We’re all going to die eventually—but what if you knew when you’d be at risk for dropping dead, based solely on the way you walk? A new study shows that measurements taken with wrist-worn motion sensors can be used to predict one’s mortality risk up to five years later. As one of the largest validations of wearable technology to date, the research raises the possibility of one day using the motion detection system in smartphones to survey patient health without the need for in-person visits to the doctor’s office.
The study, published Thursday in the journal PLOS Digital Health, was run using data from over 100,000 Britons from the massive UK Biobank project, which began collecting health and biometric information from participants in 2006 and will follow them for another 14 years. From a week of wrist sensor data, researchers at the University of Illinois at Urbana-Champaign designed a model that pares down a person’s acceleration and the distance they traveled into six-minute chunks. According to study author Bruce Schatz, a University of Illinois computer science researcher, the scientists chose this duration to mimic the six-minute walk test: a measurement of heart and lung function commonly taken during a medical appointment that tasks participants with walking at a normal pace for six minutes and compares their total distance traveled to benchmarks according to their age.
The test is “a very good external measure of what’s going on internally,” and could easily be replicated using the accelerometer present in a wrist sensor or a cheap phone, Schatz told The Daily Beast. “I know for a fact that these kinds of models will work with cheap phones.”
Predictions of future death made by the researchers’ model were correct 72 percent of the time after one year, and 73 percent after five years—a similar rate of accuracy found in a study published last year that analyzed the same data set but used hours, rather than minutes, of data. This new study, argued Schatz, is a more promising demonstration of passive monitoring technology like phone and wrist sensors as his team’s model requires less data and affords a great degree of privacy to the user.
“If you record all of the data, it’s true that people have characteristic walks and you can tell who the individual is. But it’s totally possible to take part of the signal, which is good enough to do the vitals but completely disguises who the person is,” he said.
“I know for a fact that these kinds of models will work with cheap phones.”
— Bruce Schatz, University of Illinois
Even so, using everyday technology to passively monitor patients could raise issues if users cannot give continuous informed consent, situations which could be complicated by degenerative illnesses or a lack of technological literacy. These ethical issues, said Schatz, are still speculative, but deserve coordinated thought from scientists as the research moves forward.
While the sensors used in the study were near-identical to the ones in both simple cell phones and smartphones, future work should validate this model in a large sample when users carry phones in their pockets, rather than wear sensors on their wrists. Downloading an app that can measure your health as you go about your day-to-day could be a convenient and painless way to keep people healthier, longer.
“If you want to raise the general health of the entire population, this kind of project is really important,” Schatz said.