Abstracts

Rationale: Long-term electroencephalography (EEG) monitoring (LTM) of patients in the neurointensive care unit (ICU) is a common and valuable technique for diagnosis and evaluation of seizures. However, LTM requires manual EEG interpretation by a physician, often leading to clinically significant delays in response to critical conditions such as status epilepticus. Faulty or disconnected EEG leads may go uncorrected for up to twelve hours before review, resulting in long stretches of uninterpretable data. This system presents a significant burden for caretakers and results in increased lengths of stay and suboptimal patient outcomes. Furthermore, current LTM systems provide limited trending, meta-data, or in-depth real-time analysis.Methods: In this study, we present a pipeline for real-time acquisition and analysis of ICU EEG data with automated caretaker notification of key clinical events. Natus Nicolet EEG acquisition units are used for data collection from patients in the neurointensive care unit. Monitors currently in use are automatically identified, and data are streamed directly from patient EEG monitors to a central server for format conversion and analysis. Detection of clinical events triggers text-based notification of the patient’s clinical team using our custom API to route relevant information through a password-protected smartphone application (Figure 1, patient name redacted). Interviews were conducted with members of the clinical team to optimize the notification protocol. All analysis and notification takes place within the Hospital of the University of Pennsylvania's firewall in accordance with HIPAA patient confidentiality protocols. We validate this platform through caretaker notification of faulty or disconnected EEG leads. Leads with poor data quality are identified by detection of consistent outliers in median EEG line length.Results: Data were successfully streamed from patient monitors for live analysis. The potential utility of this system was demonstrated through detection of faulty EEG leads at over 85% accuracy (n=115) compared to human markings, resulting in reliable reporting of poor signal quality within five minutes of lead malfunction.Conclusions: This platform for automated real-time collection of EEG data provides innumerable opportunities for live data analysis with direct clinical impact. This system is currently being used to validate novel algorithms for automated seizure detection, monitoring patients in burst suppression, and other use cases. These extensions of the platform promise immediate positive impact on patient care. This platform also offers the potential for rapid acquisition and storage of patient EEG for research purposes. In summary, we have successfully generated a platform for live streaming and analysis of ICU EEG and rapidly alerting the clinical team. This research was funded by the Penn Medicine Center for Health Care Innovation.