清晨好,您是今天最早来到科研通的研友!由于当前在线用户较少,发布求助请尽量完整地填写文献信息,科研通机器人24小时在线,伴您科研之路漫漫前行!

Evaluation of GPT-4 for common questions regarding anti-obesity medications. A comparative analysis

作者
Monica Aggarwal,Saam Dilmaghani,Raseen Tariq,Andres J Acosta
出处
期刊:Endoscopy [Thieme Medical Publishers (Germany)]
卷期号:56 (S 02): S191-S191
标识
DOI:10.1055/s-0044-1783120
摘要

Aims Large language models (LLM) such as GPT-4 are being increasingly used to answer common medical questions. The accuracy and understandability of responses generated by LLMs as it relates to anti-obesity medications (AOMs) is unknown. This study evaluated GPT-4 generated responses to common AOM related questions and compared these to Food and Drug Administration (FDA) patient information sheet. Methods We prompted GPT-4 LLM (Oct 2023) with five commonly asked questions by patients regarding FDA-approved AOMs (Liraglutide, Semaglutide, Phenteramine/Toiramate, Bupropion/Naltrexone and Orlistat). Prompts evaluated 5 major themes: mechanism of action, serious adverse events, concerns with pregnancy and lactation, missing a dose of drug, and expected weight loss. Responses generated by GPT-4 were compared to FDA responses to the same questions as extracted from FDA patient information sheet. In some cases, formatting (but not the content) of the FDA response was modified to preserve blinding. Each response (FDA or GPT) was evaluated by 5 physicians (3 board certified internal medicine and 2 board certified gastroenterologists). Physicians were blinded to the source of the response. Responses were graded by a 5-point Likert scale on accuracy of the content and ease in understanding, as well as the usability for patient communication (yes/no). Results Each physician graded 49 responses (25 GPT-4, 24 FDA [FDA information sheet does not contain a response regarding efficacy of Orlistat]). Pooled accuracy of identifying whether a response was GPT generated was 49.3% (95% Confidence interval, 42.9-55.8%) suggesting that the GPT responses are indistinguishable from the FDA. GPT-4 generated responses were deemed to be either mostly or completely accurate in 90.4% of cases while two responses were marked as mostly inaccurate. Both inaccurate responses were regarding expected weight loss (One each for Bupropion/Naltrexone and Semaglutide). Proportion of responses graded to be ‘easy’ or ‘very easy’ to understand were similar between GPT-4 generated and FDA responses (82.4% vs. 78.4% respectively, p=0.8). Four responses (2 each by GPT-4 and FDA) were graded as difficult to understand. Respondents would use a similar proportion of GPT-4 and FDA responses in patient communications (72.8 vs. 67.5%, p=0.3). Conclusions GPT-4 generates accurate and easy to understand responses to common questions regarding AOM. Due to their ease of understanding, GPT-4 generated responses can be used in patient communication after physician review. The expected weight loss does not appear to be reliably answered using GPT-4. Publication History Article published online: 15 April 2024 © 2024. European Society of Gastrointestinal Endoscopy. All rights reserved. Georg Thieme Verlag KG Rüdigerstraße 14, 70469 Stuttgart, Germany

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
4秒前
大胆半双完成签到,获得积分10
10秒前
安详冰夏完成签到,获得积分10
24秒前
传奇3的应助被yiyixt采纳,获得10
25秒前
28秒前
喵喵发布了新的文献求助10
32秒前
风趣馒头完成签到,获得积分10
35秒前
喵喵完成签到 ,获得积分10
42秒前
月儿完成签到 ,获得积分0
52秒前
酷炫安雁完成签到,获得积分10
57秒前
58秒前
zy997987876完成签到,获得积分10
59秒前
舒心的以莲完成签到,获得积分10
1分钟前
1分钟前
369ninja发布了新的文献求助30
1分钟前
合适的太英完成签到,获得积分10
1分钟前
muyicccccc完成签到 ,获得积分10
2分钟前
潇洒小海豚完成签到,获得积分10
2分钟前
2分钟前
谦让大米完成签到,获得积分10
2分钟前
2分钟前
无辜摇伽完成签到,获得积分10
2分钟前
2分钟前
2分钟前
3分钟前
3分钟前
3分钟前
3分钟前
3分钟前
3分钟前
3分钟前
3分钟前
3分钟前
3分钟前
3分钟前
3分钟前
3分钟前
3分钟前
Perse完成签到,获得积分10
3分钟前
3分钟前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1700
The World Atlas of Wine 9th edition 1000
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
John McDowells Bildungstheorie 800
Social Psychology in the Real World 600
八旗制度の研究 600
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8179961
求助须知:如何正确求助?哪些是违规求助? 9570183
关于积分的说明 21808480
捐赠科研通 7775832
什么是DOI,文献DOI怎么找? 3433251
关于科研通互助平台的介绍 2557373
邀请新用户注册赠送积分活动 2478958