Abstracts

Rationale: Expanding recognition of the importance of early diagnosis and treatment of status epilepticus has increased urgent use of EEG. The need for faster and 24-hour available EEG gave birth to rapid EEG devices, including Ceribell. Ceribell has filled gaps in our institution’s clinical practice, such as the paucity of EEG technicians in off-hours. This study evaluated how epileptiform abnormalities identified on Ceribell correlated with conventional EEG. Another objective was to identify and improve workflow issues.

Methods: This was a retrospective analysis of EEGs between March 2021 and May 2022. Patients were screened using the Ceribell reading website. Electronic health records were accessed to obtain demographic data and EEG reports. Those with Ceribell EEG followed by conventional EEG within 24 hours were selected for comparison. The prevalence of epileptiform abnormalities on both records as well as discordance rates between them were calculated. Discordant findings were reported as false positives and false negatives compared with conventional EEG, the gold standard. Patients with a discordant EEG had a thorough review of their charts and EEG tracings to identify causes not related to the device.

Results: A total of 189 patients was screened, some with more than one Ceribell EEG, summing up to 204 records. Those without interpretation, conventional EEG, or uninterpretable tracings were excluded, resulting in 165 records. Demographics, rate of epileptiform abnormalities, and discordance are shown in Table 1. After the second review of the discordant records, non-device variables were at times identified as the cause, including markedly different length of EEG, change in antiseizure treatment, disagreement in interpretation between readers, low duration or frequency of abnormalities, lack of video, and difference in EEG quality. These records were not considered false positives or negatives (Table 2). Twenty-five records had abnormalities present only on Ceribell EEG. Once patients with initiation of antiseizure treatment between records were excluded, 6 were deemed false positives (3.6%), with 1 Ceribell EEG inaccurately detecting a seizure (0.6%). Sixteen records had abnormalities present only on conventional EEG. Once non-device variables were excluded, 6 Ceribell records were deemed false negatives (3.6%), with 3 of them missing seizures (1.8%).

Conclusions: Ceribell EEG was accurate in detecting epileptiform abnormalities, with a discordance rate of 7.2% when compared with conventional EEG. If only considering the detection of electrographic seizures, this rate was 2.4%. These findings support that using Ceribell EEG is acceptable when conventional EEG is not available. Intrinsic limitations of the device can decrease its accuracy further. These include lack of video and quality being dependent on provider placement. This is improved by routinely checking the record and adjusting the device. Bedside evaluation of the EEG while watching the patient can partially compensate for lack of video. As with any EEG, clear communication between the treating team and reader is vital for patient care.

Funding: This study had no financial support.