The AI dispatcher copilot: beyond cardiac arrest recognition to dynamic large language model-assisted Tele-CPR

Out-of-hospital cardiac arrest (OHCA) remains a leading cause of death. Although emergency medical dispatchers represent the first link in the Chain of Survival, a persistent "AI translation gap" exists, whereby traditional machine learning models demonstrate high diagnostic performance but limited impact on clinical outcomes. This paper proposes a paradigm shift from passive AI-assisted OHCA recognition towards a dynamic, multimodal large language model (LLM)-enabled Tele-CPR system, conceptualised as an "AI dispatcher copilot".

This narrative synthesis integrates recent developments in large language models, computer vision, and cognitive load theory, aligned with the European Resuscitation Council (ERC) Guidelines 2025. It evaluates the conceptual feasibility of an integrated AI-driven decision-support architecture for emergency medical dispatch.

In contrast to narrow machine learning approaches, multimodal LLMs can integrate acoustic signals (including potential agonal breathing detection) with semantic interpretation of caller narratives to reduce ambiguity in OHCA recognition. Proposed functionalities include adaptive instruction generation tailored to caller stress to reduce cognitive load, real-time video-assisted CPR coaching with closed-loop feedback on compression quality, and automated resource orchestration through parallel activation of community first responders and automated external defibrillator (AED) routing.

The AI dispatcher copilot represents a potentially transformative evolution in Tele-CPR systems. However, translation into clinical practice requires rigorous ethical governance addressing algorithmic bias, automation bias, and data privacy, alongside prospective validation to demonstrate improvement in neurologically intact survival.