生成式人工智能在炎症性肠病中的应用：一项系统综述。

Applications of generative artificial intelligence in inflammatory bowel disease: A systematic review.

作者信息

Lusetti Francesca, Maimaris Stiliano, La Rosa Gianmaria Pio, Scalvini Davide, Schiepatti Annalisa, Biagi Federico, De Bernardi Alice, Manes Gianpiero, Saibeni Simone

机构信息

Department of Internal Medicine and Therapeutics, University of Pavia, Pavia, Italy; IBD Unit, Gastroenterology Unit, Rho Hospital, ASST Rhodense, Rho MI, Italy.

Department of Internal Medicine and Therapeutics, University of Pavia, Pavia, Italy; Istituti Clinici Scientifici Maugeri IRCCS, Gastroenterology Unit of Pavia Institute, Italy.

出版信息

Dig Liver Dis. 2025 May 9. doi: 10.1016/j.dld.2025.04.026.

DOI:10.1016/j.dld.2025.04.026

PMID:40348628

Abstract

BACKGROUND AND AIMS

Inflammatory bowel diseases (IBD) are chronic conditions that can lead to a physical, social, and economic burden. Generative artificial intelligence (AI), particularly ChatGPT, gained attention for its potential to support medical practice. However, concerns remain about the reliability and consistency of its responses. This study systematically reviews the existing evidence on the role of generative AI in IBD.

MATERIALS AND METHODS

We conducted a systematic literature review following PRISMA guidelines. Studies investigating generative AI in IBD care were identified through PubMed and Embase (Jan 2020-Sep 2024).

RESULTS

From 2875 records, 8 studies (2023-2024) met inclusion criteria: 5 on patient education, 2 on decision support, and 1 on research ideation. For patient education, ChatGPT provided clear and accurate responses, with accuracy reaching 84.2 % in a study, though sometimes lacked consistency. In decision support, ChatGPT's classifications of ulcerative colitis severity aligned with clinician assessments in 80 % of cases and in 87.8 % of cases for guideline-based dysplasia management. For research ideation, ChatGPT generated highly relevant (mean score: 4.9 ± 0.26) and clear (4.8 ± 0.41) questions, but lacked specificity (2.86/5) and originality (1.07/5).

CONCLUSIONS

Generative AI shows promise in IBD care, but concerns about accuracy, consistency, and outdated information highlight the need for expert oversight before clinical integration.

摘要

背景与目的

炎症性肠病（IBD）是慢性疾病，会导致身体、社会和经济负担。生成式人工智能（AI），尤其是ChatGPT，因其支持医疗实践的潜力而受到关注。然而，人们对其回答的可靠性和一致性仍存在担忧。本研究系统回顾了关于生成式AI在IBD中作用的现有证据。

材料与方法

我们按照PRISMA指南进行了系统的文献综述。通过PubMed和Embase（2020年1月至2024年9月）确定了调查生成式AI在IBD护理中应用的研究。

结果

从2875条记录中，8项研究（2023 - 2024年）符合纳入标准：5项关于患者教育，2项关于决策支持，1项关于研究构思。在患者教育方面，ChatGPT提供了清晰准确的回答，在一项研究中准确率达到84.2%，不过有时缺乏一致性。在决策支持方面，ChatGPT对溃疡性结肠炎严重程度的分类在80%的病例中与临床医生评估一致，在基于指南的发育异常管理方面，这一比例为87.8%。在研究构思方面，ChatGPT生成的问题高度相关（平均得分：4.9±0.26）且清晰（4.8±0.41），但缺乏特异性（2.86/5）和原创性（1.07/5）。