Fraile Navarro David, Kocaballi A Baki, Berkovsky Shlomo
Centre for Health Informatics, Australian Institute of Health Innovation, Macquarie University, 75 Talavera Road, Sydney, 2113, NSW, Australia.
Faculty of Engineering and Information Technology, University of Technology Sydney, Sydney, Australia.
J Med Syst. 2025 Aug 2;49(1):101. doi: 10.1007/s10916-025-02234-8.
This mixed-methods study evaluated clinicians' user experience (UX) with Generative AI (GenAI) in Electronic Health Record (EHR) systems across three clinical documentation tasks (Information Extraction, Summarization, and Speech-to-Text) at varying levels of user supervision (low, medium, high), focusing on workflow improvements, safety, and acceptable automation levels. Using conceptual prototyping in a usability study framework, we evaluated how incorporating GenAI into EHR could support the three documentation tasks at varying automation levels. A total of 38 clinicians interacted with the prototype and completed a questionnaire on task relevance, perceived importance, desired automation level, and EHR satisfaction. Both quantitative (descriptive statistics, Kruskal-Wallis tests, Spearman correlations) and qualitative (thematic) analyses were conducted with equal priority to explore preferences, perceived safety, and practical requirements. Clinicians showed positive reception to GenAI integration, particularly for streamlining documentation. While task relevance and importance were strongly correlated, EHR satisfaction did not significantly predict automation acceptance. Medium automation emerged as the preferred level, considered "safe with caution". Five key themes emerged from qualitative analysis: efficiency and quality benefits; system reliability concerns; safety and medico-legal considerations; automation bias and loss of nuance; and deployment requirements including adjustable settings and oversight. While clinicians welcome GenAI-driven documentation, they prefer moderate automation to balance efficiency with clinical control. Successful integration requires addressing safety concerns, conducting real-world trials, and mitigating potential biases and medico-legal challenges. These findings suggest a cautious but optimistic path forward for AI integration in EHR systems, emphasizing the importance of maintaining clinician oversight while leveraging automation benefits.
这项混合方法研究评估了临床医生在电子健康记录（EHR）系统中使用生成式人工智能（GenAI）时在三种临床文档任务（信息提取、摘要生成和语音转文本）中的用户体验（UX），这些任务处于不同程度的用户监督水平（低、中、高），重点关注工作流程改进、安全性和可接受的自动化水平。在可用性研究框架中使用概念原型，我们评估了将GenAI纳入EHR如何在不同自动化水平下支持这三种文档任务。共有38名临床医生与原型进行交互，并完成了一份关于任务相关性、感知重要性、期望的自动化水平和EHR满意度的问卷。定量分析（描述性统计、Kruskal-Wallis检验、Spearman相关性分析）和定性分析（主题分析）同等重要地进行，以探索偏好、感知安全性和实际需求。临床医生对GenAI集成表现出积极的接受态度，特别是在简化文档方面。虽然任务相关性和重要性密切相关，但EHR满意度并未显著预测对自动化的接受程度。中等自动化水平成为首选，被认为“谨慎使用时安全”。定性分析中出现了五个关键主题：效率和质量提升；系统可靠性担忧；安全和医疗法律考虑；自动化偏差和细微差别丧失；以及包括可调整设置和监督在内的部署要求。虽然临床医生欢迎由GenAI驱动的文档，但他们更喜欢适度自动化，以平衡效率和临床控制。成功集成需要解决安全问题、进行实际试验，并减轻潜在偏差和医疗法律挑战。这些发现为EHR系统中人工智能的集成提出了一条谨慎但乐观的前进道路，强调了在利用自动化优势的同时保持临床医生监督的重要性。
