What does this research mean for the field?

The PPG-AF AI model demonstrates strong correlation for AF duration (r=0.92) and good sensitivity (91.5%) and specificity (97.2%) for AF detection compared to Holter monitoring. Novelty: ClaimNovelty.NOVEL_FINDING. Consensus alignment: ConsensusAlignment.SUPPORTS_CONSENSUS.

What question did this study set out to answer?

To develop a quantitative method for assessing atrial fibrillation progression using a PPG-based smartwatch.

February 8, 2026

Multi-dimensional quantitative assessment approach of atrial fibrillation progression with photoplethysmograph-based smart watch

Key Result

PPG-based smartwatch AI model detected AF with 91.5% sensitivity, 97.2% specificity, and strongly correlated AF duration (r=0.92) to ECG Holter data.

Key Points

To develop a quantitative method for assessing atrial fibrillation progression using a PPG-based smartwatch.
Prospective cohort study with patients having paroxysmal AF monitored using PPG and Holter systems.
PPG data recorded every minute; a minimum of 500 quality readings required for analysis.
Developed PPG-AF AI model for AF detection, focusing on minimizing motion artifacts and detecting key PPG features.
Evaluated five AF progression features: number and duration of AF episodes, aggregation, circadian rhythm, and heart rate.
Total of 145 patients identified with 116 AF episodes.
Strong correlation for AF duration between PPG-AF model and ECG (r=0.92, p<0.0001).
Circadian rhythm feature showed high correlation (average real variability=0.81, p<0.0001).
Moderate correlation for number and aggregation features (r=0.709 and r=0.6576 respectively, all p<0.0001).
PPG-AF AI model had an output rate of 86.7%, with sensitivity, specificity, and accuracy at 91.5%, 97.2%, and 95.7% respectively.

Structured PICO

Does a PPG-based smartwatch with an AI model accurately assess atrial fibrillation progression compared to a 24-hour Holter monitor in patients with paroxysmal AF?

Population

145 patients with paroxysmal AF, mean age 63±14 years, 96 male, from a single center in China. Inclusion required ≥ 500 PPG recordings with adequate signal quality.

Intervention

PPG-based smartwatch with a PPG-AF AI model for multi-dimensional quantitative assessment of AF progression (measuring number, duration, aggregation, circadian rhythm, and heart rate)

Comparator

24-hour Holter monitor (simultaneous monitoring)

Outcome

Correlation between AF progression features detected by PPG-smart watch and 24-hour Holtersurrogate

A PPG-based smartwatch AI model provides reliable and valid multi-dimensional assessment of atrial fibrillation progression compared to standard 24-hour Holter monitoring.

Main Result

Absolute Event Rate: 0% vs 0%

Abstract

Abstract Background The burden of atrial ﬁbrillation (AF) is usually as the percentage of time in AF. Nonetheless , the technological innovations may enable multi-dimensional precise AF progression estimation in daily life. Objective To develop a quantitative approach for understanding the patterns of AF progression with photoplethysmography (PPG)-based smart watch. Methods In this prospective cohort study,the patients with paroxysma l AF were enrolled between March 1, 2024 to January 20, 2025 from sixth Medical center General Hospital China. All patients were simultaneously monitored cardiac rhythm with PPG-based smartwatch and 24-hour Holter. The pulse rhythm was measured every one minute with PPG-based smart watch. The patients with ≥ 500 PPG recordings with adequate signal quality were included in the analysis. We developed a PPG-AF AI model using time-series PPG data to accurately catch up the robust PPG feature for efficient AF detection through minimizing motion artifacts and the premature atrial and ventricular contractions. Then multi-dimensional quantitative assessment approach of AF progression with PPG- AF model was developed consisted of five features of AF progression (Figure 1):1) Number of AF episodes;2) Duration of AF episodes;3) AF aggregation;4) circadian rhythm;5) Heart rate. The primary endpoint was the correlation between AF progression features detected by PPG-smart watch and 24-hour Holter. Secondary outcome was the sensitivity speciﬁcity accuracy and output rate of PPG-AF AI model. Result One hundred and forty-ﬁve patients (96 male, age, mean±SD, 63±14 years) detected with total 116 AF episodes. The correlation between Duration feature detected with PPG-AF AI model and ECG was strong (AF lasting over 6 mins, spearman correlation coefﬁcient, r=0.92, p 0.0001). There was a close correlation of circadian rhythm feature between PPG-AF AI model and ECG (e.g, average real variability=0.81, p 0.0001). Number feature and Aggregation feature of PPG-AF AI model moderately correlated with ECG (AF episodes over 1 hour, r=0.709; AF density, r=0.6576, all p 0.0001) (Figure 2A-2E) . AF burden by PPG-AF AI model and ECG was consistent for 24 hours (Figure 2F). The output rate of PPG-AF AI model with was 86.7%. The sensitivity, speciﬁcity, and accuracy (%, CI) of AF detection were 0.915 0.879, 0.951, 0.9720.959,0.985, and 0.9570.940, 0.974, compared to 24-hour Holter. Conclusion Multi-dimensional quantitative assessment approach with PPG based smartwatch has good reliability and validity in the evaluation of AF progression, especially in the Duration, Circadian rhythm, Number and Aggregation of AF events, and it can better evaluate the AF burden in daily life.

Bookmark