Study: Cardiogram's DeepHeart machine learning algorithm predicts hypertension, sleep apnoea from wearable heart rate sensors

Study data presented at the American Heart Association (AHA) Scientific Sessions suggest that Cardiogram's deep neural network (DNN), DeepHeart, was able to predict hypertension and sleep apnoea from wearable heart rate sensors with high sensitivity.

The Health eHeart study aimed to evaluate whether DeepHeart could predict cardiovascular risk factors from off-the-shelf wearables with a photoplethysmographic heart rate sensor and accelerometer. The researchers enrolled 6115 active users of the Cardiogram app for Apple Watch, and collected heart rate and step counts for a mean of 8.9 weeks. The respective rates for hypertension, sleep apnoea and diabetes were 36.5 percent, 16.6 percent and 7.6 percent. Data from 70 percent of participants was used to train DeepHeart to simultaneously predict prevalent hypertension, sleep apnoea and diabetes, while test performance characteristics were assessed using the remaining participants.

Results showed that DeepHeart was able to predict hypertension and sleep apnoea from wearable heart rate sensors with 82 percent and 90 percent accuracy, respectively. However, the results were not statistically significant for diabetes. Cardiogram co-founder Brandon Ballinger noted that DeepHeart detected 52 percent of sleep apnoea with a specificity of 97 percent, adding that "if a specificity of 82 percent is acceptable, then we can detect even more sleep apnoea, about 75 percent of people." The authors recommended further research to see "whether such DNNs can provide durable and portable predictions for these conditions in other study samples."

The DNN could also be used with other over-the-counter wearables, with Cardiogram co-founder Johnson Hsieh noting "they basically all have the same technology built inside." Hsieh added "the idea here is that by screening continuously you would identify people with hypertension who might not know they have it."

In May, Cardiogram presented study data showing how the Apple Watch can be used to detect abnormal heart rhythms. The study found that the app was accurate in detecting atrial fibrillation 97 percent of the time using the smartwatch's heart rate sensor, compared to screening tests performed at the hospital.

Did you like this article?