Subscribe to eTOC

Longest Seizure Study Yet Finds Daily, Weekly, Monthly, and Yearly Cycles

Article In Brief

New research suggests cycles of focal epileptic seizures repeat not only daily or weekly but also monthly and yearly based on continuous intracranial EEG data.


Circannual cycle of seizures (A) over 9 years. Positive values indicate increased seizure counts over running mean and negative values indicate decreased seizure counts over the running mean. Gradient shows phases as months of the year. (B) Corresponding wavelet-periodogram with peak near 365 days underscoring the presence of a circannual cycle. (C) Corresponding resultant vector and circannual distribution showing that seizures occur year-round but have slight seasonal preference for spring. (D) Multidien cycle over 9 months (shaded area in panel A) apparent in daily interictal epileptiform activity (IEA; averaged over calendar days) along with wavelet-derived cycles from 4 to 45 days (blue trace), self-reported seizures (black dots), and electrographic seizures (red crosses). Higher linear density of black dots and crosses in the left half of the timeseries corresponds to increase in the seizure rate in panel A. (E) Corresponding periodogram of daily IEA counts in panel D, showing peaks at 7, 12, and 32 days. (F) Corresponding circular distribution and resultant vectors showing preferential multidien phase of occurrence for electrographic (black) and self-reported (red) seizures. (G) Circadian cycle apparent in hourly IEA over 20 days (shaded area in panel D) along with wavelet-derived circadian periodicity (blue curve) and electrographic seizures. Black boxes represent nights and white boxes represent days. (H) Corresponding periodogram of hourly IEA counts in panel G, showing peak at 24 hours.(I) Corresponding circadian distribution and resultant vector showing a preferred time for seizure occurrence around 18:00. The expected value of histogram bins for a uniform distribution depends on total number of bins: 12 monthly bins in panel C, 18 20° bins in panel F, and 24 hourly bins in panel I. au indicates arbitrary units; and PLV, phase-locking value.

The longest study to date using data from continuous intracranial EEG (cEEG) has found cycles of focal epileptic seizures that repeat not only daily or weekly but also monthly and yearly.

The data, collected between 2004 and 2018 with durations up to 10 years, involved 222 adults with medically-refractory focal epilepsy who were participating in a clinical trial of an implanted neurostimulation device, the NeuroPace RNS system.

Consistent with previous research, the study published online on February 8 on JAMA Neurology found high rates of circadian seizure cycles among participants: 89 percent. The surprise, due to the length of the study, was in the high proportion of patients—112 of 186, or 60 percent—who also had seizures that repeated approximately weekly to monthly.

Although the length of multiday cycles varied across individuals, they all tended to recur in roughly week-long increments, for reasons that remain mysterious: every seven,15, 20, or 30 days.

The study also found that 24 of the 194 patients, or 12 percent, had seizure cycles that repeated approximately once a year.

“Neurologists have not previously appreciated these longer-term cycles because we didn't have the ability to objectively track them for months and years,” said a senior author of the paper, Vikram R. Rao, MD, PhD, FAAN, associate professor of neurology at the University of California, San Francisco, and chief of its epilepsy division.

An epileptologist who has been studying seizure patterns for over a decade called the paper a “milestone.”

“This study has a huge population over a very long period of time, using the gold standard of seizure monitoring,” said Tobias Loddenkemper, MD, FAAN, director of clinical epilepsy research at Boston Children's Hospital and professor of neurology at Harvard Medical School, who was not involved with the study.

“If we dare to dream, some of these cycles really could help us rethink how we look at epilepsy treatment in general.”

Dr. Rao and other members of the study team are now planning a prospective clinical trial to establish whether people can benefit from seizure forecasting over long horizons.

Beyond the limited group of medically-refractory patients who require an implanted neurostimulation device, Dr. Loddenkemper and others are investigating less invasive approaches to monitoring seizures. Whether implanted under the scalp, adhered to the head with a removable sticker, or even worn on the arm or the wrist like a smartwatch, seizure-monitoring technologies are seeking wide clinical applications. But they are not yet ready for use outside of clinical trials, Dr. Rao said.

“It's tempting to immediately jump to clinical applications, but we need prospective trials to see if the technology really helps people manage their seizures better,” he said. “I would like to think that the knowledge of one's own patterns will ultimately be empowering for individuals. But we need to study it formally.”

Study Details

The retrospective cohort study sought to establish the prevalence, strength, and temporal patterns of seizure cycles over time periods stretching from hours to years. From a total of 256 participants in a previous clinical trial of the NeuroPace device, the group selected 222 patients based on availability of cEEG and/or self-reports of disabling seizures.

The study's finding of multi-day patterns was not the first. A 2018 paper in Lancet Neurology, for instance, analyzed data from 12 people who had been in a study of the implanted NeuroVista device, and 1,118 patients from the diary-based Seizure Tracker database. It found weekly and multi-week rhythms in just over one in five participants.

To search for fuzzier patterns outside of a strict seven-day schedule, the new paper used a technique called “circular statistics.”

“The problem is that not every cycle is precisely seven days long,” Dr. Rao said. “Sometimes it's 6.7 days, other times it's 7.5. Biology is not metronomic; there's a bit of jitter. That variation is why someone with an approximately seven-day cycle doesn't have a seizure every Sunday morning. It can drift by half a day. When we used the circular statistics to catch the wobble, we found that these patterns are much more prevalent than previously seen.”

Rather than the approximately one in five rate seen in the 2018 paper in Lancet Neurology, the new paper found that six in ten have multi-day cycles.

“Across individuals,” the paper reported, “circadian seizure cycles showed five peaks: morning, mid-afternoon, evening, early night, and late night.” Multi-day (also known as “multidien”) cycles of epileptiform activity (IEA), the study found, “showed peak periodicities centered around seven, 15, 20, and 30 days. Independent of multidien period length, self-reported and electrographic seizures consistently occurred during the days-long rising phase of multidien cycles of IEA.”

In other words, regardless of the length of the cycle, seizures were increasingly likely to occur as the cycle approached, rather than receded from, its multi-day peak.

The study's finding that 12 percent of patients had annual cycles of seizures “represents the first quantitative investigation of ‘seasonal epilepsy,’” the paper stated.

Practicing neurologists have commonly encountered patients reporting seizures occurring at the same time of day, month, or even year, but such anecdotal observations have seldom been quantified. As a result, some patients have been misled by “spurious correlations with behavioral or environmental factors,” the paper stated. “Our results help demystify fluctuations in seizure rates and refocus the dialogue between clinicians and patients on objective metrics associated with long-term disease course.”

An immediate implication of the findings, the paper noted, is that diagnostic studies intended to capture seizures, such as inpatient video EEG, might have a higher yield if they are scheduled based on times of highest seizure likelihood.” Likewise, “given the variability of seizure rates across and within patients,” future clinical trials could benefit from studying patients when they are most likely to have seizures.

Another paper from the same group, published in the February issue of Lancet Neurology, attempted to exploit the finding of seizure cycles in adults with focal epilepsy to make patient-specific forecasts. Eighteen patients with cEEG and electrographic seizure data were used to develop the forecasting models, and another 157 patients with cEEG and self-reported seizure data were used to validate the models.

“Models incorporating information about multidien IEA cycles alone generated daily seizure forecasts for the next calendar day with IoC [improvement over chance] in 15 (83 percent) patients in the development cohort and 103 (66 percent) patients in the validation cohort,” the paper reported. “The forecasting horizon could be extended up to 3 days while maintaining IoC in two (11 percent) of 18 patients and 61 (39 percent) of 157 patients. Forecasts with a shorter horizon of one h, possible only for electrographic seizures in the development cohort, showed IoC in all 18 (100 percent) patients.”

For both papers, the co-senior author with Dr. Rao was Maxime O. Baud, MD, PhD, epileptologist at the Sleep-Wake-Epilepsy Center in the department of neurology at the University of Bern in Switzerland.

Expert Commentary

“Seeing this new stream of data confirming cycles of seizures is highly satisfying,” said Mark Cook, MD, professor of neurology at the University of Melbourne in Australia.

Dr. Cook was the senior author of the 2018 paper that had found seizure cycles in data from NeuroVista and Seizure Tracker and was chief investigator of the NeuroVista trial, which demonstrated that prospective seizure prediction is possible. He is now conducting a clinical trial using a seizure monitor that is inserted under the scalp without penetrating the skull.

“We hope to get up to 50 people, but have been hamstrung by COVID-19,” he said.

The value of finding predictable cycles when seizures are more or less likely to occur, he said, is to reduce the unpredictability of the disorder.

“It can be socially disabling for some people,” Dr. Cook said. “The unpredictability affects their employment and driving ability, and creates a significant physical risk. If we can solve that problem, it would give them back their independence.”

Gregory L. Barkley, MD, a neurophysiologist at Henry Ford Hospital who specializes in the treatment of epilepsy, said he is already looking at the seizure patterns of his patients who have the NeuroPace device to help them forecast their times of low and high risk.

“I am personally training about two dozen patients with the NeuroPace device,” he said. “The limitation of this study, however, is that most people don't have the continuous intracranial monitoring you need to do this mathematical modeling. There are also people with multi-focal epilepsy, who do not have the NeuroPace device and thus were not in this study. We don't know what these findings will mean for them.”

The annual patterns seen in the study did not come as a great surprise, but may be due to other triggers, Dr. Barkley said.

“I have a group of patients without the NeuroPace device who have clusters of seizures every year around Christmas,” he said. “For most of them, it has to do with the excitement of the holiday, staying up late, and perhaps drinking alcohol. When you take that trifecta of factors, you get more seizures.”

Selim R. Benbadis, MD, FAAN, professor of neurology at the University of South Florida and director of the Tampa General Hospital's Comprehensive Epilepsy Program, said that he and other neurologists treating epilepsy have routinely heard patients say that they have seizure cycles. But, he said, “Until now, we were skeptical. Some of us rolled our eyes and didn't believe it. Now we are more likely to believe and perhaps to adjust diagnostic and therapeutic interventions. Maybe now we can say that during the second half of April, or the third week of June, is when a patient needs stronger treatment.”

In December, he was senior author of a paper in Epilepsia that used machine learning from a wristband sensor to provide noninvasive seizure forecasting. The study of 69 patients ran over the course of 2,311 hours, during which 452 seizures were observed. The deep-learning approach was able to accurately predict seizures in 43 percent of patients and did not differ between generalized and focal seizure types.


Dr. Rao has received grants and consulting fees from NeuroPace, Inc. during the conduct of the study. Dr. Cook had no disclosures. Dr. Barkley disclosed that he was an investigator in the NeuroPace RNS Feasibility Trial, the NeuroPace RNS Pivotal Trial, and the NeuroPace RNS Long-Term Treatment Trial, and his center at Henry Ford Hospital had the largest number of patients in the feasibility trial. These trials ran from 2004-2018 so some of the data in the article may be from patients he treated in the trials. The data were de-identified, and he has no relationship with the authors, the company, and he had no part in the manuscript. Dr. Benbadis disclosed the following (none of which are relevant to this article): He has been a consultant or served on advisory boards for Bioserenity, Ceribell, Eisai, Greenwich, LivaNova, Neurelis, Neuropace, SK Life Science, and Sunovion. He has participated in the speakers bureau for Alliance, Aquestive, Bioserenity, Eisai, Greenwich, LivaNova, Neurelis, SK Life Science, and Sunovio. He is the national medical director for RSC Diagnostic Services (EEG), the Florida medical director for Stratus/Alliance (EEG), Nexus Neuro (EEG), and is a member of the Epilepsy Study Consortium. Dr. Benbadis has received grant support from Cerevell, Neuropace, Greenwich, Otsuka, SK Life Science, Takeda, and Xenon; he has also received royalties as an author or editor for Emedicine-Medscape-WebMD, and UpToDate. Dr. Loddenkemper has received past funding by Empatica for a wristworn sensor, is listed on two NIIH grants, together with Epitel, and is an inventor through Boston Children's Hospital on several patents related to seizure detection and prediction, but none of the patents have been licensed to date.

Link Up for More Information

• Leguia MG, Andrzejak RG, Rummel C, et al. Seizure cycles in focal epilepsy JAMA Neurol 2021; Epub 2021 Feb 8.
    • Proix T, Truccolo W, Leguia MG, et al. Forecasting seizure risk in adults with focal epilepsy: A development and validation study Lancet Neurol 2021;20(2):127–135.
    • Karoly PJ, Goldenholz DM, Freestone DR, et al. Circadian and circaseptan rhythms in human epilepsy: A retrospective cohort study Lancet Neurol 2018;17(11):977–985.
    • Duun-Henriksen J, Baud M, Richardson MP, et al. A new era in electroencephalographic monitoring? Subscalp devices for ultra-long-term recordings Epilepsia 2020;61(9):1805–1817.
    • Meisel C, El Atrache R, Jackson M, et al. Machine learning from wristband sensor data for wearable, noninvasive seizure forecasting Epilepsia 2020;61(12):2653–2666.