Optimizing entropy-based heart rate variability metrics: fixed tolerance improves congestive heart failure classification

OBJECTIVE: Entropy-based heart rate variability (HRV) metrics are widely used to characterize heartbeat dynamics and distinguish physiological from pathological states, but their classification performance is highly sensitive to parameter selection. This study investigated a fixed-tolerance approach derived from a healthy reference population as an alternative to conventional subject-specific tolerance normalization and assessed how entropy-parameter selection influences classification performance. Approach: Approximate entropy (ApEn), sample entropy (SampEn), and multiscale entropy (MSE) were evaluated using both conventional and fixed-tolerance formulations in 24-hour Holter recordings from 274 healthy subjects and 946 patients with congestive heart failure (CHF). Classification performance was quantified using the area under the receiver operating characteristic curve (AUC). Parameter sensitivity was assessed through systematic sweeps of the embedding dimension, tolerance coefficient, and MSE scale range. Main results: Classification performance varied substantially across the parameter space, and the embedding-dimension and tolerance values commonly used in the literature were generally suboptimal. The fixed-tolerance implementation consistently outperformed the conventional subject-specific approach, achieving AUC values of 0.78 (+0.03) for ApEn, 0.77 (+0.09) for SampEn, and 0.74 (+0.14) for MSE. For MSE, tolerance selection had a substantially larger influence on classification performance than scale-dependent parameterization. Significance: The results suggest that a substantial portion of the classification potential of entropy-based HRV analysis remains unexploited under conventional tolerance normalization and parameter selection. More generally, entropy-based HRV metrics appear to be more sensitive to methodological choices than is commonly assumed. The fixed-tolerance approach provides a simple and computationally inexpensive strategy for improving entropy-based CHF classification, although validation in independent datasets is needed to establish its broader applicability.