HCOE: Hyperbolic Clinical Ontology Embeddings from Biomedical Language Models

2026-09-24

AI Summary

This study addresses the limitation of existing biomedical language models in explicitly preserving the hierarchical structure of medical codes. To this end, we propose HCOE, a framework that maps frozen BioBERT embeddings onto the Poincaré ball, leveraging hyperbolic geometry to naturally model tree-like ontological relationships. Methodologically, HCOE introduces dual-ended contrastive learning and a coarse-to-fine ontology path aggregation mechanism, deeply integrating ICD and ATC hierarchical information for hierarchy-aware representation learning. Experiments demonstrate that HCOE achieves state-of-the-art performance across clinical relation prediction, hierarchical transfer, and multiple MIMIC-IV tasks. These results confirm that the proposed approach effectively enhances the structured representational capacity of medical code embeddings.

Abstract

Biomedical language models (LMs) encode textual semantics but do not explicitly preserve medical code hierarchies. We present Hyperbolic Clinical Ontology Embeddings (HCOE) for hierarchy-aware clinical concept representation. HCOE maps frozen BioBERT embeddings into a Poincare ball, combining parent-side and child-side ontology-guided contrastive learning with coarse-to-fine ontology-path aggregation. It uses International Classification of Diseases (ICD) codes organized by Clinical Classifications Software (CCS) and Anatomical Therapeutic Chemical (ATC) medication hierarchies. Evaluations show that HCOE performs best on ICD/ATC clinical relation prediction and CCS-to-PheCode hierarchy transfer. On the MIMIC-IV dataset, HCOE also achieves the best performance on mortality prediction, readmission prediction, medication recommendation, and rare drug prediction.

Problem

• Biomedical Language Models
• Clinical Ontology
• Hierarchy-aware Representation
• Medical Code Hierarchies

Innovation

• Hyperbolic Embeddings
• Ontology-guided Contrastive Learning
• Clinical Ontology
• Biomedical Language Models
• Poincare Ball