Large language model-driven agents in nursing practice: A scoping review

This review aimed to systematically analyze the technological frameworks, application scenarios, and outcomes of large language model-driven agents (LLMDAs) in nursing practice, and to summarize ethical, technological, and practical challenges, guiding future research and clinical implementation.

This scoping review was conducted following the JBI guidelines. Five databases (PubMed, Embase, Web of Science, APA PsycNet, Cochrane Library) were systematically searched for peer-reviewed English-language studies from inception until September 9, 2025. Eligible studies were screened by title and abstract, with full-text assessments conducted independently by two reviewers.

Twenty-five studies published between 2023 and 2025 were included, involving nine countries, primarily China (n = 9) and the United States (n = 9). Technological architectures were categorized into three types: collaborative models for solving complex tasks through multi-agent division of labor; augmentative models to enhance the accuracy of information outputs; and interactive models focusing on natural interactions and robotic task execution. Application scenarios included clinical, home-based, and community care. Studies indicated that LLMDAs can enhance diagnostic accuracy, optimize resource allocation, and improve patient experience. Primary ethical challenges identified included data privacy, reliability of generated content, and ambiguous attribution of responsibility.

LLMDAs offer a novel paradigm for intelligent transformation in nursing care through integrative technological frameworks. They demonstrate considerable potential in enhancing clinical decision-making accuracy, efficiency of care delivery, and patient satisfaction. Addressing existing ethical, technical, and practical challenges is essential for facilitating broader clinical adoption.