Planning the development of an AI-driven decision support architecture for the recognition of sudden cardiac arrest by 9-1-1 telecommunicators: report of a community engagement and brainstorming meeting.

OBJECTIVES: Sudden cardiac arrest is a leading cause of mortality in Canada (40,000 treatable deaths annually). Despite the benefits of 9-1-1 telecommunicator-assisted CPR, agonal (reflexive/ineffective) breathing often leads to misdiagnosis and delayed intervention. Artificial intelligence (AI), particularly large language models, may improve cardiac arrest recognition during 9-1-1 calls. This study aimed to identify community-informed requirements for integrating AI models into 9-1-1 ambulance communications centers to enhance rapid cardiac arrest recognition and timely response. METHODS: We hosted a CIHR-funded planning and dissemination meeting at the Ottawa Paramedic Service. The qualitatively driven study involved diverse community members including emergency physicians, telecommunicators, paramedics, cardiac arrest survivors, AI model developers, and policymakers. We used a structured Miro board exercise before, during, and after the meeting to facilitate participant input. Five pre-identified themes guided the exercise: (1) essential must-haves in a useful AI application, (2) considerations for integration with 9-1-1 system, (3) policy/ethical considerations, (4) burning questions, and (5) missing community members. RESULTS: Participants emphasized the need for high cardiac arrest interpretation accuracy, low false activation rates, multilingual capabilities, and explainable AI model outputs. Integration concerns included compatibility with computer-assisted dispatch systems, platform independence, and redundancy during outages. Ethical considerations focused on privacy, data sovereignty, liability, and bias mitigation. Participants raised questions about AI model's ability to interpret voice cues, background audio, and caller distress. Gaps in community representation were identified, including accessibility experts and Ministry information technology specialists. CONCLUSION: This planning exercise identified critical technical, ethical, and operational requirements for AI model integration into 9-1-1 ambulance communications centers. Community member insights will inform future development and validation of an AI-driven decision support architecture aimed at improving cardiac arrest recognition, reducing time to CPR initiation, and enhancing equity in emergency response. The findings support a patient/telecommunicator-centered, responsible approach to AI model deployment in prehospital care.