Clarity Interpretation of Ceribell Rapid Response Electroencephalography Use Shows Limited Influence on Patient Disposition and Outcome
Michael Connerney1, Victor Lami2, Melissa Reider-Demer3, Marc Nuwer2, Inna Keselman2
1UCLA Department of Neurology, 2UCLA, 3UCLA Medical Group
Objective:
To report how Clarity artificial intelligence (AI) interpretation of Ceribell rapid electroencephalography influences overall patient management and disposition in the acute emergency department (ED) setting.
Background:
Ceribell rapid response electroencephalography (EEG) was developed to rapidly detect epileptiform abnormalities (EA) in patients with a high suspicion of status epilepticus (SE). Conventional EEG, although the gold standard for diagnosis of seizures and EA, requires a dedicated technician to setup and can take several hours before data is available. Ceribell with clinical neurophysiologist (CN) interpretation has been shown to increase the sensitivity and specificity of seizure diagnosis in the acute setting. Clarity is an artificial intelligent (AI) software associated with Ceribell designed to recognize epileptiform abnormalities and inform treating physicians of discharge burden, developed so that healthcare teams may act on suspected SE more acutely than awaiting the read by a CN. However, less is known regarding the effect of Clarity on management of patients in suspected SE in the ED.
Design/Methods:
Observational retrospective comparison between patients with and without Clarity reported EA at UCLA ED.
Results:
Out of 82 patients who had Ceribell use, 34% had ≥ 10% seizure burden detected by Clarity, and 12% had EA on Ceribell as read by a CN. There was a 50% match rate between Clarity and Ceribell read by CN. Our data suggests that presence of increased seizure burden as interpreted by Clarity AI did not correlate with changes in management or patient disposition by UCLA ED physicians. In contrast, Ceribell detected EA as interpreted by a CN lead to further escalation of seizure management, often with continuous EEG and anti-seizure medications.
Conclusions:
EEG data acquired by Ceribell agreed with subsequent conventional EEG as interpreted by a clinical neurophysiologist and influenced patient management by ED. The same was not seen when Clarity AI alone was used for interpretation.