F1000Research, Journal Year: 2025, Volume and Issue: 10, P. 401 - 401

Published: April 8, 2025

Background The reliable and usable (semi) automation of data extraction can support the field systematic review by reducing workload required to gather information about conduct results included studies. This living examines published approaches for from reports clinical Methods We systematically continually search PubMed, ACL Anthology, arXiv, OpenAlex via EPPI-Reviewer, dblp computer science bibliography databases. Full text screening are conducted using a mix open-source commercial tools. update includes publications up August 2024 content September 2024. Results 117 in this review. Of these, 30 (26%) used full texts while rest titles abstracts. A total 112 (96%) developed classifiers randomised controlled trials. Over entities were extracted, with PICOs (population, intervention, comparator, outcome) being most frequently extracted. Data available 53 (45%), code 49 (42%) publications. Nine (8%) implemented publicly Conclusions presents an overview (semi)automated data-extraction literature interest different types identified broad evidence base describing interventional reviews small number extracting other study types. Between updates, large language models emerged as new tool extraction. While facilitating access automated extraction, they showed trend decreasing quality reporting, especially quantitative such recall lower reproducibility results. Compared previous update, trends transition relation sharing datasets stayed similar.

Language: Английский