Researchers at the University of Pennsylvania have developed an artificial intelligence platform allowing epilepsy patients to communicate directly with their medical devices via smartphone. Tested in a hospital monitoring unit, the system successfully answered questions about brain activity and helped reduce automated seizure false alarms by 80 percent.
A clinical pilot study conducted at the Hospital of the University of Pennsylvania has introduced a conversational interface designed to bridge the gap between continuous neurological monitoring and patient awareness. Researchers enrolled 13 adults undergoing continuous electroencephalogram monitoring in the hospital’s epilepsy monitoring unit. Through a smartphone application connected to hospital systems, participants could ask questions about their brain activity, request visualizations of data trends, report symptoms, and receive personalized alerts or surveys.
University of Pennsylvania Trial Connects Patients to Live EEG Data
Current medical devices collect extensive physiological data that is typically reviewed periodically by clinicians rather than shared with patients in real time. To overcome this limitation, the research team built a platform that streams scalp and intracranial EEG data from clinical recording stations into a cloud analytics environment. A version of OpenAI’s GPT-4o-mini handled the conversational interactions through a private endpoint contractually barred from retaining or training on patient data.
During the hospital stay, participants and the system exchanged 1,307 messages covering topics such as seizure events, sleep patterns, mood, and energy levels. Simple chat replies took about two seconds, while data queries requiring database searches took approximately nine seconds. Senior author Brian Litt, a professor of neurology, neurosurgery, and bioengineering at the university, explained the rationale behind the interface.
These devices are made first and foremost for doctors, however, they do see a lot of information that patients would benefit from knowing. There are insights buried there about how their behavior impacts health, but they currently have no way of learning them in the moment.
Brian Litt, M.D., professor of neurology, neurosurgery, and bioengineering at the University of Pennsylvania
Patient Feedback Cuts False Alarms by Eighty Percent
Automated algorithms aboard monitoring devices frequently flag suspicious electrical signals that turn out to be benign activity, requiring clinical staff to manually disregard false detections. In the new pilot study, the application sent 58 real-time seizure alerts to patients. Participants responded to about two-thirds of these detections, marking them as true seizures, false alarms, or uncertain.
Researchers utilized this patient-generated feedback to retrain the underlying detection model. In a retrospective simulation, incorporating patient responses reduced the system’s false alarm rate by 80 percent across the study period.
Plans for Outpatient Expansion and Parkinson’s Disease Applications
Several participants expressed interest in continuing to use the system after leaving the hospital, with co-first author Zack Goldblum noting that several patients asked if they could take the system home with them. Because clinical deployment remains premature, researchers instructed participants that their care team was not actively monitoring the app in real time and directed them to press standard unit alarms for suspected seizures.
The study team is currently preparing for large-scale outpatient evaluations. Simultaneously, investigators are working to adapt the conversational platform to facilitate communication between deep brain stimulation devices and patients managing Parkinson’s disease, alongside broader exploratory applications for cardiac devices and insulin pumps.
