Machine learning identifies diabetics with 85% accuracy using wearable heart rate monitors, study says
Photo by Shutterstock

Machine learning identifies diabetics with 85% accuracy using wearable heart rate monitors, study says

February 8, 2018

The Knife Media

Machine learning identifies diabetics with 85% accuracy using wearable heart rate monitors, study says

The Raw Data

Unspun and unbiased. These are the facts.

On Wednesday, App developer Cardiogram announced results of a study that showed 85 percent accuracy in detecting diabetes using smart watches or activity trackers that monitor heart rate and activity. The study took more than 200 million data points from 14,000 Apple Watch and Android Wear users, and put them into a neural network program called DeepHeart. According to Cardiogram, DeepHeart has been used in past studies to detect atrial fibrillation, hypertension and sleep apnea.

Cardiogram performed the study with the University of California San Francisco (UCSF). Researchers used some of the available data to “train” Cardiogram’s DeepHeart program. Once the program was “trained” to detect diabetes, a separate set of data was used to establish the program’s accuracy in identifying which of the participants were diabetics.

In the past, heart beat behavior has been found to correlate with diabetes. A 2015 Framingham Heart Study determined that “low” heart rate variability and “high” resting heart rate correlated with the development of Type 2 diabetes over a 12-year period.

According to the Centers for Disease Control and Prevention (CDC), more than 100 million U.S. adults have diabetes or prediabetes. The CDC said about 25 percent of  diabetes cases are undiagnosed.

Sources: CNET, MacRumors, Upbeat