Post-discharge recovery management requires continuous representation of patient conditions, temporal observations, and clinical interventions after hospital discharge. This study proposes a graph-augmented case retrieval framework for post-discharge decision support. The framework models each patient as a longitudinal recovery case in a knowledge graph, where the discharge state serves as the baseline and subsequent monitoring records are represented as time-indexed snapshots. A patient-centered ontology is developed to integrate diagnoses, medications, symptoms, abnormalities, behavior factors, lifestyle recommendations, and recovery status. Large language models are used to support schema-conformant case construction, case update, and Cypher query generation from semi-structured discharge summaries and patient recovery data. During follow-up encounters, similar historical cases are retrieved through a structured similarity process that combines baseline profile similarity, clinical similarity, recovery trajectory similarity, and observation similarity. A hypertension recovery case study is presented to demonstrate graph-based case construction, three-day snapshot modeling, similar case retrieval, and decision-support output generation. The results illustrate how graph representation can support longitudinal recovery modeling and retrieval-augmented post-discharge decision support.

Keywords

Post-discharge decision support, patient recovery journey, knowledge graph, large language model, case retrieval