How Accurately Can We Detect Atrial Fibrillation Using Photoplethysmography Data Measured in Daily Life?

Photoplethysmography (PPG) is an unobtrusive measurement modality recently explored for the detection of atrial fibrillation (AF). When used in wrist-worn applications, PPG-monitoring can be used for long-term monitoring in daily life, which is beneficial when aiming to detect AF. The objective of this study was to investigate whether the performance of an AF detection model trained and tested on short measurements is generalizable to measurements in daily life. PPG, accelerometer, as well as reference ECG data were measured from 32 subjects (13 continuous AF, 19 no AF) in 24-hour monitoring in daily life. An AF detection model combining inter-pulse interval features was trained to classify AF or non-AF. Short measurements were obtained by selecting a 5-minute segment from each 24-hour recording and used for training the model. The accuracy was tested on both 5-minute segments and 24-hour data. Sensitivity, specificity, and accuracy of the model were 98.90%, 99.03%, and 98.98% with 5-minute data and 96.94%, 91.99%, and 93.91% with 24-hour data. False positive detections per patient worsened from being on average none during short recordings to (mean ± sd) 467 ± 328 in daily life. Thus, testing the AF detection models intended for long-term PPG-monitoring is essential with data from daily life in order to obtain a realistic estimate of the accuracy.