Ultra-Short HRV from holter ECG cannot replace 5-Minute HRV for short-term AF onset prediction

Introduction The autonomic nervous system plays a key role in the initiation of atrial fibrillation (AF), and its modulation is commonly inferred from heart rate variability (HRV). In remote monitoring and wearable devices, HRV is frequently computed from ECG or photoplethysmography (PPG) over recording periods shorter than five minutes. Although 5 minutes remains the recommended standard for short-term HRV, prior validation studies in sinus rhythm under resting/stationary conditions, and in wearable PPG settings (e.g., during sleep), suggest that ultra-short 1-minute RMSSD (often lnRMSSD) can provide estimates comparable to standard 5-minute measures. However, whether this interchangeability persists in the dynamic, potentially non-stationary period immediately preceding AF onset is unknown, and this question is critical for HRV-based short-term AF onset prediction using wearable-grade recordings. Objective To evaluate whether ultra-short-term HRV features extracted from 1-minute ECG windows can replace standard 5-minute HRV features for HRV-based short-term AF onset prediction. Methods We analysed two-lead Holter ECG recordings from 927 patients (987 recordings). We selected 520 patients (569 recordings) who had an AF episode lasting >5 minutes and preceded by ≥35 minutes of normal sinus rhythm. Time- and frequency-domain HRV parameters were computed using sliding windows of 5 and 1 minute, advanced minute-by-minute to characterise HRV dynamics over the 30 minutes directly preceding AF onset. AF episodes were excluded when at least one HRV computation window contained >50% ectopic beats. The final analysis included 807 episodes (538 recordings, 495 patients). Agreement between 1-minute and 5-minute HRV estimates was assessed using Pearson correlation and Bland–Altman limits of agreement (LoA). Results The average Pearson correlation coefficient was <0.65 for all HRV parameters except mean RR intervals, indicating weak-to-moderate association between ultra-short and standard estimates. Bland–Altman analyses showed wide LoA for all parameters, consistently exceeding the typical pre-onset variation observed for each feature, indicating poor agreement between 1-minute and 5-minute HRV measurements in the 30 minutes preceding AF onset (Fig. 1- 2). Conclusion One-minute HRV windows are not interchangeable with standard 5-minute HRV in the pre-AF period and fail to provide reliable estimates for HRV-based short-term AF onset prediction, supporting the need for longer recordings for robust autonomic assessment.