Reliable Detection of Atrial Fibrillation with a Medical Wearable during Inpatient Conditions

Area of Science:

• Cardiology
• Medical Technology
• Biomedical Engineering

Background:

• Atrial fibrillation (AF) is a prevalent arrhythmia significantly impacting patient morbidity and mortality.
• Detecting asymptomatic AF presents a clinical challenge, necessitating improved diagnostic tools.

Purpose of the Study:

• To evaluate the sensitivity and specificity of a non-invasive medical wearable for detecting atrial fibrillation (AF).
• To assess the performance of different algorithms, including deep neural networks, for AF detection using wearable photoplethysmography data.

Main Methods:

• An observational trial involving 102 patients with AF, comparing a novel medical wearable against a standard ECG Holter over 24 hours.
• The wearable utilizes photoplethysmography (PPG) technology to analyze pulse rates and inter-beat intervals from 5-minute datasets.
• Deep neural networks and other algorithms were applied to PPG data for automated AF detection.

Main Results:

• Over 2306 hours of parallel recordings, 1781 hours (77.2%) were interpretable by algorithms.
• The wearable achieved a sensitivity of 95.2% and specificity of 92.5% (AUC 0.97) for AF detection.
• Deep neural network algorithms improved sensitivity to 96.0% and specificity to 99.0% (AUC 0.98).

Conclusions:

• Non-invasive AF detection using a medical wearable is feasible, particularly in hospitalized, physically active patients.
• Deep neural network integration significantly enhances the reliability and continuous monitoring capabilities for AF detection.
• This technology offers a promising approach for early and ongoing detection of atrial fibrillation.