Year: 2023
Type: dissertation
Study design: Observational Study
Abstract: Atrial fibrillation (AF) is the most common tachyarrhythmia and a significant cause of cardioembolic strokes.AF is often intermittent and asymptomatic, making its detection a major clinical challenge.Photoplethysmography (PPG) provides a promising option for atrial fibrillation detection.The primary objective of this dissertation was to evaluate the accuracy of a PPG wrist band in an algorithmbased detection of AF.Secondary objectives included signal quality assessment and evaluation of the feasibility and user experience of the method. The dissertation comprises three clinical trials and original publications (I-III).Patients with AF or sinus rhythm (SR) were recruited in the emergency care departments and cardiological wards in three university or central hospitals in Finland between May 2017 and December 2019.A total of 1504 patients were screened, 793 recruited and 745 (AF n = 351, sinus rhythm, SR, n = 394) included in the final analyses.PPG was recorded with a wrist band device.Simultaneously recorded electrocardiogram (ECG) data served as the gold standard for rhythm analysis in all studies.The study setting in all studies was a prospective casecontrol diagnostic accuracy clinical study.I) The PPG signal was evaluated in detection of individual pulses, and two AF detection algorithms originally developed for ECG signals were used for the PPG-based detection of AF.II) Ten robust PPG signal features were evaluated for the detection of AF, including pulse interval variability and amplitude features, and a novel morphology-based autocorrelation (AC) feature.A novel AF detection algorithm was developed and validated.III) Continuous PPGs of ambulatory patients were recorded for 24 h and split into 10-, 20-, 30-and 60-minute time frames.The best time frame was 8 selected, and AF alarms were generated to simulate continuous real-time AF monitoring.Sensitivities and positive predictive values (PPVs) of the AF alarms were evaluated.User experiences of the patients were collected.I) The wrist band PPG achieved individual pulse detection with a sensitivity of 91.7% and a PPV of 97.5% for AF, and with a sensitivity of 99.4% and PPV of 98.1% for SR.The pulse detection sensitivity was lower 86.7% with AF duration < 48 h than with duration ≥ 48 h 95.1%.AF detection sensitivities were 96.2% and 95.3%, and specificities 98.1% with the two algorithms.II) The univariate predictor model based on the novel PPG signal morphology-based AC feature detected AF with an area under the receiver operating characteristic curve (AUC) value of 0.982, which was higher than with any other pulse interval or amplitude-based univariate model.The AC PPG feature was also the most significant individual feature (p < 1 x 10 -5 ) included in the multivariate predictor model.This optimized AF detection algorithm achieved AUC of 0.993 for the detection of AF.III) Of the evaluated time frames, the 30-minute time frame yielded the highest F1-score (0.9536), and the AF alarms indentified AF correctly in 72 out of 76 true AF patients (sensitivity 94.7%), and only three out of 97 SR patients received a false AF diagnosis (specificity 96.9%).PPG signal quality was significantly better during nighttime than daytime with 67.3% of signal data of sufficient quality signal data for rhythm analysis vs. 30.5%.The sensitivity and PPV of the simulated AF alarms were 78.2% and 97.2% at night, and 49.3% and 97.0% during the daytime.Most of the patients (82%) were willing to use the device at home.In conclusion, I) a PPG wrist band detected individual pulses reliably, but the sensitivity was lower in recent-onset AF than in late AF.The algorithms detected AF reliably from the PPG recordings.II) The novel AC feature based on pulse wave morphology detected AF independently with high sensitivity and specificity without the need for pulse detection.The established algorithm combining pulse wave morphology with information from pulse-interval variability achieved in detection of AF with high accuracy.III) Continuous PPG wrist-band monitoring with an automated algorithm provided reliable AF detection alarms.The PPV remained high throughout the 24-h monitoring regardless of lower signal quality during daytime.In addition, the positive user experience, and AF detection results suggest that the method could enable continuous rhythm monitoring and screening for AF.
Author: Väliaho, Eemu-Samuli
Institution: University of Eastern Finland
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7215505653