From Notes to Nodes: Patient-Centric Knowledge Graph for Targeted Glaucoma Care

Unstructured clinical notes, often composed of free-text descriptions by healthcare providers, are rich in detailed patient information but lack a consistent format, making them challenging to analyze. Integrating Knowledge graphs (KG) into healthcare provides significant advantages, especially in managing chronic conditions like glaucoma. The treatment decisions for such cases depend on various clinical factors. KGs are an essential tool for modeling complex healthcare relationships. It provides a structured and interpretable framework for understanding medical concepts and their connections. The graph structure enhances NLP techniques, efficiently extracting critical information from unstructured clinical data. This paper presents the Patient-Centric Knowledge Graph (PaCKG). It is constructed from unstructured clinical notes focusing on glaucoma patients. Our method, PaCKG, aims to help clinicians make medical decisions by mapping relationships between patients, their symptoms, and corresponding treatments. The PaCKG follows three steps. First, preprocessing the clinical notes; second, Extracting relevant information using Named Entity Recognition (NER); and finally, Constructing the KG, which includes three main node types—patients, symptoms, and treatments (drugs). By structuring unstructured clinical notes, PaCKG enables efficient retrieval and relational mapping of critical healthcare information. The key metrics analyzed include the total number of nodes, symptoms, treatments, and the total number of edges and relationships. Further insights are derived from the average degree, degree distribution, and the number of weakly connected components, which provide a comprehensive view of the graph’s structure. Performance is demonstrated through accurate extraction of symptoms and treatments, along with high connectivity and appropriate density of links, effectively linking patients to their symptoms and treatments in the knowledge graph.