Background and Aim. Poincaré plot (PP) analysis provides a nonlinear framework to characterize the temporal structure of physiological signals. While widely applied to heart rate variability, its direct application to atrial fibrillatory waves (f-waves) remains unexplored. A critical yet overlooked aspect is the selection of the lag parameter, as a suitable value is required to obtain a meaningful PP representation of the underlying atrial dynamics. This study evaluates different lag-selection criteria and their impact on PP-derived descriptors for discriminating clinical outcomes in persistent atrial fibrillation (AF).
Methods. Preprocedural ECG recordings from 52 patients with persistent AF undergoing catheter ablation (CA) were retrospectively analyzed. Up to three 6 s f-wave segments per patient were extracted from lead V1. Lag values were estimated using well-established criteria, including fractions of the dominant cycle length (AFCL/4), autocorrelation-derived measures, such as the first zero crossing (AFC-ZC) and the first sign change in the second derivative of the AFC (AFC-2ndD), and the first local minimum of auto–mutual information function (AMIF-min). Patient-specific and population-averaged lags were determined and used as embedding delay to construct PPs. Standard ellipse-based PP descriptors were calculated and their capability to discriminate AF recurrence after CA was evaluated.
Results. Lag selection influenced PP-derived descriptors and their ability to discriminate between recurrence and non-recurrence groups. Tailored lag-selection criteria yielded improved discriminative performance compared to fixed lag values. Among the evaluated criteria, AFC-2ndD yielded the strongest statistical significance across descriptors. The fitted ellipse axis ratio (SD1/SD2) was the most discriminative descriptor (p-value < 1e-8) and achieved the highest classification performance (AUC = 83.2%).
Conclusions. Lag selection is a critical factor in PP analysis of f-waves in persistent AF. Patient-specific strategies improved the predictive value of PP-derived descriptors, supporting their potential as non-invasive markers for CA outcome stratification.