Trends of Authors' Conflicts of Interest in Clinical Trials Published in the Journal of Clinical Oncology: A Large Language Model-Assisted Longitudinal Study.
Jiasheng Wang, Pedro C Silberman, Sayan Mullick Chowdhury, Bradley W Blaser
{"title":"Trends of Authors' Conflicts of Interest in Clinical Trials Published in the <i>Journal of Clinical Oncology</i>: A Large Language Model-Assisted Longitudinal Study.","authors":"Jiasheng Wang, Pedro C Silberman, Sayan Mullick Chowdhury, Bradley W Blaser","doi":"10.1200/OP-25-00293","DOIUrl":null,"url":null,"abstract":"<p><strong>Purpose: </strong>Conflicts of interest (COIs) between clinical trial investigators and biopharmaceutical companies have raised concerns about potential bias in research. This study aimed to systematically analyze the prevalence and trends of COIs in oncology clinical trials published in the <i>Journal of Clinical Oncology</i> (JCO) for the past 15 years and to demonstrate the utility of large language models (LLMs) for automated data extraction in this context.</p><p><strong>Methods: </strong>We identified clinical trials published in the JCO from 2010 to 2025 using PubMed. We extracted publication data and author disclosures from the JCO Web site. OpenAI's GPT-4o was used to identify the main medical product studied and the related biopharmaceutical company and their variants in author disclosures. We then analyzed COI trends across three time periods (2010-2015, 2015-2020, 2020-2025).</p><p><strong>Results: </strong>GPT-4o demonstrated close to 95% accuracy in identifying medical products and companies. Of the 2,583 clinical trials, 2,219 (85.9%) involved a medical product. Among these, 1,610 (72.6%) had at least one author with a COI related to the associated biopharmaceutical company. COI prevalence increased from 70.0% (2010-2015) to 77.0% (2015-2020), and then decreased to 72.0% (2020-2025). Company employment, advisory roles, and honoraria were common COI types and followed similar trends. US-led studies had a significantly higher COI prevalence than those from other regions (77.6% <i>v</i> 67.3%; <i>P</i> < .001). Additionally, 61.9% of first or last authors had a COI, which increased consistently over three time periods.</p><p><strong>Conclusion: </strong>This study reveals widespread COIs in oncology clinical trials, particularly in US-led studies and among leading authors, with discernible temporal patterns. The LLM-based method provides an efficient solution for COI monitoring, promoting transparency in biomedical research.</p>","PeriodicalId":14612,"journal":{"name":"JCO oncology practice","volume":" ","pages":"OP2500293"},"PeriodicalIF":4.6000,"publicationDate":"2025-07-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"JCO oncology practice","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1200/OP-25-00293","RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"ONCOLOGY","Score":null,"Total":0}
引用次数: 0
Abstract
Purpose: Conflicts of interest (COIs) between clinical trial investigators and biopharmaceutical companies have raised concerns about potential bias in research. This study aimed to systematically analyze the prevalence and trends of COIs in oncology clinical trials published in the Journal of Clinical Oncology (JCO) for the past 15 years and to demonstrate the utility of large language models (LLMs) for automated data extraction in this context.
Methods: We identified clinical trials published in the JCO from 2010 to 2025 using PubMed. We extracted publication data and author disclosures from the JCO Web site. OpenAI's GPT-4o was used to identify the main medical product studied and the related biopharmaceutical company and their variants in author disclosures. We then analyzed COI trends across three time periods (2010-2015, 2015-2020, 2020-2025).
Results: GPT-4o demonstrated close to 95% accuracy in identifying medical products and companies. Of the 2,583 clinical trials, 2,219 (85.9%) involved a medical product. Among these, 1,610 (72.6%) had at least one author with a COI related to the associated biopharmaceutical company. COI prevalence increased from 70.0% (2010-2015) to 77.0% (2015-2020), and then decreased to 72.0% (2020-2025). Company employment, advisory roles, and honoraria were common COI types and followed similar trends. US-led studies had a significantly higher COI prevalence than those from other regions (77.6% v 67.3%; P < .001). Additionally, 61.9% of first or last authors had a COI, which increased consistently over three time periods.
Conclusion: This study reveals widespread COIs in oncology clinical trials, particularly in US-led studies and among leading authors, with discernible temporal patterns. The LLM-based method provides an efficient solution for COI monitoring, promoting transparency in biomedical research.