Abstracts

Rationale: Recent studies have shown associations between seizures and epileptiform patterns on electroencephalography (EEG) and worse clinical outcomes as estimated by automated algorithms run in continuous EEG recordings (Parikh, et al, 2023). Point-of-care (POC)-EEG is a faster-access diagnostic tool that can run Clarity (Ceribell, Inc), a machine-learning algorithm that monitors seizure burden (SzB). We pose that both total time in seizure and maximum seizure burden within a 5-minute window, per Clarity, will be associated with worse functional outcomes at discharge.


Methods: We performed the analysis on the POC-EEG recordings from the SAFER-EEG retrospective study, which included adult patients admitted to the hospital who received POC-EEG. We ran the latest version of Clarity, which reports SzB as the percent of 10-second segments likely containing seizure patterns within a 5-minute rolling window to users at the bedside, and a more sensitive seizure-detection algorithm, which provides additional insight to ClarityPro users viewing the recording in the cloud portal, post-hoc in all recordings. We obtained the maximum 5-min SzB and estimated the cumulative time in seizure over the entire recording. Our primary outcome was the modified Rankin Scale (mRS) at discharge, defined as “unfavorable” for scores ≥4 (moderate-severe or worse disability). Multivariate logistic regression models were used to estimate adjusted odds ratio (OR), controlling for age, clinical presentation, recording duration and door-to-EEG time.

Results: A total of 359 patients were included in the analysis; a majority (59.1%) had 0% SzB throughout their recordings per Clarity, 12.8% had SzB ≥ 50%, and 7% had SzB ≥ 90%. After adjusting for clinical covariates, patients with SzB > 50% had 3.5-fold increase in odds of poor outcome, compared to those without seizure activity (adj. OR = 3.52 [1.48 - 8.87]; p = 0.002); and the subset with SzB ≥ 90% had a higher increase (adj. OR = 4.91 [1.56 - 17.9], p = 0.009). Combining both the seizure burden algorithm and the more sensitive detection algorithm, 28.7% of patients didn’t have any positive findings (SzB = 0%), 18.9% had SzB ≥ 50%, and 13.6% had SzB ≥ 90%. For this combined output, patients with SzB ≥ 50% also had significantly higher odds of mRS ≥ 4 (adj. OR = 4.2 [1.6-11.8]; p = 0.005). The sub-population with SzB ≥ 90% had 6-fold increase in odds, compared to those without seizures (adj. OR = 6.10 [2.09-20.3]; p = 0.002).

For the cumulative time in seizure, we found a 45% relative increase in odds of poor outcome for every 30 min in seizure (adj. OR = 1.45 [1.07-2.16], p = 0.035). Combining the algorithms, we found a similar trend where increased time with a positive finding increase the odds of poor mRS at discharge (adj. OR = 1.21 [1.03-1.48]; p = 0.037).



Conclusions: We showed that seizure burden, as calculated by the automated algorithms on POC-EEG, is associated with worse functional outcomes at discharge, even after controlling for initial clinical presentation, recording duration and time to EEG. This was true when using total time in seizure or maximum 5-minute seizure burden.

Funding: Funded by Ceribell, Inc. as an investigator-initiated study.