Concurrent Clinical Predictions via Dual-Space Message Passing and Ontology Alignment
Year: 2026
Type: conference-paper
Abstract: The growing availability of electronic health records (EHR) presents increasing opportunities for data-driven applications in healthcare. These records, however, are represented using disparate coding systems and often suffer from fragmentation and structural inconsistencies. Such data fragmentation can lead to substantial performance gaps across different predictive tasks. In this work, a framework is proposed in order to mitigate these artefacts and enable multi-task predictions. To address performance gaps, we introduce a concurrent loss that guides the learning process over fragmented entities. To address structural misalignment, we propose a neural message-passing algorithm that enables the integration of ontological information. Experiments conducted on cardiovascular disease (CVD) patients demonstrate that the proposed system overcomes prediction gaps across multiple clinical tasks, including laboratory tests, diagnoses, medications, and mortality prediction, performed concurrently. The results highlight the potential of ontology-aligned representation learning for high-precision decision-making in healthcare systems.
Authors: Shervin Mehryar, Michel Dumontier
Institution: Maastricht University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7222949227
