Teperikidis Lefteris, Boulmpou Aristi, Papadopoulos Christodoulos, Biondi-Zoccai Giuseppe
Third Department of Cardiology, Ippokratio General Hospital, Aristotle University of Thessaloniki, Thessaloniki, Greece -
Clinical Research Unit, Special Unit for Biomedical Research and Education (SUBRE), School of Medicine, Aristotle University of Thessaloniki, Thessaloniki, Greece -
Minerva Cardiol Angiol. 2024 Dec;72(6):547-567. doi: 10.23736/S2724-5683.24.06568-2. Epub 2024 Jul 26.
This tutorial provides a comprehensive guide on leveraging ChatGPT for systematic literature reviews, leveraging actual applications in cardiovascular research. Systematic reviews, while essential, are resource-intensive, and ChatGPT offers a potential solution to streamline the process. The tutorial covers the entire review process, from preparation to finalization. In the preparation phase, ChatGPT assists in defining research questions and generating search strings. During the screening phase, ChatGPT can efficiently screen titles and abstracts, processing multiple abstracts simultaneously. The tutorial also introduces an intermediate step of generating study summaries that leads to the generation of reliable data extraction tables. For assessing the risk of bias, ChatGPT can be prompted to perform these tasks. Using each tool's explanation document to generate an appropriate prompt is an efficient method of reliable risk of bias assessments using ChatGPT. However, users are cautioned about potential hallucinations in ChatGPT's outputs and the importance of manual validation. The tutorial emphasizes the need for vigilance, continuous refinement, and gaining experience with ChatGPT to ensure accurate and reliable results. The methods presented have been successfully tried in several projects, but they remain in nascent stages, with ample room for improvement and refinement.
本教程提供了一份全面指南，介绍如何利用ChatGPT进行系统文献综述，并展示了其在心血管研究中的实际应用。系统综述虽然至关重要，但资源消耗大，而ChatGPT为简化这一过程提供了一种可能的解决方案。该教程涵盖了从准备到完成的整个综述过程。在准备阶段，ChatGPT协助定义研究问题并生成检索词。在筛选阶段，ChatGPT能够高效地筛选标题和摘要，同时处理多篇摘要。本教程还引入了生成研究总结的中间步骤，这有助于生成可靠的数据提取表。对于评估偏倚风险，可促使ChatGPT执行这些任务。利用每个工具的解释文档来生成合适的提示，是使用ChatGPT进行可靠偏倚风险评估的有效方法。然而，需提醒用户注意ChatGPT输出中可能出现的幻觉以及人工验证的重要性。本教程强调了保持警惕、持续完善以及积累使用ChatGPT经验的必要性，以确保结果准确可靠。所介绍的方法已在多个项目中成功尝试，但仍处于初期阶段，有很大的改进和完善空间。
