I’ve been thinking a lot lately about the rise of wearable fitness technology, and whether its benefits will be able to match its hype. In many respects, I remain a skeptic that technology is really an improvement over learning to monitor your own feelings and sensations.
But, as I wrote in the New York Times over the weekend, I’m starting to think that the true power of wearable fitness tech could be its ubiquity. It may be hard to extract meaning from any one individual’s data—but collectively, the millions of people wearing self-monitoring devices amount to “the largest and most comprehensive observational health trial ever conducted.”
What does this mean? After the article was published, I got an e-mail from Brandon Ballinger, an ex-Google engineer who focuses on “machine learning”—think, for example, speech recognition and spam detection. He has spent a couple of years working with cardiologists at the University of California San Francisco, and recently co-founded Cardiogram, an app that collects and analyses heart-rate data from the Apple Watch and Android Wear.
Of course, there are lots of apps that monitor heart rate. But the UCSF medical researchers, led by Greg Marcus and José Sanchez of UCSF’s Heart eHealth team, along with Cardiogram, just launched a big study called mRhythm to tap into all this data in order to detect atrial fibrillation. Using a machine-learning technique called “semi-supervised deep learning,” they’ll look for variations in heart rate collected by Apple Watch wearers to see if they can pick up telltale signals of atrial fibrillation.
This should be of particular interest to endurance athletes, because atrial fibrillation seems to occur much more frequently in endurance athletes than non-athletes. One question that remains to be answered is whether the higher rates of a-fib diagnosis in athletes are partly because athletes tend to be examined more closely. Because the symptoms are often very mild, many people have atrial fibrillation but don’t even realise it—until it leads to more serious consequences like a stroke.
This is a great example of the type of big-data analysis that becomes possible (and financially feasible) with the proliferation of wearables.
One other related note: Observational trials (i.e., have lots of people wear a monitoring device, then look for patterns in their data) have plenty of limitations, even with millions of people participating. One of the examples I gave in the Times article was an analysis of 4.2 million MyFitnessPal users that yielded some interesting insights about those who succeeded in hitting weight-loss goals versus those who failed. The successful ones ate more fibre and less meat, and increased their consumption of grains and cereals, for example.
But it’s hard to distinguish between causation and correlation in observational data. Another approach is to conduct “micro-randomised trials,” like an ongoing University of Michigan, US, study in which users of a tracker-linked health app are randomly assigned to receive different fitness prompts multiple times a day.
By analysing the increases or decreases in step count after hundreds of randomisations per subject over the course of the study, the researchers will be able to tease out how factors like the weather and how busy the user’s calendar is influence the effectiveness of the exercise suggestion.
The bottom line? I think there are some pretty exciting possibilities with the continuing development of wearable tech. I just hope we don’t completely forget to pay attention to how a run feels.