Development and performance of a generative pretrained transformer for diabetes care

IF 7.4 3区医学 Q1 ENDOCRINOLOGY & METABOLISM

Diabetes research and clinical practice Pub Date : 2025-08-22 DOI:10.1016/j.diabres.2025.112425

Miguel Garrido-Bueno , Pilar Santa Cruz-Álvarez , Manuel Pabón-Carrasco , Rocío Romero-Castillo

{"title":"Development and performance of a generative pretrained transformer for diabetes care","authors":"Miguel Garrido-Bueno , Pilar Santa Cruz-Álvarez , Manuel Pabón-Carrasco , Rocío Romero-Castillo","doi":"10.1016/j.diabres.2025.112425","DOIUrl":null,"url":null,"abstract":"<div><h3>Aims</h3><div>To design and evaluate the performance of a diabetes-related Generative Pretrained Transformer (GPT).</div></div><div><h3>Methods</h3><div>A prompt-engineered layer over GPT was developed in four stages: (1) literature review on GPT tools development; (2) selection and preprocessing of 65 information sources about diabetes care strategies, patient education, diabetes technologies, and cultural care, among others; (3) prototype development; and (4) final tool evaluation using 420 diabetes-related questions adapted from three validated instruments. Outcomes were accuracy, rationale, citations, disclaimers, and emoji exclusion. Statistical analyses included descriptive statistics, chi-square tests and bias assessment. Compliance with data protection regulations and ethical standards was ensured.</div></div><div><h3>Results</h3><div>Diabetes Help GPT showed high overall accuracy (91.7 %), with 100 % rationale inclusion, 93.3 % citations, 84.8 % disclaimers, and minimal emoji use (13.3 %). Accuracy was highest in general diabetes knowledge and nutrition questions; slightly lower in insulin-related items (82.3 %). Disclaimer and emoji usage varied significantly by question format (p = 0.026 and p < 0.001). No accuracy bias was detected.</div></div><div><h3>Conclusions</h3><div>Diabetes Help GPT delivers accurate, well-sourced responses, supporting healthcare professionals in diabetes care. Unlike existing GPT models in medicine, it was developed through a transparent, expert-led process with curated content and iterative validation. It should complement, and not replace, professionals’ criteria.</div></div>","PeriodicalId":11249,"journal":{"name":"Diabetes research and clinical practice","volume":"227 ","pages":"Article 112425"},"PeriodicalIF":7.4000,"publicationDate":"2025-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Diabetes research and clinical practice","FirstCategoryId":"3","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0168822725004395","RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"ENDOCRINOLOGY & METABOLISM","Score":null,"Total":0}

引用次数: 0

Abstract

Aims

To design and evaluate the performance of a diabetes-related Generative Pretrained Transformer (GPT).

Methods

A prompt-engineered layer over GPT was developed in four stages: (1) literature review on GPT tools development; (2) selection and preprocessing of 65 information sources about diabetes care strategies, patient education, diabetes technologies, and cultural care, among others; (3) prototype development; and (4) final tool evaluation using 420 diabetes-related questions adapted from three validated instruments. Outcomes were accuracy, rationale, citations, disclaimers, and emoji exclusion. Statistical analyses included descriptive statistics, chi-square tests and bias assessment. Compliance with data protection regulations and ethical standards was ensured.

Results

Diabetes Help GPT showed high overall accuracy (91.7 %), with 100 % rationale inclusion, 93.3 % citations, 84.8 % disclaimers, and minimal emoji use (13.3 %). Accuracy was highest in general diabetes knowledge and nutrition questions; slightly lower in insulin-related items (82.3 %). Disclaimer and emoji usage varied significantly by question format (p = 0.026 and p < 0.001). No accuracy bias was detected.

Conclusions

Diabetes Help GPT delivers accurate, well-sourced responses, supporting healthcare professionals in diabetes care. Unlike existing GPT models in medicine, it was developed through a transparent, expert-led process with curated content and iterative validation. It should complement, and not replace, professionals’ criteria.

查看原文本刊更多论文

糖尿病护理生成式预训练变压器的开发和性能

目的设计并评估糖尿病相关生成式预训练变压器（GPT）的性能。方法分四个阶段构建GPT上的即时工程层：(1)综述GPT工具开发的相关文献；(2)糖尿病护理策略、患者教育、糖尿病技术、文化护理等65个信息源的选择与预处理；(3)原型开发；(4)最后的工具评估使用420糖尿病相关问题改编自三个验证的工具。结果是准确性、基本原理、引用、免责声明和表情符号排除。统计分析包括描述性统计、卡方检验和偏倚评估。确保遵守数据保护条例和道德标准。结果diabetes Help GPT显示出较高的总体准确性（91.7%），其中100%的基本原理包含，93.3%的引用，84.8%的免责声明和最少的表情符号使用（13.3%）。一般糖尿病知识和营养问题的准确性最高；胰岛素相关项目略低（82.3%）。免责声明和表情符号的使用因问题格式而有显著差异（p = 0.026和p <； 0.001）。未发现准确性偏差。结论diabetes Help GPT提供准确、来源良好的响应，支持医疗保健专业人员进行糖尿病护理。与医学中现有的GPT模型不同，它是通过一个透明的、专家主导的过程开发的，具有精心策划的内容和反复验证。它应该补充而不是取代专业人士的标准。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Diabetes research and clinical practice 医学-内分泌学与代谢

CiteScore

10.30

自引率

3.90%

发文量

862

审稿时长

32 days

期刊介绍： Diabetes Research and Clinical Practice is an international journal for health-care providers and clinically oriented researchers that publishes high-quality original research articles and expert reviews in diabetes and related areas. The role of the journal is to provide a venue for dissemination of knowledge and discussion of topics related to diabetes clinical research and patient care. Topics of focus include translational science, genetics, immunology, nutrition, psychosocial research, epidemiology, prevention, socio-economic research, complications, new treatments, technologies and therapy.