Highlight
A novel AI-enabled point-of-care (POC) EEG system quantifies seizure burden in critically ill adults.
The AI-derived seizure burden correlates dose-dependently with poor functional outcomes at hospital discharge.
Combining a bedside and more sensitive portal algorithm strengthens the association with clinical prognosis.
This study offers initial evidence supporting AI-based EEG interpretation to guide acute neurological care and prognostication.
Study Background
Seizure monitoring in critically ill patients is essential as seizures and status epilepticus can significantly worsen neurological outcomes. Electroencephalography (EEG) is the gold standard for detecting seizures, but conventional EEG access and interpretation are often limited by specialized equipment and trained neurophysiologists. This gap leads to delays or under-recognition of seizures, hampering timely intervention. Recently, portable point-of-care (POC) EEG devices coupled with artificial intelligence (AI)-based algorithms have emerged to facilitate rapid bedside seizure detection. However, the clinical relevance of AI-detected EEG seizure patterns and their relationship to meaningful outcomes requires further evaluation.
Study Design
This work represents a secondary cohort analysis from the retrospective, multicenter Seizure Assessment and Forecasting with Efficient Rapid-EEG (SAFER-EEG) study. EEG and clinical outcomes data were collected from three academic medical centers that integrated Ceribell’s POC EEG system into routine neurological care. The study enrolled 400 adult patients undergoing EEG monitoring; complete outcome and clinical data were available for 359 individuals. Seizure burden (SzB) was quantified by two AI algorithms: a “bedside” algorithm designed for real-time clinical use and a more sensitive “portal” algorithm used offline. The primary outcome was functional status at discharge measured by the modified Rankin Scale (mRS), with unfavorable outcome defined as mRS greater than 3. Associations between seizure burden metrics and outcomes were adjusted for relevant clinical confounders.
Key Findings
The bedside AI algorithm identified a peak 5-minute seizure burden greater than zero percent in 39.8% of patients. More prolonged seizure activity measured by the bedside algorithm was strongly associated with poor functional outcomes at discharge. Specifically, for each additional hour of seizure detected, there was a near doubling of odds for unfavorable outcome (adjusted odds ratio [aOR], 1.98; 95% confidence interval [CI], 1.11-4.29). Patients exhibiting a peak 5-minute seizure burden ≥90% had a 3.4-fold increased odds of poor outcome compared to those with no detected seizure burden.
Moreover, incremental increases in seizure activity as brief as 30 seconds per hour in the maximum hourly SzB were independently associated with worsening outcomes (aOR, 1.02; 95% CI, 1.00-1.03). Critically, combining the outputs of the bedside and portal AI algorithms improved the predictive strength for outcome, especially in patients with high seizure burdens (peak 5-minute SzB ≥90%), resulting in an aOR of 4.4 (95% CI, 1.66-12.69) for poor outcome.
Expert Commentary
This study provides pioneering clinical validation of AI-based seizure burden quantification from POC EEG and its prognostic relevance. Importantly, the dose-response relationship highlighted here suggests not only that seizure presence but also the temporal seizure load impacts functional recovery. The use of both real-time bedside and offline sensitive algorithms offers a balanced approach between feasibility and sensitivity, enhancing clinical decision-making.
However, limitations include the retrospective design and potential center-specific management variations. While mRS is a standard neurological outcome measure, longer-term functional and cognitive outcomes remain to be studied. Further prospective trials are needed to determine whether AI-guided seizure detection and interventions can improve patient prognosis.
Nonetheless, these findings support integrating AI-enabled POC EEG into critical care workflows to improve timely seizure identification and guide tailored treatment strategies.
Conclusion
The AI-derived seizure burden from POC EEG systems is significantly associated with functional outcomes at hospital discharge in critically ill adult patients. The relationship is dose-dependent and remains robust after adjustment for confounding factors. This study represents an important validation step for AI-enhanced EEG interpretation, emphasizing its promise for augmenting neurological monitoring and guiding management to potentially improve outcomes in seizure-prone populations.
Funding and Clinical Trials Registration
The SAFER-EEG study was supported by institutional and federal grants; details are available in the original publication. This analysis was retrospective and secondary; no new clinical trial registration applies.
References
Parvizi J, Armenta Salas M, Aparicio MK, et al. Point-of-Care EEG Artificial Intelligence Measure of Seizure Burden Associates With Clinical Outcome at Discharge. Crit Care Med. 2026 Jun;54(7):1710-1720. PMID: 42223304.
Claassen J, Hirsch LJ, Kreiter KT, et al. Prognostic significance of continuous EEG monitoring in patients with seizures after traumatic brain injury. Neurology. 2004;62(10):1568-1574.
Hirsch LJ, Brenner RP. Continuous EEG monitoring in the intensive care unit: an overview of what clinicians need to know. Am J Electroneurodiagnostic Technol. 2006;46(1):8-19.