Most medical devices are silent data collectors. Penn researchers want yours to talk back, and a 13-patient study suggests the basic idea works.
Penn Medicine announced a platform, published in Nature Medicine, that lets people message their EEG monitor through a smartphone app. A ChatGPT-style conversational agent sits on top of a HIPAA-compliant cloud pipeline that analyzes brain data in near real time.
What the study tested
Thirteen adults in Penn’s Epilepsy Monitoring Unit used the app. They could ask about their brain activity, request charts, report symptoms, and answer surveys three times a day. Researchers analyzed 1,307 messages, mostly about seizures, sleep, mood, and energy.
The more interesting part is the feedback loop. Patients flagged which detected seizures were real and which were false alarms, and the system improved. Clinicians can do the same through an “expert in the loop” algorithm that limits their input to high-value labels.
Why it matters
Consumer wearables from Apple, Oura, and Dexcom already push alerts, but they rarely learn from the individual in a conversational way. This is closer to a personalized clinical model that adapts per patient. That could cut down on false alarms, which is a real problem for seizure detection.
Still, caveats are big. The sample is tiny, patients were in a hospital unit, and no one has shown fewer seizures or better outcomes yet. Putting a language model near clinical devices also raises safety and regulatory questions the study doesn’t answer.
What’s next
The team plans to tune deep brain stimulation for Parkinson’s, add wearables, and test outpatient use. Litt also points to insulin pumps and cardiac devices. The work is backed by a $5.6 million NIH Pioneer Award.
Is this a product? No. But if the outpatient trials show real clinical benefit, expect device makers like Medtronic to pay attention.
