![[2506.06977] Discovering Hierarchy-Grounded Domains with Adaptive Granularity for Clinical Domain Generalization](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2506.06977] Discovering Hierarchy-Grounded Domains with Adaptive Granularity for Clinical Domain Generalization
Summary
The paper presents UdonCare, a novel method for domain generalization in healthcare that utilizes medical ontologies to enhance predictive model performance across varying patient groups.
Why It Matters
This research addresses significant challenges in predictive healthcare, where traditional domain generalization methods often fail due to the lack of domain labels and clinical insights. By leveraging medical ontologies, this study offers a promising approach to improve model robustness and accuracy in clinical settings, which can ultimately lead to better patient outcomes.
Key Takeaways
- UdonCare introduces a hierarchy-pruning method for dynamic patient domain discovery.
- The method outperforms eight existing baselines in clinical prediction tasks.
- Utilizing medical ontologies can significantly enhance model generalization.
- The research highlights the importance of integrating clinical insights into machine learning.
- Addressing domain shifts in healthcare is crucial for predictive model effectiveness.
Computer Science > Machine Learning arXiv:2506.06977 (cs) [Submitted on 8 Jun 2025 (v1), last revised 13 Feb 2026 (this version, v3)] Title:Discovering Hierarchy-Grounded Domains with Adaptive Granularity for Clinical Domain Generalization Authors:Pengfei Hu, Xiaoxue Han, Fei Wang, Yue Ning View a PDF of the paper titled Discovering Hierarchy-Grounded Domains with Adaptive Granularity for Clinical Domain Generalization, by Pengfei Hu and 3 other authors View PDF HTML (experimental) Abstract:Domain generalization has become a critical challenge in predictive healthcare, where different patient groups often exhibit shifting data distributions that degrade model performance. Still, regular domain generalization approaches often struggle in clinical settings due to (1) the absence of domain labels and (2) the lack of clinical insight integration. To address these challenges in healthcare, we aim to explore how medical ontologies can be used to discover dynamic yet hierarchy-grounded patient domains, a partitioning strategy that remains under-explored in prior work. Hence, we introduce UdonCare, a hierarchy-pruning method that iteratively divides patients into latent domains and retrieve domain-invariant (label) information from patient data. On two public datasets, UdonCare shows superiority over eight baselines across four representative clinical prediction tasks with substantial domain gaps, highlighting the potential of medical knowledge for enhancing model generalization. ...