Authors :
Presenting Author: Boyu Zhang, M.S. – Massachusetts Institute of Technology
Weixuan Vincent Chen, Ph.D. – Empatica; Giulia Regalia, Ph.D. – Empatica; Daniel Goldenholz, M.D., Ph.D. – Department Of Neurology – Beth Israel Deaconess Medical Center; Rosalind Picard, Sc.D. – Massachusetts Institute of Technology, Empatica
Rationale:
Epilepsy patients are advised to maintain an adequate amount of sleep every day to reduce seizure risk. Despite the long-lasting notion that sleep deficiency may induce seizures, existing literature has not clarified the relationship between them with objective evidence. We performed a large-scale longitudinal observational ambulatory study to investigate sleep deficiency and the occurrence of seizures using wearable monitoring.
Methods:
Patients diagnosed with epilepsy characterized by generalized tonic-clonic seizures (GTCS) wore a smart wristband at home for at least three months. The total sleep time (TST) of each sleep episode was computed by an FDA-cleared activity-based rest detection algorithm. Sensor data from the wristband was also used to detect GTCS events using an FDA-cleared algorithm.
The adequate amount of sleep humans should get each day varies by age. We followed the age groups and TST bounds suggested by the National Sleep Foundation, the American Academy of Sleep Medicine, and the Sleep Research Society (see Table 1). Subsequently, we labeled each sleep episode as undersleep (above the suggested upper TST bound), oversleep (below the suggested lower TST bound), or appropriate (within the suggested TST bounds). Data from a patient were analyzed for next-day GTCS only if they had at least 30 sleep episodes in a given category.
We computed three corresponding risk values for each patient: the percentage of next-day containing at least one GTCS following undersleep (R1), oversleep (R2), and appropriate sleep (R3); the exact formulas are shown in Figure 1. Two-tailed paired t-tests assessed if the population-level difference between every two risk values (empirically Gaussian) was significant with the Holm-Bonferroni method. The magnitude of effect (MoE) was computed as the average of paired differences among all patients.
Results:
After excluding naps, insufficiently labeled false alarms, or other low-quality data, 140,678 seizures (median=38, range 18 to 1,982) and 331,814 sleep episodes (mean=159, range 60 to 597) were recorded from 2,397 patients. Overall, significantly elevated risks were found after undersleep compared to after appropriate sleep among all age groups: less than seven hours for ages 25 to 60 (p< 0.001, MoE=2.68%), less than seven hours for ages 17 to 25 (p< 0.001, MoE=3.17%), less than eight hours for ages 13 to 17 (p< 0.001, MoE=3.14%), and less than nine hours for ages 6 to 13 (p< 0.001, MoE=2.68%). Oversleep is not followed by a change in risk in any age group. The complete results are given in Table 1.