Multimodal AI for automated resuscitation debriefing: Feasibility of timeline reconstruction using commercially available tools

Abstract

Resuscitation debriefing requires accurate timeline reconstruction for effective team learning. Standard documentation captures medications and interventions but lacks granular communication patterns, team dynamics, and teachable moments essential for educational debriefing. Video review addresses this gap but requires extensive manual annotation, limiting scalability. We developed a structured prompting methodology using commercially available multimodal AI (Google Gemini). This reproducible approach enables educators to generate automated debriefing timelines by uploading resuscitation recordings and applying a structured prompt framework, without requiring custom software development or programming expertise. Feasibility was validated using three simulation videos (8–11 min each): two ACLS cardiac arrests and one postpartum hemorrhage. Videos were processed using a 14-question structured prompt framework. Across scenarios, the approach detected most clinician-identified events with temporal accuracy of ±10–15 seconds. Automated timeline generation using existing AI tools is feasible across diverse acute care contexts. Uploading complete videos after the event significantly outperformed real-time analysis. The prompt framework is adaptable to other multimodal AI platforms, democratizing access to high-quality debriefing. Next steps include clinical validation using real resuscitation events, integration with electronic medical records for added context, and assessment of impact on team learning outcomes. The complete prompt framework and implementation guide are available on GitHub.

Ähnliche Arbeiten