Association of AF pattern properties and accuracy of AF burden estimation: a computational analysis

Background/Introduction Atrial fibrillation (AF) burden (the percentage of time spent in AF over a certain follow-up period) has emerged as an important metric for AF diagnosis and may have prognostic relevance. While continuous rhythm monitoring is the gold-standard to determine the true AF burden, its use is restricted to select subpopulations. Intermittent rhythm monitoring (IRM), e.g., using handheld ECGs or Holter monitors can provide an estimate, but the accuracy of different IRM strategies for estimating AF burden and its dependence on different temporal patterns of AF episodes are unknown. Purpose To explore how different AF episode patterns and different IRM strategies influence the estimation of AF burden using a computational approach to enable perfect control over AF pattern properties. Methods Four datasets of virtual patients with 5% AF burden were generated to investigate the impact of episode duration, frequency, and AF density (i.e., clustering of AF episodes) on AF Burden estimation. These include low-frequency long-lasting episodes (LFLE; episode duration 720±120 min) and high-frequency short-lasting episodes (HFSE; episode duration 120±20 min), each with a high or low AF density (Figure 1). Intermittent ECG recordings of 30 seconds were simulated once every 6 weeks, 4 weeks, 2 weeks, 1 week, or twice per week. Secondly, Holter recordings of 12h, 24h, 48h, 72h, 120h, performed five times per year were simulated. Results (Figure 2) LFLE and HFSE traces had an average of 37 and 209 episodes per year, respectively. Increasing the frequency of ECG-based IRM resulted in higher precision (narrower confidence interval; i.e., fewer virtual patients with an estimate far from the true AF burden) of the estimated AF burden for each AF pattern (e.g. estimated AF burden in LFLE high density of 8.62±5.42% and 5.38±1.58% with once every 6 weeks and twice-weekly assessment). By contrast, an increase in Holter duration had limited effect on the precision in high density populations (e.g. LFLE high density, 24h Holter: 4.89±7.23% and 120h Holter: 4.96±5.81%). Performing ECG-based IRM twice per week results in little difference in precision of the estimated AF Burden between different AF patterns (low density: 4.88±2.02% (LFLE) and 4.69±2.05% (HFSE), high density: 5.38±1.58% (LFLE) and 5.17±1.59% (HFSE)). Performing 120h Holter recordings in the HFSE low density population results in an estimated burden with high precision 4.78±0.55% while in the LFLE low density population, the precision is lower (4.71±2.29%). Conclusion At a population level, both ECG (once every 6 weeks to twice per week) and Holter (5 per year) IRM strategies provide an accurate estimation of average AF burden. However, the precision of the different approaches depends on the AF pattern. For the patterns assessed, a twice weekly ECG IRM strategy is the most precise approach.