Cardiogram announced Wednesday that its DeepHeart neural net can detect diabetes and prediabetes using data collected from Apple Watch and Android Wear smartwatches. The company said a study by its own researchers and researchers at the University of California San Francisco found DeepHeart could accurately identify diabetes with 85% accuracy using only heart rate and step count measurements.
Cardiogram is a fitness app for iOS and Android with apps for smartwatches, too. The company recruited more than 14,000 users of its apps and gathered some 33,628 person-weeks of health sensor data to train a deep neural net. The neural net was also given samples of “people with and without diabetes, hypertension, sleep apnea, atrial fibrillation, and high cholesterol.”
Wednesday’s announcement focused on the diabetic portion of the study, and Cardiogram will present its results at the Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18) in New Orleans on Wednesday.
DeepHeart’s Big News
This is a pretty big deal, especially in the U.S., where a crap tonne of us are diabetic and prediabetic. According to Cardiogram, 25% of people with diabetes don’t know it, while “88.4% of people with prediabetes don’t realize they have it.”
In other words, a lot of people are walking around—or not walking around—with diabetes completely undiagnosed. Cardiogram believes that being armed with knowledge might help some folks.
This is a remarkable bit of news. 85% accuracy is far from perfect, but identifying—Cardiogram is careful in its announcement not to say they can diagnose diabetes—diabetes and prediabetes just from fitness tracking data is great.
It’s not the same thing, however, as Apple’s rumored project for Apple Watch to do noninvasive glucose monitoring. That’s a separate effort, and even if Apple accomplishes that moonshot goal, Cardiogram’s achievement will be separately useful and beneficial.