BEACON-SP Uses Ontology-Grounded GraphRAG for Clinical Suicide Risk Assessment

Summary

BEACON-SP is an ontology-grounded GraphRAG framework designed to support clinicians assessing suicide risk in behavioral health settings. It combines patient knowledge graphs with ontology-guided retrieval so the system can connect diagnoses, medications, risk and protective factors, life events, and temporal relationships through multi-hop reasoning. The framework uses a unified suicide prevention ontology that brings together the Three-Step Theory, the Integrated Motivational-Volitional Model, and the Suicide Social Determinants of Health Ontology. In an evaluation with 1,500 queries covering 15 clinical categories and 100 patients, BEACON-SP was compared with a vector-based RAG baseline. Under a corrected comparative evaluation protocol, it improved completeness, clinical relevance, and evidence grounding, while producing only a small gain in factual accuracy. In paired criterion-level comparisons, the GraphRAG system was preferred in 76.4% of cases. The findings indicate that structured ontological representations can help organize contextual patient evidence for clinician-facing decision support, although the abstract does not report deployment outcomes or clinical validation beyond the benchmark.