Next Visit Diagnosis Prediction via Medical Code-Centric Multimodal Contrastive EHR Modelling with Hierarchical Regularisation
Heejoon Koo
TL;DR
This paper addresses predicting a patient’s next-visit diagnosis from heterogeneous EHR data by introducing NECHO, a medical code-centric multimodal framework that fuses medical codes, demographics, and clinical notes through cross-modal transformers and a multimodal adaptation gate. It adds two asymmetric bimodal contrastive losses anchored on the medical-code representation and employs hierarchical regularisation via parental ICD-9 codes to regularise each modality-specific encoder. The approach achieves state-of-the-art performance on MIMIC-III for next-visit diagnosis prediction and is supported by extensive ablations and a case study showing qualitative improvements. The framework demonstrates the value of prioritising medical-code representations and leveraging ICD-9 hierarchy to improve robustness and generalisation in multimodal EHR modelling.
Abstract
Predicting next visit diagnosis using Electronic Health Records (EHR) is an essential task in healthcare, critical for devising proactive future plans for both healthcare providers and patients. Nonetheless, many preceding studies have not sufficiently addressed the heterogeneous and hierarchical characteristics inherent in EHR data, inevitably leading to sub-optimal performance. To this end, we propose NECHO, a novel medical code-centric multimodal contrastive EHR learning framework with hierarchical regularisation. First, we integrate multifaceted information encompassing medical codes, demographics, and clinical notes using a tailored network design and a pair of bimodal contrastive losses, all of which pivot around a medical codes representation. We also regularise modality-specific encoders using a parental level information in medical ontology to learn hierarchical structure of EHR data. A series of experiments on MIMIC-III data demonstrates effectiveness of our approach.