Mutual Integration of Patient Journey and Medical Ontology for Healthcare Representation Learning

View PDF of the article titled MIPO: Mutual Integration of the Patient Journey and Medical Ontologies for Learning Healthcare Representation, by Xueping Peng, Guodong Long, Tao Shen, Sen Wang, Chengqi Zhang, Allison Clarke, and Clement Schlegel

View PDF HTML (beta)

a summary:Representation learning in electronic health records (EHRs) plays a vital role in downstream medical prediction tasks. Although natural language processing techniques, such as recurrent neural networks and self-attention, have been adapted to learn medical representations from time-stamped, hierarchical electronic health record data, they often face difficulty when general or task-specific data are limited. Recent efforts have attempted to mitigate this challenge by incorporating medical ontologies (i.e., knowledge graphs) into self-supervised tasks such as diagnosis prediction. However, two major issues remain: (1) small and uniform ontologies that lack diversity for robust learning, and (2) insufficient attention to critical contexts or dependencies underlying patient journeys, which can further enhance ontology-based learning. To address these gaps, we propose MIPO (Mutual Integration of Patient Journey and Medical Ontology), a robust, comprehensive framework that uses a transformer-based architecture for representation learning. MIPO emphasizes task-specific representation learning through a sequential diagnosis prediction task, while also incorporating an ontology-based disease typing task. A graph embedding module was introduced to integrate information from patient visit records, thus alleviating data deficiency. This setup creates a mutually reinforcing loop, where both including the patient journey and including ontology benefit from each other. We validate MIPO on two real-world benchmark datasets, showing that it consistently outperforms baseline methods under conditions of sufficient and limited data. Furthermore, the resulting diagnostic embeddings provide improved interpretability, confirming the promise of MIPO for real-world healthcare applications.