Predictive diagnosis of fatal heart rhythms using wearables

Sudden cardiac death causes more than 300,000 deaths annually in the US. Our research goal is to develop a continuous cardiac monitoring system that utilizes current wearable devices and is capable of not just detecting arrhythmia but also to predict life-threatening arrhythmia a few minutes before it would actually happen. The monitoring system should provide a diagnosis based on analyzing a few-minutes of heart-rate data streams. In order to verify the feasibility of this approach, we have developed a prototype and evaluated its capabilities. The prototype is based on a two-tier data analytics approach and utilizes multiple gradient boosting machine learning models. The system was tested for predicting four different life-threatening arrhythmias solely on realistic heart-rate readings and also tested the atrial fibrillation recognition capability. The prototype scored 91.6% and 93.9% accuracy respectively. These preliminary results validate the feasibility of our approach to predict arrhythmia in real-time from heart-rate observations.