Ates Tunahan, Tamkac Nezih, Sukur Ibrahim Halil, Ok Fesih
Department of Urology, Defne State Hospital, Hatay, 31000, Turkey.
Department of Urology, Adana City Training and Research Hospital, Adana, Turkey.
Urolithiasis. 2025 May 15;53(1):92. doi: 10.1007/s00240-025-01761-w.
This review aimed to investigate the role of large language models (LLMs) in clinical decision support, patient counseling, and patient education in the management of urolithiasis. Eleven eligible studies were assessed following a comprehensive search of the Scopus and Web of Science databases. In the realm of clinical decision support, large language models (LLMs), particularly ChatGPT-4, have shown efficacy in areas such as the diagnosis of urolithiasis and initial treatment planning. An evaluation of the models' adherence to the European Association of Urology guidelines revealed that ChatGPT and Perplexity outperformed Bard. For patient counseling, Bing AI exhibited a robust capacity to deliver resource-based information. The studies have yielded conflicting results regarding ChatGPT versions' ability to empathize. In the context of patient education, models such as Claude-3 and ChatGPT-4 have demonstrated the capability to provide accurate and comprehensible answers to patient questions. However, it was noted that the quality of information is occasionally conveyed using complex language. LLMs have considerable potential as assistive tools in the management of urolithiasis; however, their limitations necessitate expert supervision, especially in complex cases. In the future, it is anticipated that these limitations will be mitigated through improved training and integration of these models into clinical practice. Consequently, LLMs should be employed as auxiliary tools rather than primary instruments in clinical decision-making.
本综述旨在探讨大语言模型（LLMs）在尿石症管理中的临床决策支持、患者咨询和患者教育方面的作用。在对Scopus和Web of Science数据库进行全面检索后，评估了11项符合条件的研究。在临床决策支持领域，大语言模型（LLMs），特别是ChatGPT-4，在尿石症诊断和初始治疗计划等方面已显示出有效性。对这些模型遵循欧洲泌尿外科学会指南情况的评估表明，ChatGPT和Perplexity表现优于Bard。在患者咨询方面，必应人工智能表现出强大的提供基于资源信息的能力。关于ChatGPT各版本共情能力的研究结果相互矛盾。在患者教育方面，Claude-3和ChatGPT-4等模型已证明有能力为患者问题提供准确且易懂的答案。然而，有人指出，信息质量偶尔会用复杂的语言来传达。大语言模型在尿石症管理中作为辅助工具具有相当大的潜力；然而，其局限性需要专家监督，尤其是在复杂病例中。未来，预计通过改进训练并将这些模型整合到临床实践中，这些局限性将得到缓解。因此，在临床决策中，大语言模型应作为辅助工具而非主要工具使用。
