Presentation (CTI)

OBJECTIVES: The accelerating volume and complexity of oncology evidence have made traditional manual SLRs increasingly difficult to conduct, maintain, and update. This study describes the design, implementation, and performance of a proprietary DMS developed to automate the most resource-intensive SLR components while preserving methodological rigor, transparency, and auditability.
METHODS: The DMS architecture is built on a MongoDB NoSQL database with an Angular-based data editor supporting full CRUD (Create, Read, Update, Delete) operations. Automated components include structured search execution in PubMed+conferences, deterministic rule-based deduplication, and automated master records (MRs) creation for each unique study. An integrated agentic large language model (aLLM) system with retrieval-augmented generation (RAG) architecture supports screening+extraction, with human quality control and strategic data aggregation. Each master record is cross-linked to external regulatory, guideline, and reimbursement resources.
RESULTS: System implementation resulted in three major efficiency gains. First, automated search execution, deduplication, screening, and data extraction are pre-scheduled and completed in ~5 minutes, compared with 6 hours required for a manual daily update (98.6% time reduction). Second, the DMS automatically generates MRs that aggregate study/publication-level data while prioritizing recent/relevant efficacy; human review of a pre-generated MRs requires ~5 minutes, compared with 15 minutes for manual aggregation (66.7% time reduction). Third, the DMS functions as a centralized data governance framework, passively maintaining data hygiene, versioning, and linkage of publications to unique studies (~20% reduction vs. ongoing project management and quality control effort). The system currently supports Living-SLRs containing 3,898 unique studies across four cancers: 1,527 non-small-cell lung, 924 breast, 738 prostate cancer, and 709 multiple myeloma, with new evidence incorporated daily.
CONCLUSIONS: A purpose-built DMS substantially improves the scalability, efficiency, and sustainability of oncology SLRs. Combining programmatic workflows, rule-based automation, and aLLM/RAG-assisted data review+extraction, enables true Living-SLR and supports faster HTA, and regulatory decision making in rapidly evolving cancers.