According to a study conducted by researchers at the University of California SF in partnership with the team behind the Cardiogram app for Apple Watch, the heart rate monitor on the device can detect abnormal heart rate with an accuracy of 97 percent.
Over the duration of the study, over 139 million heart rate and step count measurements were collected by just under 10,000 users who enrolled in a Health eHeart Study. The data was trained using DeepHeart, Cardiogram’s deep neural network.
Once trained, the machine learning model was able to detect atrial fibrillation with an accuracy of 97 percent.
The study suggests that the Apple Watch can predict this abnormal heart rate with higher accuracy than the FDA-approved KardiaBand. In a statement, Cardiogram co-founder Johnson Hsieh said:
97% accuracy refers to the c-statistic, or area under the sensitivity-specificity curve. Surprisingly, both the sensitivity and specificity of DeepHeart were even higher than an FDA-cleared Apple Watch ECG attachment — 98% (vs 93%) sensitivity and 90% (vs 84%) specificity.
The study was published in JAMA Cardiology this morning. Cardiogram said that the study marks the first peer-reviewed study in a medical journal that demonstrates that these wearable devices can detect a major health condition.