Artificial intelligence for evidence synthesis
Period: January 2024 - December 2024
In recent years, the number of scientific publications has increased significantly, partly driven by a growing number of scientific journals, many of which are primarily published in digital formats. This increase presents an additional workload in evidence synthesis, where scientific findings are aggregated and summarized. In healthcare, systematic reviews are conducted to synthesise the current body of knowledge, providing a foundation for (political) decisions. These reviews involve searching scientific databases for specific keywords, followed by a screening process in which the abstracts of the publications and then the full texts are evaluated to determine which publications are to be included or excluded from the review. Due to the large number of results, analysing the abstracts is often the most time-consuming step in preparing such reviews.
For systematic reviews that focus on specific questions, tools based on artificial intelligence (AI) are increasingly being used to streamline the process. Scoping reviews, however, address broader research questions and use search strategies based on concepts (e.g., ‘digital tools’ rather than ‘digital patient documentation’). As a result, they have not benefited as much from AI support.
The rise of chatbots powered by large language models offers new opportunities for the use of AI in the process of conducting scoping reviews, possibly enhancing efficiency. This project aims to evaluate how such chatbots compare to conventional AI tools and human researchers in terms of their efficiency in abstract screening and their reliability in reproducing results. Additionally, the potential time and cost savings of using this technology will be analysed.
