Exploration of AI-Tools Used in Systematic Reviews and Evidence Synthesis

OBJECTIVES: The most reliable method for synthesizing evidence is systematic review (SR). Despite being a crucial component of evidence-based decision-making, SRs are the traditional, expensive, time-consuming, and resource-intensive approach. To aid in the process, many tools with integrated artificial intelligence (AI) are now available. But there is lack of literature on the capabilities of the available tools. Hence, we conducted a review to investigate the AI features of the various web-based and software tools and their capabilities in the entire SR process.

METHODS: We conducted a search of previously published literature and a google search to identify the available AI tools used for conducting SR. We included only tools that do not require coding for conducting SR or evidence synthesis. Tools that required coding were not considered.

RESULTS: We found 37 SR tools, of which, 11 were excluded as they did not provide details on AI integration. Of the remaining 26, 23 tools used AI integration for screening, 10 for extraction, and 2 for report writing. DistillerSR, Covidence, Nested Knowledge, SWIFT-Active Screener, EasySLR, and LASERAI integrated AI or ML to support prescreening, screening and extraction (auto-populating highlighted qualitative information). Abstrackr, Litsuggest, Rayyan, SysRev, Research Screener, AS Review, SR Accelerator, LitStream, SyRF, and Sorcero iSLR tools were predominantly focused on the integration of AI during the screening. Covidence, DistillerSR, Nested Knowledge, Rayyan, RobotAnalyst, SysRev, and EasySLR have integrated AI to autonomously train the machine language-based model to predict the likelihood that records are screened based on the previous decisions made by the human reviewer.

CONCLUSIONS: Most of the available tools integrated AI for screening abstracts and full texts whereas very few tools explored AI for the data extraction process. Perhaps recommendations from HTA agencies on acceptance of AI augmented SRs may encourage AI developers to build more robust tools.