Karino Mayuko, So Clara, Jinta Torahiko, Tanaka Michiho, Nakamura Tomoaki, Okafuji Kohei, Kitamura Atsushi, Tomishima Yutaka, Nishimura Naoki
Department of Pulmonary Medicine, Thoracic Center, St. Luke's International Hospital, Tokyo, JPN.
Department of Pulmonology, St. Luke's International Hospital, Tokyo, JPN.
Cureus. 2025 May 14;17(5):e84133. doi: 10.7759/cureus.84133. eCollection 2025 May.
Improvements in operations through the use of artificial intelligence (AI) are expected in various fields. Chat Generative Pre-trained Transformer (ChatGPT), released in November 2022, has gained rapid popularity and is widely used; however, from a privacy perspective, research on the use of AI in medicine is limited. In this study, we compared 68 discharge summaries generated by ChatGPT based on interim summaries (hereafter, "AI summaries") with those generated by junior residents (hereafter, "resident summaries") and investigated the challenges and benefits of using ChatGPT. Resident summaries significantly outperformed AI summaries in quality (p = 0.011), factuality (p < 0.002), and completeness (p < 0.001) but not in readability (p= 0.08). For hospitalizations exceeding three months (n = 12), no significant differences were observed between resident and AI summaries on any measure. These findings suggest that AI can produce discharge summaries comparable to those of residents in complex, long-term cases. Thus, responsible use of ChatGPT, ensuring data privacy, can aid in efficient discharge summary creation.
预计人工智能（AI）的应用将在各个领域改善手术效果。2022年11月发布的聊天生成预训练变换器（ChatGPT）迅速走红并得到广泛应用；然而，从隐私角度来看，AI在医学领域的应用研究有限。在本研究中，我们将ChatGPT根据临时总结生成的68份出院小结（以下简称“AI小结”）与低年资住院医师生成的出院小结（以下简称“住院医师小结”）进行了比较，并探讨了使用ChatGPT的挑战和益处。住院医师小结在质量（p = 0.011）、事实性（p < 0.002）和完整性（p < 0.001）方面显著优于AI小结，但在可读性方面无显著差异（p = 0.08）。对于住院时间超过三个月的患者（n = 12），住院医师小结和AI小结在任何指标上均未观察到显著差异。这些发现表明，在复杂的长期病例中，AI可以生成与住院医师相当的出院小结。因此，在确保数据隐私的前提下，负责任地使用ChatGPT有助于高效创建出院小结。
