LLM-augmented Knowledge Graphs for Cross-system Patient Record Reconciliation

This article presents a novel approach to addressing patient record fragmentation across healthcare systems by integrating large language models (LLMs) with knowledge graphs. The fragmentation of health information presents significant challenges for care coordination, clinical decision-making, and data-driven healthcare initiatives. The hybrid architecture leverages the contextual reasoning capabilities of LLMs to interpret and reconcile semantically similar but structurally diverse clinical data elements across disparate electronic health record systems. The knowledge graph component provides a structured, queryable representation of patient information that preserves complex clinical relationships while supporting sophisticated entity resolution. This article moves beyond traditional demographic matching methods by incorporating clinical content awareness into the reconciliation process, enabling more accurate identification of patient records across institutional boundaries. Evaluation across multiple healthcare environments demonstrates substantial improvements in matching accuracy compared to conventional approaches, particularly for patients with complex medical histories. The article shows promise in reducing duplicate records, enhancing interoperability, and supporting unified care delivery through comprehensive longitudinal patient views. While implementation challenges remain, including privacy considerations and computational requirements, this article establishes a foundation for more intelligent and effective health information exchange that could significantly improve continuity of care and clinical workflow efficiency.