Skip to content

Circadian Phase Locking of Epilepsy Seizures in Wearable Data: A Single-Patient Case Study

Home / Papers / Circadian Phase Locking of Epilepsy Seizures in Wearable Data: A Single-Patient Case Study

Circadian Phase Locking of Epilepsy Seizures in Wearable Data: A Single-Patient Case Study

Authors: Berenika Ewart-James, Matthew Wragg, Nawid Keshtmand, Amberly Brigden, Paul Marshall, Raúl Santos‐Rodríguez Date: 2026-04-20 Paper ID: openalex:2604.18297

Summary

This paper investigates the relationship between epilepsy seizure onsets and circadian rhythms derived from wearable inter-beat interval (IBI) data. By applying band-limited filtering and Hilbert-based phase estimation to 176 days of continuous wearable data, the authors identify significant circadian phase concentration for seizure events. The findings suggest that modeling physiological phase offers a more interpretable and effective bridge between continuous sensing and sparse clinical events than traditional clock-time approaches. This approach provides a foundation for enhancing seizure forecasting pipelines through multi-scale phase representation.

Key Contributions

  • Introduces a methodology for aligning discrete clinical seizure events with continuous physiological rhythms using Hilbert-based phase estimation.
  • Demonstrates statistically significant circadian phase concentration of seizures in a single-patient wearable dataset.
  • Provides evidence that explicit physiological phase representations offer predictive structure beyond linear clock-time baselines.

Open Questions & Future Work

Archivist Review

The paper proposes using physiological phase (via Hilbert transform) to improve seizure forecasting, which is a domain-specific application rather than a foundational ML concept. I approved the open questions regarding multi-scale rhythm interaction and nonstationary analysis as they represent significant, reusable bottlenecks in physiological time-series forecasting.

Approved Open Questions

  • Multi-scale Rhythm Interaction Modelling: Current models often miss the multi-scale interactions that likely characterize the underlying physiological processes of epilepsy, limiting the accuracy of seizure forecasting systems.
  • Nonstationary Rhythm Analysis Methods: Improving phase estimation accuracy in the presence of nonstationarity is critical for moving beyond simple, stationary assumptions in longitudinal digital health monitoring.

Rejected Candidates

  • [concept] Physiological Phase Representation (physiological-phase-representation) - paper_local: While useful, this is a specific application of circular statistics and signal processing phase estimation rather than a novel, reusable ML architecture or mechanism.

Metadata & Links

url
https://arxiv.org/abs/2604.18297
paper_id
2604.18297
paper_source
openalex
domain
time-series
tags
time-series-forecastinghealthcare-analyticsphysiological-sensing
architectures
datasets
concept_slugs
dataset_slugs
skill
TimeSeriesSkill
created_at
2026-04-23T05:48:11Z