Presentation (CTI)

OBJECTIVES: Oncology real-world evidence (RWE) generation is structurally inefficient; each new question often triggers repeat screening despite substantial overlap in the underlying evidence. RWE heterogeneity complicates organization across projects, while current artificial intelligence (AI) tools largely prioritize single-project throughput over evidence retention. This proof-of-concept proposes a living evidence library to support the internal evidence needs of pharmaceutical manufacturers and health technology assessment submissions.
METHODS: A reusable evidence-management framework was developed as an oncology RWE evidence library. Study-level records were stored, screened, and extracted into a filterable repository spanning clinical, economic, humanistic, and epidemiologic domains. Large language models (LLMs) supported structured extraction and classification of study attributes, including population, intervention, comparator, outcomes, and study design (PICOS), enabling consistent indexing across heterogeneous study types. Human expertise was embedded throughout library development to curate extracted evidence and resolve contextual interpretation.
RESULTS: Unlike single-question living reviews, the living evidence library enabled reuse across research questions. Broad PICOS criteria captured evidence within and across oncology indications, while AI-enabled processing made it feasible to organize large literature volumes in a single reusable repository rather than repeat separate searches for related questions. When a new research question arose, teams queried the repository by filtering relevant attributes instead of initiating a new evidence-generation cycle. Review effort shifted towards newly published studies not yet captured in the library and targeted quality checks tailored to the question. Periodic updates supported incremental repository growth without repeating prior work.
CONCLUSIONS: Oncology RWE synthesis has traditionally followed project-by-project logic, generating evidence that is rarely reused. The living evidence library changes this model. Each review cycle expands an organized evidence base rather than producing another one-time output. As submission demands and evidence volumes grow, retrieving previously curated evidence instead of rebuilding it offers a practical and sustainable foundation for oncology evidence generation.