Presentation (CTI)

OBJECTIVES: To automate the end-to-end Systematic Literature Review (SLR) process by combining unbiased AI agents with human expertise through a PRISMA-guided workflow. The platform tends to enhance SLR quality through Quality Assurance (QA) checkpoints across SLR steps, while ensuring secured access via role-based access control (RBAC).
METHODS: The platform employed a three-phase approach: in phase 1, multiple human reviewers work in parallel with an unbiased AI agent to perform title and abstract screening followed by QA checkpoints and conflict resolution via an interactive dashboard by SME. Phase 2 involved full-text search followed by screening of full-text articles with AI agent and humans in parallel. In the final phase, a RAG-enabled AI chatbot powered by AWS Claude and integrated with a Lang Chain-based agent framework performs quick data extraction and analysis across multiple studies, delivering comprehensive results within seconds. The platform facilitates project-specific team creation for secured access and workflow management through an interactive interface.
RESULTS: Embase®, Medline®, and Cochrane databases were searched to identify relevant randomized controlled trials in the schizophrenia. The performance of the unbiased AI agent evaluated and achieved 94.69% screening accuracy, 88.59 sensitivity and 95.78% specificity and processed 980 abstracts in 45 minutes (6-10 seconds/batch, running 3 documents in batches). The automated workflow reduced review completion time by 40% compared to traditional methods while maintaining compliance with recommended two review and QC process along with multiple QA checks.
CONCLUSIONS: MetaSLR demonstrates the integration of AI-powered review with human expertise, achieving significant time reduction while maintaining PRISMA compliance. Future enhancements will optimize performance through increased parallel agent processing and batch operations, improved scalability, and extending functionality to include automated SLR report generation, further streamlining the end-to-end review process.