Dynamic heart rate variability for classification of arrhythmia burden

Purpose: Heart rate variability (HRV) may provide information beyond heart rate (HR) for identifying arrhythmia burden, but the value of temporal HRV variation remains unclear. This study developed a dynamic HRV measure quantifying the interquartile range of window-level root mean square of successive RR-interval differences (IQRRMSSD) and evaluated its utility in classifying total abnormal-beat and premature ventricular contraction (PVC) burden compared with the conventional RR-based HRV measure, RMSSD. Methods: Manually annotated 30-minute ECG recordings from 43 individuals were analyzed using overlapping windows. A linear support vector classifier was used to classify individuals as having high or low arrhythmia burden, defined using 1% and 10% burden thresholds. Four feature sets were evaluated: HR alone, HR with RMSSD, HR with IQRRMSSD, and all three measures. Performance was assessed using repeated stratified five-fold cross-validation. Results: HRV-containing feature sets generally outperformed HR alone, depending on outcome and threshold. At the 1% threshold, IQRRMSSD showed particularly high sensitivity for total abnormal-beat and PVC burden ($0.970 \pm 0.064$ and $0.882 \pm 0.193$, respectively). At the 10% threshold, conventional RMSSD showed the strongest overall discrimination, achieving a ROC-AUC of 0.947 for PVC burden. Combined features sometimes improved the sensitivity-specificity balance but not overall discrimination. Conclusion: These findings suggest that a conventional RR-based HRV measure is informative for elevated ectopic burden, while a dynamic HRV measure may provide complementary information for lower-burden abnormalities.