Medical journals serve as the cornerstone for the transmission of medical knowledge, clinical decision-making, and the formulation of public health policies. The quality and integrity of the research they publish directly impact the credibility of scientific progress and patient safety. Within the current scholarly publishing system, peer review is regarded as the core mechanism safeguarding this quality—a critical process for the systematic and critical evaluation of original research. However, this system is under dual pressure. On one hand, the exponential growth of global research output makes it difficult for reviewers to comprehensively track the vast volume of multilingual literature, amplifying the inherent limitations of their knowledge breadth. On the other hand, technological misuse has given rise to novel forms of academic misconduct, ranging from the mass production by "paper mills" to the use of generative AI for semantic paraphrasing to evade plagiarism detection, making the identification of manuscript authenticity and originality unprecedentedly complex. The rise of generative artificial intelligence technology, represented by large language models (LLMs), presents a transformative and complex solution to this predicament. It demonstrates potential as a powerful auxiliary tool: by rapidly parsing and synthesizing vast amounts of text, it can provide reviewers with more comprehensive research background support; through automated screening, it can assist in detecting methodological flaws, anomalies in statistical analysis, or adherence to reporting guidelines (e.g., CONSORT, PRISMA), thereby enhancing the rigor and efficiency of the review process. Recent studies indicate that advanced LLMs have shown impressive accuracy in automatically assessing the compliance of research reports with these guidelines. Yet, the same technology can also be leveraged to generate or deeply alter academic text, posing a new threat to the integrity of journal content and making the development and deployment of effective AI-generated content detection tools an urgent need for editorial offices. This novel risk—how to effectively identify content generated or substantially modified by AI—has become a new challenge facing editors. Confronted with this technological wave, the international publishing community has responded swiftly, though consensus and norms are still evolving. From the principle statements of organizations like the International Committee of Medical Journal Editors (ICMJE) and the Committee on Publication Ethics (COPE), to the varied usage policies of top-tier journals such as Nature, Science, and The Lancet, a global discussion and practical exploration on how to responsibly integrate AI is intensifying. Although related discussions and practices are burgeoning, the academic community still lacks a systematic synthesis of this field. Existing literature primarily consists of technical reports focusing on single application scenarios (e.g., language polishing, plagiarism detection) or opinion pieces centered on ethics and policies, lacking integrative research that provides a panoramic view of the application status, evidence base, practical models, and core controversies of AI across the entire workflow of medical journal editing—from initial manuscript screening to post-publication dissemination. Therefore, this study aims to employ a scoping review methodology to, for the first time, systematically synthesize the existing evidence in this field. It seeks to clarify the application landscape of AI technology across the complete editorial chain, identify research hotspots and evidence gaps, and thereby provide a solid foundation for evidence-based decision-making by journal editorial offices, direction-setting for technology developers, and the refinement of relevant academic policies.
Bingyi Wang (Thu,) studied this question.