November 17, 2026 · Resuscitation · DOI: 10.1016/j.resuscitation.2026.111398

Commercially available multimodal LLMs for EEG-based prognostication after cardiac arrest.

Listen to this summary

This exploratory study asked whether two commercially available multimodal LLMs (ChatGPT and Claude) could predict neurological outcome after cardiac arrest directly from images of a single 10-second EEG segment, without task-specific signal processing. Across five prompting strategies in 33 comatose patients, AUCs ranged from 0.74 to 1.00, with the best accuracy (93.9%, achieved by both models using few-shot prompting) approaching but not matching a published dedicated deep-learning model (100%). The authors conclude that the approach shows variable but promising discrimination, that prompt design strongly affects performance, and that results need confirmation in a larger cohort.

Katy Machado Monteiro, Andria Pelentritou, Lucas Gruaz, Manuela Iten, Matthias Haenggi, Andrea O Rossetti, Gert Krummrey, Hans Pihan, Marzia De Lucia, Frederic Zubler

This is one of 33,000+ journals available on OSLR. Try it free for 14 days.

Free 14-day trial. 33,000+ journals. Cancel anytime.

14-day free trial. No commitment.

“

"Oslr has become part of my weekly routine on my day off. The clinical relevance of the summaries is outstanding — I'd rate it 9/10. Being able to consume research hands-free is a huge advantage for busy physicians."

Dr. Jennifer Thompson

Dr. Jennifer Thompson

Portland, OR

Stay current without falling behind

33,000+ journals. 3-minute audio summaries. Free for 14 days.

Download on the App StoreGet it on Google Play