JMIR Publications

ABSTRACT

Systematic literature reviews (SLRs) are critical for evidence synthesis and play a central role in supporting research, policy development, and evidence‑based decision‑making across healthcare and related disciplines. However, traditional SLRs are resource-intensive and time-consuming, often requiring months of manual effort to screen thousands of records, extract data, and maintain methodological rigor. As global scientific output continues to grow exponentially, these operational challenges have intensified, contributing to longer completion timelines, greater workforce burden, and a heightened risk that reviews become outdated shortly after publication. In response, there is growing interest in artificial intelligence (AI) approaches to enhance the efficiency and scalability of SLRs. AI tools now offer support for a broad range of review tasks, including assisting with search strategy development, identifying relevant concepts, prioritizing records for screening, and supporting data extraction and risk‑of‑bias assessments. AI can significantly accelerate labor-intensive stages, reduce human error during repetitive tasks, and enable the synthesis of evidence bases that might otherwise be impractical to review manually. However, these efficiencies must be balanced against the risks associated with AI, including bias, lack of transparency, variable outputs, and hallucinations (outputs that appear plausible but are factually incorrect). A human-in-the-loop approach is therefore essential to validate AI outputs and maintain scientific integrity. Human expertise remains critical for defining research questions, validating search strategies, confirming study eligibility, interpreting nuanced data, and making final judgments on quality and risk of bias. Clear methodological guidance is required to support teams in integrating AI tools responsibly, transparently, and reproducibly into SLR workflows. Methodological considerations include selecting appropriate tools, defining oversight strategies, and applying performance metrics such as precision and recall. This paper aims to provide methodological guidance on the effective integration of AI into each stage of the SLR process, drawing on both published literature and the authors’ real-world experience. We outline key considerations for selecting and implementing AI tools while maintaining human oversight. We also discuss how to maintain transparency, auditability, and alignment with established standards, including PRISMA‑P, PRISMA‑trAIce, and emerging guidance from regulators and health technology assessment bodies. We also present future directions for responsible AI use in SLRs. AI should complement, not replace, human judgment. When implemented within a human-in-the-loop framework, AI has the potential to accelerate evidence synthesis, enabling faster, scalable, and rigorous reviews while preserving transparency and reproducibility.