Building a Living, Ontology-Informed and Large Language Model-Assisted Knowledge Base for Cancer Screening Implementation: A Study Protocol

<ns5:p>Background Although numerous implementation strategies have been evaluated to improve cancer screening uptake, the evidence base remains fragmented, inconsistently described, and difficult to apply to local contexts. This project will develop a living, ontology-informed, and large language model (LLM)-assisted knowledge base for cancer screening implementation. This resource will serve as a structured, continuously updated, and computable evidence infrastructure, enabling evidence mapping, synthesis, and retrieval of context -sensitive implementation strategies to support evidence-informed implementation decision-making . Methods First, a cancer screening implementation ontology will be developed, building on the Behavior Change Intervention Ontology to formally represent screening-specific pathways, modalities, implementation determinants, strategies, and outcomes. Second, guided by the Cochrane Effective Practice and Organization of Care method, we will systematically identify randomized controlled trials evaluating cancer screening implementation strategies from four databases: PubMed, EMBASE, CINAHL, and PsycINFO, from database inception to May 13, 2026. Information from eligible studies will be extracted and standardized using ontology. All the above steps will be supported by a human-LLM collaborative workflow that incorporates one or more LLMs with source-grounded verification and expert review to improve efficiency while maintaining human accountability. The outputs will include an interactive knowledge base that supports evidence visualization, synthesis, structured comparison, and future decision support. The knowledge base will be continuously updated and progressively expanded to incorporate additional forms of implementation evidence. Workflow performance and knowledge base quality will be evaluated using predefined process and outcome metrics. Discussion The proposed knowledge base will improve the standardization, comparability, and usability of implementation evidence in cancer screening by transforming heterogeneous trial evidence into structured and computable evidence units. Future prediction or decision-support applications, when developed, will contribute to emerging efforts in precision implementation science by enabling a more context-sensitive understanding of cancer screening strategies.</ns5:p>