Wristband-based automated cardiac arrest detection and alerting: Algorithm performance in relation to the left ventricular function

Introduction Unwitnessed out-of-hospital cardiac arrest (OHCA) has poor survival rates, due to the delayed recognition. Wearable technologies offer a promising approach to automatically detect and alert for cardiac arrest, potentially improving survival by faster emergency response. Purpose In the DETECT-studies, a wristband-based photoplethysmography (PPG) algorithm for cardiac arrest detection has been developed. Since left ventricular function influences the PPG signal, this study evaluates algorithm performance in relation to the left ventricular ejection fraction (LVEF). Methods This post hoc analysis included patients previously enrolled in the DETECT-1a and DETECT-1b studies, comprising of individuals undergoing ventricular tachycardia (VT) ablation, transcatheter aortic valve implantation (TAVI), and S-ICD implantation. Patients without known calculated LVEF were excluded. Cardiac arrest events included pulseless ventricular tachycardia (pVT) during VT ablation, rapid ventricular pacing (RVP) with aortic balloon inflation during TAVI, and ventricular fibrillation (VF) induction during S-ICD implantation. During the procedures, patients wore a PPG-based wristband. Cardiac arrest was confirmed using continuous ECG monitoring and invasive arterial blood pressure measurements. All recorded PPG data were analysed using the previously developed cardiac arrest detection algorithm. Patients were categorized into poor/moderately reduced LVEF (LVEF <40%) and mildly reduced/normal LVEF (LVEF ≥40%). Sensitivity for cardiac arrest detection and false-positive rates were assessed. Results A total of 219 patients were included, of whom 42 had an LVEF <40% and 177 had an LVEF ≥40%. The median LVEF was 30% (IQR 23–35%) and 55% (IQR 49–60%) respectively. Patients in the LVEF <40% group more frequently underwent VT ablation and S-ICD procedures than in the LVEF>40% group. In the LVEF <40% group, 45 cardiac arrests were induced; 25 pVT, 3 VF, and 17 RVP events. In the LVEF ≥40% group, a total of 184 CAs occurred; 10 pVT, 7 VF, and 167 RVP events. In patients with LVEF <40%, the algorithm correctly identified 40 CAs, corresponding to a sensitivity of 89% (95% CI 80–95%). In the LVEF ≥40% group, 181 CAs were correctly identified, yielding a sensitivity of 98% (95% CI 97-100%), see Figure 1 (p=0.028 for comparison between LVEF groups). Four false-positive alarms occurred during 107 hours of PPG monitoring in patients with LVEF <40%, compared with 12 false positives during 346 hours in the LVEF ≥40% group. Conclusions Sensitivity for detecting cardiac arrest was lower in patients with LVEF <40%, although performance in this group remained clinically acceptable. Missed cardiac arrest events were primarily VTs, which may explain the lower sensitivity. Future studies should validate performance in real-life cardiac arrests and aim to reduce false-positive alarms during daily-life use.Sensitivity in LVEF groups