Artificial intelligence in acute and critical care: current challenges and strategic solutions

The Emergency Department (ED) and Intensive Care Unit (ICU) are high-acuity environments where rapid decision-making and clinical precision are fundamental to patient survival. Artificial Intelligence (AI) offers transformative potential by providing real-time data synthesis, advanced pattern recognition, and personalized decision support-capabilities essential for optimizing clinical efficiency and patient outcomes. This narrative review synthesized the applications, clinical benefits, and implementation challenges of AI within acute and critical care.

A comprehensive literature search was conducted across international and Chinese databases, including PubMed, Web of Science, China National Knowledge Infrastructure (CNKI), and Wanfang Data, yielding 281 English and 136 Chinese studies. Furthermore, official policy documents from provincial Health Commissions were analyzed to evaluate the regulatory landscape for AI deployment.

AI showed potential across multiple acute and critical care scenarios, including early warning of sepsis, chest pain assessment, stroke imaging, electrocardiogram interpretation for acute coronary syndrome, ARDS subphenotyping, cardiogenic shock risk stratification, treatment support, and workflow coordination. However, translation into real-world practice remained limited by major challenges, including poor data quality, heterogeneity and fragmentation of clinical data, limited model explainability, insufficient prospective validation, workflow integration barriers, clinician training gaps, and unresolved ethical and regulatory issues.

Medical AI held substantial promise for improving decision-making efficiency, workflow optimization, and patient outcomes in acute and critical care. However, its safe and effective implementation required more than predictive performance alone. Future progress depended on trustworthy, explainable, workflow-integrated, and prospectively validated systems supported by stronger data infrastructure, interdisciplinary collaboration, and clearer ethical and regulatory oversight. This review proposed a translational framework and a general workflow for data-driven AI development in acute and critical care.