Using ChatGPT to perform a systematic review: a tutorial

Abstract

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.