• Center on Health Equity & Access
  • Clinical
  • Health Care Cost
  • Health Care Delivery
  • Insurance
  • Policy
  • Technology
  • Value-Based Care

Automated Processing EEG Algorithm Predicts 1-Year Seizure Recurrence in Patients

News
Article

Predictions for identifying 1-year seizure recurrence performed significantly better in electroencephalography (EEG) without interictal epileptiform discharges.

An automated processing algorithm during routine electroencephalography (EEG) screening may help neurologists to better identify patients with epilepsy who are at high-risk for recurrence of seizures, according to one study.

“In this paper, we develop and validate predictive models for the prediction of seizure recurrence at 1-year based on the computational extraction of biomarkers from the routine EEG signal,” wrote the researchers of the study. “We train the model on a large retrospective cohort of consecutive patients undergoing routine EEG and validate the predictive accuracy on a temporally shifted cohort of patients. We investigate whether predictive accuracy is independent of IEDs [interictal epileptiform discharge] and other clinical confounders.”

This retrospective cohort study is published in Scientific Reports.

Patients with epilepsy often experience recurring seizures, in which predicting the risk of recurrence is crucial to the diagnosis and management of this patient group. Prior studies have shown that visual identification of interictal epileptiform discharges (IEDs) on during routine EEG for patients with epilepsy doubles the risk that the patient with have seizure recurrence in the next years.

However, sensitivity of EEG for predicting seizure recurrence is limited, with IEDs often subjected by overinterpretation, and the possible misdiagnosis of epilepsy.

In this study, the researchers aimed to identify a biomarker of seizure propensity that is automated, objective, and without IEDS, with the intent of reducing diagnostic error, accelerating treatment among high-risk patients, avoiding the consequences of overdiagnosis, and improving monitor disease activity.

Patients who had underwent a routine EEG between January 2018, and June 2019, from the University of Montreal Hospital Center were recruited into the study, retrospectively. EEGs recorded between January 2018, to December 2018, were part of the training set, while EEGs recorded between January 2019, and June 2019, made up the held-out testing set.

The researchers reviewed a total of 816 patient medical and clinical information, including age, sex, comorbidities at the time of the EEG, epileptogenic factors, reason for EEG, presence of focal brain lesion on neuroimaging, and number of antiseizure medication (ASM). From the EEG report, the researchers included the type of recording (awake or sleep deprived), deepest sleep stage achieved, presence of IED, and presence and degree of abnormal slowing.

The researchers trained machine learning algorithms on multichannel EEG segments, with the aim of predicting seizure recurrence 1-year after evaluation.

The median (SD) age of patients was 50 (IQR, 33-62) years. Additionally, the median follow-up after EEG was 100 (IQR, 42-135) weeks, 248 (72%) of EEGS did not show IEDs in people with epilepsy, and the EEG was part of the initial evaluation of suspected seizures in 286 cases.

A total of 517 patients undergoing routine EEG, with 549 EEG recordings were assigned to the training set. Among this cohort, 132 (24%) EEGs were from patients who had seizure recurrence after EEG screening.

Additionally, the testing set included 301 EEGs from 261 patients, with 32% prevalence of seizure recurrence after EEG.

Analysis showed the receiver operating characteristics area under the curve (AUC) for seizure recurrence after EEG in the testing set (AUC, 0.63; 95% CI, 0.55-0.71), with predictions significantly above chance in EEGs with no IEDS.

The researchers acknowledged some limitations to the study, including being based on a single center, the data collection being retrospective, and that for some patients, follow-up may have been too short to detect seizure recurrence.

Despite these limitations, the researchers believe the study suggests that the use of automated EEG algorithms may perform above change in EEGs with no IEDs, and that other changes in the EEG signal may be at play in seizure propensity.

“In conclusion, we demonstrate that there are changes other than IEDs in the EEG signal embodying seizure propensity,” wrote the researchers. “These changes have a predictive horizon of 1-year after the EEG and their significance is independent of IEDs, age, and number of antiseizure medications.”

Reference

Lemoine É, Toffa D, Pelletier-Mc Duff G, et al. Machine-learning for the prediction of one-year seizure recurrence based on routine electroencephalography. Sci Rep. 2023;13(1):12650. doi:10.1038/s41598-023-39799-8

Related Videos
Glenn Balasky, executive director of the Rocky Mountain Cancer Center.
Klaus Rabe, MD, PhD, chest physician and professor of medicine, University of Kiel
Klaus Rabe, MD, PhD, chest physician and professor of medicine, University of Kiel
April Armstrong, MD, MPH, chief of dermatology, UCLA
Toby Maher, MD, PhD, professor of clinical medicine, Keck Medicine of USC
Benjamin Scirica, MD, MPH, associate professor of medicine at Harvard Medical School and director of quality initiatives at Brigham and Women’s Hospital’s Cardiovascular Division
Glenn Balasky during a video interview
April Armstrong, MD, MPH, chief of dermatology, UCLA
dr joseph alvarnas
Toby Maher, MD, PhD, professor of clinical medicine, Keck Medicine of USC
Related Content
© 2024 MJH Life Sciences
AJMC®
All rights reserved.