, New York, Illinois, and Pennsylvania.
The investigators compared sensor data with the weekly estimates of ILI rates reported state-by-state by the US Centers for Disease Control and Prevention (CDC). Then they looked at the weeks in which Fitbit users had elevated resting heart rates and increased sleep rates.
In total, the investigators identified 47,249 users across these 5 states who wore a Fitbit consistently during the study period. This consisted of more than 13.3 million total resting heart rates and sleep measures.
Using this data, the team modeled ILI case counts for each state along with a negative binomial model including 3-week lagged ILI data. Investigators also evaluated weekly change in ILI rates using linear regression and Pearson correlation was used to compare predictions vs CDC reported ILI rates.
The investigators found that the Fitbit data significantly improved ILI predictions with an average increase in Pearson correlation of .12 [standard deviation: .07] over baseline models, corresponding to an improvement of 6.3–32.9%.
It is also reported that correlations of the final models with the ILI rates from the CDC ranged from .84 to .97. Week-to-week changes in the proportion of Fitbit users with abnormal data were found to be associated with week-to-week changes in ILI rates in most cases.
“This information could be vital to enact timely outbreak response measures to prevent further transmission of influenza cases during outbreaks,” the authors concluded.
