Automated cardiac arrest detection using wrist-derived photoplethysmography during withdrawal of life-sustaining treatment: a prospective clinical validation study

Background: Automated cardiac arrest detection and alerting using wearable technology has the potential to shorten recognition delays for unwitnessed out-of-hospital cardiac arrest. In DETECT-1A and -1B, a photoplethysmography-based detection model was developed and validated in patients with induced cardiac arrest. This study evaluates model performance in true cardiac arrests following withdrawal of life-sustaining treatment. Methods: Prospective, single-center study in adult ICU patients with planned withdrawal of life-sustaining treatment. Patients wore a photoplethysmography-wristband (CardioWatch) until death. Continuous ECG and invasive arterial pressure served as reference standards. The previously developed rule-based algorithm was refined using two training cohorts and evaluated in a separate test cohort. Endpoints were sensitivity for cardiac arrest detection and false positive alerts. Findings: Forty-four patients were included (training 1: n = 10; training 2: n = 11; test: n = 23), median age 65 years; 75% male, all with non-shockable cardiac arrest. Sensitivity for cardiac arrest detection was 100% (10/10; 95% confidence interval [CI] 66–100%) and 90% (9/10; 95% CI 54–99%), in training 1 and 2, respectively. In the test set, sensitivity was 100% (23/23; 95% CI 82–100%), with one false positive alert. Cardiac arrest was detected at a mean arterial pressure of 30 mmHg (IQR 24–35) and pulse pressure of 13 mmHg (IQR 11–19). Interpretation: Cardiac arrest can be detected with high sensitivity using wrist-derived photoplethysmography, providing first evidence on model performance in true cardiac arrest, specifically in non-shockable cases. Findings support further development of wearable-based cardiac arrest detection technologies to enable earlier recognition for unwitnessed cardiac arrest. Funding: Dutch Heart Foundation, Radboudumc.