Construction of personal health knowledge graphs for clinical data harmonization in breast cancer

Integrating heterogeneous personal health data and maintaining clinical registries for breast cancer management is labor-intensive. This study presents an AI-powered virtual assistant transforming identifiable breast cancer patient data into interoperable personal health knowledge graphs for secondary use. The first-generation prototype was deployed across three hospitals to extract and ingest patient data under informed consent, mapped to ontology-compliant knowledge graphs, with SHACL-based structural validation and human-in-the-loop review. Curated data covered demographics, diagnosis, pathology, laboratory results, surgery, therapy, and follow-up, from which local SPARQL queries generated per-site standardized registries. Ingestion and curation results were completed at two sites: ingestion success of 51–94%, curation success of 59–64% and automated curation of 41–47%. Overall, SHACL-based structural data quality was 94.5–96.1% (validity: >98%; completeness: 76–81%). These results demonstrate the feasibility of semi-automated health data harmonization and knowledge graph construction, representing an initial step toward scalable, interoperable registry infrastructure in breast cancer.