UdonCare: Hierarchy Pruning for Unseen Domain Discovery in Predictive Healthcare

2025-06-08

AI Summary

Clinical prediction models suffer from degraded generalization due to patient population distribution shifts across healthcare sites, while existing domain generalization (DG) methods are hindered by the absence of explicit domain labels and lack of clinical knowledge integration. To address this, we propose a medical knowledge–driven unsupervised domain discovery framework. Our approach introduces, for the first time, an iterative pruning mechanism grounded in the ICD-9-CM hierarchical ontology to enable flexible, clinically semantically aligned domain partitioning. It jointly employs an ontology-guided domain encoder and a Siamese-style disentangled inference architecture to synergistically model clinical priors and data-driven features. Extensive experiments on MIMIC-III and MIMIC-IV demonstrate that our method significantly outperforms state-of-the-art DG baselines, achieving up to a 3.2% AUC improvement under large domain shifts. This work provides the first empirical validation that structured medical ontologies enhance both robustness and interpretability in healthcare DG.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Knowledge Representation and Reasoning: Ontologies
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Domain generalization has become a critical challenge in clinical prediction, where patient cohorts often exhibit shifting data distributions that degrade model performance. Typical domain generalization approaches struggle in real-world healthcare settings for two main reasons: (1) patient-specific domain labels are typically unavailable, making domain discovery especially difficult; (2) purely data-driven approaches overlook key clinical insights, leading to a gap in medical knowledge integration. To address these problems, we leverage hierarchical medical ontologies like the ICD-9-CM hierarchy to group diseases into higher-level categories and discover more flexible latent domains. In this paper, we introduce UdonCare, a hierarchy-guided framework that iteratively prunes fine-grained domains, encodes these refined domains, and applies a Siamese-type inference mechanism to separate domain-related signals from patient-level features. Experimental results on clinical datasets (MIMIC-III and MIMIC-IV) show that the proposed model achieves higher performance compared to other domain generalization baselines when substantial domain gaps presents, highlighting the untapped potential of medical knowledge for enhancing domain generalization in practical healthcare applications.

Problem

• Addressing domain generalization challenges in clinical prediction models
• Overcoming lack of patient-specific domain labels in healthcare data
• Integrating medical ontologies to improve domain discovery and model performance

Innovation

• Leverages hierarchical medical ontologies for domain discovery
• Prunes fine-grained domains iteratively for refinement
• Uses Siamese-type inference to separate domain signals