Article Impact Level: HIGH Data Quality: STRONG Summary of Nature, 1–3. https://doi.org/10.1038/s41586-025-08810-9 Dr. Kamal Shah et al.
Points
- The study developed a machine learning algorithm using photoplethysmography (PPG) to detect sudden loss of pulse, a key indicator of out-of-hospital cardiac arrest (OHCA).
- The algorithm was tested on 220 daily users, 135 controlled setting participants with induced arterial occlusion, and 21 stunt performers simulating OHCA collapse.
- The system achieved 99.99% specificity for non-events, a sensitivity of 67.23% for pulselessness detection, and one false emergency call per 21.67 user-years.
- The algorithm detected pulselessness within 57 seconds, followed by a 20-second user confirmation check before initiating an emergency call. Its sensitivity ranged from 53% (simulated collapse) to 72% (motionless pulseless events).
- While the system could significantly enhance survival rates for unwitnessed cardiac arrests, further refinements are needed to reduce false positives and improve sensitivity in real-world conditions.
Summary
This study investigates using a machine learning-based algorithm running on a smartwatch to detect sudden loss of pulse, a key indicator of out-of-hospital cardiac arrest (OHCA). Researchers developed an algorithm using photoplethysmography (PPG) to measure pulse changes associated with pulselessness, such as that caused by ventricular fibrillation (VF) or arterial occlusion. The algorithm was trained using data from controlled environments, such as ventricular fibrillation simulations and real-world conditions. The team aimed to validate the algorithm’s ability to autonomously detect these events and trigger emergency responses, thereby improving survival rates for OHCA patients.
The algorithm was evaluated across six cohorts, including 220 participants who wore the smartwatch during daily activities, 135 participants in a controlled setting with induced arterial occlusion, and 21 stunt performers simulating OHCA collapse. The system demonstrated high specificity, with 99.99% of non-event cases correctly classified and a moderate sensitivity of 67.23% (95% CI, 64.32%–70.05%) in detecting induced pulselessness. Regarding false positives, there was one unintentional emergency call per 21.67 user-years. The sensitivity for detecting simulated collapse events was 53%, while it was 72% for motionless pulseless events. The algorithm detected pulselessness within 57 seconds, followed by a 20-second user response check before initiating an emergency call.
The findings suggest that a smartwatch-based system for automated detection of cardiac arrest could significantly improve survival rates by enabling faster emergency responses, especially for unwitnessed events. However, further refinement is needed to reduce false positives and increase sensitivity in real-world conditions. This research highlights the potential of wearable devices in advancing pre-hospital emergency care while addressing the challenge of minimizing unnecessary alerts and associated healthcare system burdens.
Link to the article: https://www.nature.com/articles/s41586-025-08810-9
References Shah, K., Wang, A., Chen, Y., Munjal, J., Chhabra, S., Stange, A., Wei, E., Phan, T., Giest, T., Hawkins, B., Puppala, D., Silver, E., Cai, L., Rajagopalan, S., Shi, E., Lee, Y.-L., Wimmer, M., Rudrapatna, P., Rea, T., … Sunshine, J. (2025). Automated loss of pulse detection on a consumer smartwatch. Nature, 1–3. https://doi.org/10.1038/s41586-025-08810-9
