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

Healthcare representation learning on the Electronic Health Records is crucial for downstream medical prediction tasks in health informatics. Many NLP techniques, such as RNN and self-attention, have been adapted to learn medical representations from hierarchical and time-stamped EHRs data, but fail when they lack either general or task-specific data. Hence, some recent works train healthcare representations by incorporating medical ontology, by self-supervised tasks like diagnosis prediction, but (1) the small-scale, monotonous ontology is insufficient for robust learning, and (2) critical contexts or dependencies underlying patient journeys are barely exploited to enhance ontology learning. To address the challenges, we propose a Transformer-based representation learning approach: Mutual Integration of Patient journey and medical Ontology (MIPO), which is a robust end-to-end framework. Specifically, the proposed method focuses on task-specific representation learning by a sequential diagnoses predictive task, which is also beneficial to the ontology-based disease typing task. To integrate information in the patient's visiting records, we further introduce a graph-embedding module, which can mitigate the challenge of data insufficiency in healthcare. In this way, MIPO creates a mutual integration to benefit both healthcare representation learning and medical ontology embedding. Such an effective integration is guaranteed by joint training over fused embeddings of the two modules, targeting both task-specific prediction and ontology-based disease typing tasks simultaneously. Extensive experiments conducted on two real-world benchmark datasets have shown MIPO consistently achieves better performance than state-of-the-art methods no matter whether the training data is sufficient or not. Also, MIPO derives more interpretable diagnose embedding results compared to its counterparts.

PDF Abstract

Datasets


Results from the Paper


  Submit results from this paper to get state-of-the-art GitHub badges and help the community compare results to other papers.

Methods


No methods listed for this paper. Add relevant methods here