Out-of-hospital cardiac arrest is a critical emergency that demands immediate recognition and response; in cases of sudden, unwitnessed cardiac arrest, survival chances are extremely low. One of the key indicators of cardiac arrest is the sudden loss of a pulse. The use of automated biosensors to detect unwitnessed cardiac arrest and prompt medical assistance could significantly improve survival rates, given the critical role of time in determining outcomes. However, for such systems to be effective, minimizing the false positive burden on emergency medical services is essential. Researchers have now developed an artificial intelligence (AI) system for a smartwatch that can detect signs of cardiac arrest and automatically alert emergency responders. This machine-learning algorithm, partially trained on data from patients whose hearts were intentionally stopped during medical procedures, is capable of recognizing the key indicators of cardiac arrest. While the researchers believe this system could potentially save lives, further testing is necessary to confirm its effectiveness.

The smartwatch algorithm, which leverages machine learning, was developed by Google Research (Mountain View, CA, USA), and is designed to meet performance standards suitable for large-scale use. The researchers first utilized photoplethysmography (PPG) to demonstrate that PPG measurements of peripheral pulselessness (induced by an arterial occlusion model) mirror the signal pattern of pulselessness seen in ventricular fibrillation (VF), a common arrhythmia associated with cardiac arrest. By analyzing the similarity between the PPG signals from VF and those from the occlusion model, they developed and validated a pulse loss detection algorithm using data collected from both the simulated model and real-world conditions.

Once the algorithm was developed, the researchers tested it in real-world scenarios. The results, published in Nature, showed that the system generated one unintentional emergency call per 21.67 user-years across two prospective studies. The algorithm’s sensitivity was 67.23% (with a 95% confidence interval of 64.32%–70.05%) when tested in a simulated arterial occlusion model of cardiac arrest. These findings indicate that this new technology offers a promising solution for wearable detection of sudden pulse loss while minimizing the societal impact of false alarms.

http://www.allmycbs.com/en/index.asp