Prediction of 30-day hospital readmission with clinical notes and EHR information

2025-03-29

AI Summary

This study addresses the problem of predicting 30-day hospital readmission risk for inpatients. We propose a multimodal graph neural network (GNN) framework that, for the first time, jointly models large language model (LLM)-encoded clinical notes and heterogeneous electronic health record (EHR) fields as nodes in a heterogeneous graph—enabling cross-modal semantic alignment and joint reasoning between structured and unstructured data. The method integrates LLM-based text encoding, multi-source feature embedding, and GNN-based message passing. Evaluated on a real-world EHR dataset, our model achieves an AUROC of 0.72 and a balanced accuracy of 66.7%, significantly outperforming unimodal baselines. Our key contribution is the introduction of the first heterogeneous graph learning paradigm specifically designed for synergistic modeling of EHRs and clinical notes, offering a novel, interpretable, and scalable approach to multimodal clinical risk prediction.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

High hospital readmission rates are associated with significant costs and health risks for patients. Therefore, it is critical to develop predictive models that can support clinicians to determine whether or not a patient will return to the hospital in a relatively short period of time (e.g, 30-days). Nowadays, it is possible to collect both structured (electronic health records - EHR) and unstructured information (clinical notes) about a patient hospital event, all potentially containing relevant information for a predictive model. However, their integration is challenging. In this work we explore the combination of clinical notes and EHRs to predict 30-day hospital readmissions. We address the representation of the various types of information available in the EHR data, as well as exploring LLMs to characterize the clinical notes. We collect both information sources as the nodes of a graph neural network (GNN). Our model achieves an AUROC of 0.72 and a balanced accuracy of 66.7%, highlighting the importance of combining the multimodal information.

Problem

• Predict 30-day hospital readmissions using clinical notes and EHR data
• Integrate structured EHR and unstructured clinical notes for prediction
• Develop a GNN model combining multimodal data for accurate forecasting

Innovation

• Combines clinical notes and EHR data
• Uses LLMs to characterize clinical notes
• Integrates data via graph neural network