Apple Watch Detects Diabetes, But Not By Measuring Blood SugarKatie Taylor
The Apple Watch has been creating a lot of buzz lately. It’s growing beyond a fitness tracker into a health tracker—like a mini-checkup on your arm. The news reports and flashy commercials are hard to miss, and people are wondering just how much this piece of wrist-bling can really do.
According to a new report, the Apple Watch can identify diabetes with 85% accuracy. That’s impressive, but people that already know they have diabetes aren’t interested in being told so again. They want answers to the big question: can this watch monitor my glucose levels?
As the technology currently stands, the Apple Watch alone cannot monitor glucose. It can, however, be linked with several third-party accessories, such as the Dexcom CGM system, which includes a sensor worn just under the skin and a small transmitter that sends glucose readings to a receiver or compatible device (such as an iPhone or Apple Watch). That could be hugely convenient, but the Apple Watch is still not a noninvasive glucose monitor.
So how in the world can it detect diabetes? It does so by looking at a person’s heart rate, or more specifically, their heart rate variability.
The nerve damage cause by diabetes can cause an irregular heartbeat. Cardiogram, a company that offers a free app that collects and organizes heart rate data, partnered with UC San Francisco to collect enough data to train their technology to learn what the heart patterns of a person with diabetes look like.
Computers, and in this case, apps, learn to look for patterns, recognize trends, and make decisions using artificial neural networks. Think about facial recognition technology on social media—after being fed hundreds of thousands of images, a computer can recognize what a face is, and whose face is whose, even when other elements in the image change. It might get things wrong now and then (just like a human brain), but it’s much more often right.
Neural networks and their accompanying algorithms allow computers to take huge amounts of data and recognize patterns that humans might not. But the computer needs massive amounts of data in order to identify those patterns, and that’s where the UC San Francisco study comes in. About 40,000 participants shared their medical information with the Cardiogram app. As the app collected heart rate information, it was able to learn which heart rate variability numbers were associated with diabetic participants.
Hence the Apple Watch’s ability, when paired with an artificially intelligent app, to recognize a diabetic heartbeat 85% of the time.
Apple is quick to advise that the watch is not a medical device and cannot replace doctor visits. And it’s not clear what exactly about heart rate variability is correlated with diabetes. So while the app and watch can match the heart rate variability with diabetes, researchers aren’t sure why.
Mark Pletcher, one the the principal investigators of the study, is concerned that machines are learning patterns but not able to help humans understand those patterns. “It makes me nervous, frankly,” Pletcher told Wired, “We’ve had a lot of internal discussions about whether this could be picking up medications diabetics use or some other extraneous factor.”
Eric Topal, Director of the Scripps Translational Science Institute, was more direct, “This combines features of the black box of algorithms and the black box of biology. It’s unconvincing and shaky.”
That’s disappointing, especially since early detection of diabetes could have a huge impact on Americans’ health. But Brandon Ballinger, co-founder of Cardiogram, thinks that if a simple screening from an Apple Watch could get someone to their doctor to test for a serious health condition such as diabetes, then the app has still made a positive impact.
It’s estimated that 1 in 4 people with prediabetes or diabetes in the United States go undiagnosed. At this juncture it seems the Apple Watch may have potential to change that and perhaps combat other health conditions. The Apple Watch paired with Cardiogram can detect abnormal heart rhythms with up to 97% accuracy and hypertension with 82% accuracy.
For now it looks like the healthcare world will be in a “wait and see” pattern until more studies are completed. Having health information at one’s fingertips (or wrist) could make a huge difference in how Americans are diagnosed and how early they can begin treatment, but only if that information is reliable and interpretable.