ChatGPT 能否帮助患者了解男科疾病?

Can ChatGPT help patients understand their andrological diseases?

机构信息

Kızılcahamam State Hospital, 06890 Ankara, Turkey.

Etlik City Hospital, 06010 Ankara, Turkey.

出版信息

Rev Int Androl. 2024 Jun;22(2):14-20. doi: 10.22514/j.androl.2024.010. Epub 2024 Jun 30.

Abstract

We aimed to assess the reliability of Chat Generative Pre-training Transformer (ChatGPT)'s andrology information and its suitability for informing patients and medical students accurately about andrology topics. We presented a series of systematically organized frequently asked questions on andrology topics and sentences containing strong recommendations from the European Association of Urology (EAU) Guideline to ChatGPT-3.5 and 4.0 as questions. These questions encompassed Male Hypogonadism, Erectile Dysfunction and Sexual Desire Disorder, Disorders of Ejaculation, Penile Curvature and Penile Size Abnormalities, Priapism, and Male Infertility. Two expert urologists independently evaluated and assigned scores ranging from 1 to 4 to each response based on its accuracy, with the following ratings: (1) Completely true, (2) Accurate but insufficient, (3) A mixture of accurate and misleading information, and (4) Completely false. A total of 120 questions were included in the study. Among these questions, 50.0% received a grade of 1 (completely correct) (55.4% for 4.0 version). The combined rate of correct answers (grades 1 and 2) was 85.2% for frequently asked questions (88.8% for 4.0 version) and 81.5% for questions obtained from the guideline. The rate of completely incorrect answers (grade 4) was 1.8% for frequently asked questions (0% for 4.0 version) and 5.2% for questions based on strong recommendations. The response rate of version 4.0 to questions created from sentences containing strong recommendations from the EAU guideline was the same as version 3.5. ChatGPT provided satisfactory answers to the questions asked, although some responses lacked completeness. It may be beneficial to utilize ChatGPT under the guidance of a urologist to enhance patients' comprehension of their andrology issues.

摘要

我们旨在评估 ChatGPT 在男科信息方面的可靠性,以及其是否适合准确地向患者和医学生提供男科知识。我们向 ChatGPT-3.5 和 4.0 提出了一系列经过系统组织的男科常见问题,以及包含欧洲泌尿外科学会 (EAU) 指南强烈推荐的句子。这些问题涵盖了男性性腺功能减退症、勃起功能障碍和性欲障碍、射精障碍、阴茎弯曲和阴茎大小异常、阴茎异常勃起和男性不育症。两位专家泌尿科医生独立评估并根据其准确性为每个回复分配 1 到 4 分的评分,以下是评分标准:(1)完全正确,(2)准确但不充分,(3)准确与误导信息的混合,以及(4)完全错误。该研究共包含 120 个问题。其中,50.0%的问题获得了 1 分(完全正确)(4.0 版本为 55.4%)。正确答案(1 分和 2 分)的综合比例为常见问题的 85.2%(4.0 版本为 88.8%)和指南问题的 81.5%。完全错误答案(4 分)的比例为常见问题的 1.8%(4.0 版本为 0%)和指南问题的 5.2%。4.0 版本对基于 EAU 指南强烈推荐的句子生成的问题的回复率与 3.5 版本相同。ChatGPT 对提出的问题提供了满意的答案,尽管有些回答不够完整。在泌尿科医生的指导下使用 ChatGPT 可能有助于提高患者对其男科问题的理解。

文献AI研究员

20分钟写一篇综述,助力文献阅读效率提升50倍

立即体验

用中文搜PubMed

大模型驱动的PubMed中文搜索引擎

马上搜索