Abstracts

Rationale: Circadian disruption has been reported in many conditions including epilepsy and is often associated with poor health outcomes. Here, we present a novel computational method to measure variability in circadian properties over time using continuous heart rate recordings, as a biomarker for day-to-day circadian variability and disruption. We apply this to long-term data recorded from 164 people with epilepsy (PWE) and 36 controls using a wearable smartwatch in free-living conditions.

Methods: Using singular spectrum analysis (SSA), we extract a circadian rhythm from the 5-minute-averaged heart rate time-series of each participant. We then derive estimates of 3 circadian properties: period (duration), acrophase (peak time), and amplitude (strength), for each circadian cycle (approximately each day). Circadian variability was estimated as the standard deviation of each property across all circadian cycles. We compared circadian variability between PWE and controls using the two-sided Mann-Whitney U test.

Results:
Accounting for the effect of recording duration, we found significant increases in the standard deviation of period (p=0.004) and acrophase (p=0.003) for PWE, but not amplitude (p=0.192).




Conclusions:
Our result provides further evidence for circadian disruption in epilepsy, described specifically as increased variation in circadian period and acrophase compared to controls in free living conditions. The mechanisms driving increased circadian variability remain uncertain – future work will investigate how this relates to self-reported seizure measures in this cohort. Our findings demonstrate the potential of this method, which is compatible with affordable, non-invasive wearable devices, to facilitate the study of circadian disruption across various conditions at scale.




Funding: Y.W. is supported by UKRI Future Leaders Fellowships (MR/T04294X/1, MR/V026569/1).