A reproducible event-driven analytical framework for ECG-derived skin sympathetic nerve activity analysis: towards afferent signal processing for closed-loop neuromodulation

Background Quantifying skin sympathetic nerve activity (SKNA) is challenging because sympathetic signals are noisy, burst-like, and processing-dependent. Electrocardiogram-derived skin sympathetic nerve activity (ECG-derived SKNA) provides a non-invasive surrogate of sympathetic activity. Many existing SKNA analysis methods rely on semi-manual processing or static measurements, with limited attention to dynamic event-based physiology, reproducibility across large cohorts, or near-real-time applications. Objective To develop and evaluate a reproducible event-driven computational framework for analysing ECG-derived SKNA during transient sympathetic activation driven by simulated apnoeas, with potential application to future closed-loop neuromodulation. Methods The framework integrates electrocardiography preprocessing, multi-channel SKNA extraction, adaptive burst detection, event-centred physiological analysis, and sequential replay. It was developed using recordings from 50 pigs during apnoeas induced by intermittent negative upper airway pressure (INAP). Adaptive thresholds were defined as the maximum of a two-component Gaussian mixture model intersection, the baseline 95 th percentile, and a baseline variability criterion. Sequential offline replay was performed on the pre-processed SKNA, with each replay decision restricted to SKNA samples available up to that decision point. Burst occupancy was calculated in a 30-s trailing window updated every 10 s, with trigger activation requiring occupancy above 5% in two consecutive windows. The pipeline was also applied to an independent multi-event INAP recording. Results During sequential replay, a trigger was present during INAP in 43 of 50 recordings (86%). Among the 49 recordings in which the detector was OFF at INAP onset, 42 (85.7%) developed a new post-onset trigger activation, with a median detection latency of 24.0 s (IQR 18.9–54.3 s). In the independent multi-event recording, all three INAP events generated replay triggers, with detection latencies of 28.4, 30.0, and 20.0 s. Continuous replay of an intervention recording also showed event-specific SKNA and trigger outputs before and during dorsal root ganglion (DRG) stimulation. Conclusion This study presents a reproducible event-driven framework for analysing SKNA. The framework captures heterogeneous SKNA responses and allows sequential evaluation as new SKNA samples become available. By enabling dynamic SKNA responses to be monitored during physiological perturbations, the framework provides a practical foundation for prospective real-time validation and future closed-loop neuromodulation using SKNA-derived physiological feedback.