Peer-reviewed medical journal JAMA Cardiology today published a study suggesting that an Apple Watch equipped with Cardiogram, an AI-assisted heart monitoring app, is more effective at detecting atrial fibrillation than a dedicated electrocardiogram (EKG) wristband, and nearly as effective as a 12-lead EKG machine. As spotted by AppleInsider, the study raises the prospect that advanced AI will enable wearable devices to detect dangerous medical conditions using only basic sensors.
The study, “Passive Detection of Atrial Fibrillation Using a Commercially Available Smartwatch,” included users of the Apple Watch app Cardiogram and its DeepHeart neural network, co-developed with researchers at the University of California, San Francisco. DeepHeart was trained to detect irregularities in the timing of heartbeats, canvassing over 139 million heart rate measurements from nearly 10,000 people to create fibrillation risk scores.
Researchers found that the Cardiogram/DeepHeart pairing had 98 percent sensitivity and 90.2 percent specificity in detecting atrial fibrillation compared with a 12-lead EKG machine, versus 67.7 percent sensitivity and 67.6 percent specificity coming from participants who self-reported persistent fibrillations. In other words, passively wearing an Apple Watch with Cardiogram installed was nearly as accurate as being actively checked by a medical-grade apparatus, and far more accurate than attempting to detect an issue without sensors.
Additionally, Cardiogram’s 98/90.2 percent results appear to be superior to those obtained with the KardiaBand, a $200 EKG wristband accessory released for the Apple Watch. Using a simpler two-lead design, the KardiaBand received FDA clearance specifically for detecting atrial fibrillation, achieving accuracy rates of 93/84 percent on its own, or 99/83 percent with a physician interpreting the recording. Kardia notably charges fees for physician EKG reads.
The DeepHeart researchers say that this is “the first large-scale, peer-reviewed study to appear in a medical journal demonstrating consumer wearables’ ability to detect a major health condition.” They also noted that DeepHeart’s pre-training and semi-supervised learning yielded “a significant increase in accuracy compared to similar artificial intelligence research,” suggesting that having skilled researchers involved in the initial setup of the AI can considerably improve its utility for health diagnoses.
Apple has reportedly been working on expanded heart monitoring hardware for Apple Watches, and is currently working on a similar heart study with Stanford University, though it’s open by invitation only. Cardiogram is available for everyone as a free download from the iOS App Store.