Presentation (CTI)

OBJECTIVES: Cardiovascular-kidney-metabolic (CKM) syndrome comprises interconnected diseases spanning multiple organ systems, creating challenges for evidence synthesis and health economic model development. We developed a large-scale knowledge graph, CKM-KG, and evaluated its utility for identifying clinically meaningful disease relationships to inform HEOR model conceptualization using hypertension as a case study.
METHODS: CKM-KG was developed using a multi-agent AI framework combining large language model-based information extraction, biomedical entity normalization, and automated knowledge integration across over one million CKM-relevant publications, clinical trials, and curated databases. Qwen model, fine-tuned using LoRA and optimized with Group Relative Policy Optimization, was used to identify and structure diseases, risk factors, biomarkers, outcomes, and interventions into provenance-preserving knowledge units. Unsupervised graph-learning and clustering methods characterized CKM-domain connectivity patterns, classified biomedical entities by functional role, and identified cross-system disease relationships.
RESULTS: CKM-KG comprised 6.6 million relationships connecting 1.2 million biomedical entities. Hypertension emerged as a central cross-system hub linking cardiovascular, kidney, and metabolic disease pathways. Among 23,646 entities acting as risk factors for at least one CKM domain, only 1,411 (6%) were shared drivers across all three domains, with hypertension, ranking first. Risk factor analysis showed that hypertension's influence was concentrated toward cardiovascular and kidney pathways, with less support directed toward metabolic pathways, suggesting that hypertension primarily functions as an upstream driver of cardiac and renal injury while simultaneously representing a downstream consequence of metabolic dysfunction. Functional-role analyses classified hypertension as a disease driver, distinct from blood pressure, which was clustered as a biomarker.
CONCLUSIONS: CKM-KG provides a systematic framework for translating complex biomedical evidence into actionable HEOR insights. Using hypertension as a proof-of-concept, CKM-KG recovered clinically recognized disease relationships and generated mechanistic insights into cross-system disease progression. KG approach may support development of CKM economic models by informing model structure, outcome selection, disease interactions, and evidence-gap.