Impact of ambulatory ECG monitoring duration on precision of AI-based AF burden estimation

Atrial fibrillation (AF) burden has emerged as a prognostic and therapeutic target in patients with paroxysmal AF. Studies have shown that longer ECG monitoring improves AF detection and precision of AF burden estimation [1, 2]. However, majority of these studies relied on simulated ambulatory ECG (AECG) strategies rather than observed recordings. There is limited real-world evidence for how monitoring duration impacts AF detection and burden estimation. We analysed 7-day AECG recordings to estimate detection sensitivity and burden accuracy by monitoring duration.

We retrospectively screened 1,886 ambulatory ECG (AECG) recordings of ≥7-day duration collected between 2020 and 2025. To limit bias stemming from hardware, electrode choice, or clinical workflows, we selected six centres in Switzerland, Denmark, and Greece.

AECG inclusion criteria were adults with an AI-derived 7-day AF burden >0% and <100% (excluding permanent AF).

AF burden was calculated as the proportion of time in AF during a specified monitoring period. Cumulative monitoring periods were specified as: first 24, 48, 72, 96, 120, and 144 hours. Full 7-day monitoring period served as the reference.

We used Cardiomatics, a CE-marked software platform powered by a validated deep learning algorithm to estimate AF burden. Cardiomatics previously shown to match expert cardiologists in performance [3].

Intraclass correlation coefficient (ICC), coefficient of determination (R²),and Bland–Altman analysis was used to assess agreement between shorter-duration burden estimates and the 7-day reference.

After applying inclusion criteria 143 recordings with full 7-day monitoring period remained for analysis (63.4% men; 70.7% aged ≥65 years).

- Agreement metrics improved steeply after day 3: R² increased from 0.56 to 0.98 from one day to six days (Figure 1).

- AF burden estimation decreased with increasing monitoring duration. Bland–Altman analysis showed the mean bias shrinking from +3.3% after one day to ~0% after six days, with limits of agreement narrowing correspondingly (Figure 2).

- Bland–Altman limits of agreement show that the absolute error in AF burden estimation can reach 40 percentage points after one monitoring day, but this range contracts markedly with longer recordings: shrinking to 5 pp by day 6.

Study shows that the accuracy of AI-based atrial fibrillation (AF) burden estimation rises markedly as ambulatory ECG (AECG) monitoring is extended from one to seven days. Although short recordings may suffice for simple AF detection, our data demonstrate that reliable quantification of AF burden clearly requires longer monitoring periods.

These findings demonstrate the advantage of long-duration AECG and AI-enabled analysis as applicable solution for precise AF burden assessment.Figure 1.Figure 2.