Abstracts

Rationale: Patient safety is paramount while recording seizures in the epilepsy monitoring unit (EMU). Prompt seizure detection is critical for alerting appropriate healthcare personnel (HCP) to attend to patients experiencing seizures. Despite best efforts, seizures may go undetected in the EMU. Past surveys have documented a variety of clinical and automated seizure detection methods in use in EMUs across the United States. In this study, we examine the reliability of various seizure detection methods utilized for patients admitted in a typical university hospital EMU.Methods: We reviewed all adult EMU admissions featuring epileptic or non-epileptic seizures during a six-month time period at our institution. All events were then further reviewed on video-EEG. We compiled and tabulated the following: event types, event onset and offset times, methods of detection (automated seizure detection alarm – Persyst Insight II ver.11 with default settings, patient push button, HCP detection), HCP response times after detection, and false positive detections.Results: A total of 151 (111 epileptic and 40 non-epileptic) events were reviewed. Patient push-button sensitivity for epileptic seizures was 100%, 14.5%, and 13.3% for auras, complex partial seizures, and convulsions, respectively. Positive predictive value (PPV) of push-button for either complex partial or generalized tonic-clonic seizure was 31%, while for non-epileptic seizure it was 40%. Automated alarm sensitivity was 20%, 87%, and 87%, respectively, for auras, complex partial seizures, and convulsions. PPV of automated alarm for any epileptic seizure was only 28%, with corresponding false positive rate of one false alarm every 14 hours. The average latency from seizure onset to patient push button was 34 seconds, while automated alarm latency after seizure onset averaged 97 seconds. The average latency from alarm sounding to staff arriving at bedside was 18 seconds for push-button alarms, vs. 106 seconds for automated alarms. Unattended seizures during this period numbered 12 out of 52 complex partial seizures (23%), and 2 out of 28 convulsions (7%).Conclusions: Seizures were detected in our institution by a variety of methods. Automated alarms have the best sensitivity (detecting 73% of all epileptic seizures) but sound late during the seizure, resulting in staff arriving typically after a seizure has ended. Push-button alarms sound earlier in the course of a seizure, but have low sensitivity (detecting only 25% of epileptic seizures). Relying solely on alarms to alert HCP to seizure activity will result in missed opportunities for ictal exams and unattended seizures, including convulsions. As maximizing safety in the EMU requires attending to all seizures, these findings reinforce the need for human personnel for continuous monitoring in the EMU. The metrics reported here may be useful for standardized assessment of EMU quality and safety.