MedFuse: Multiplicative Embedding Fusion For Irregular Clinical Time Series

2025-11-12

AI Summary

Clinical time series—such as electronic health record (EHR) data—exhibit irregular sampling, high missingness, and dynamic heterogeneity across features, rendering conventional additive embedding methods inadequate for capturing value-dependent, cross-feature interactions. To address this, we propose MuFuse, a multiplicative modulation-based embedding fusion framework that explicitly models higher-order feature dependencies via element-wise multiplication of feature-identity and value embeddings, enabling nonlinear coupling. This design enhances representational capacity, supports cross-dataset pretraining, and seamlessly integrates with mainstream irregular-time-series modeling architectures. Extensive experiments on three real-world clinical datasets demonstrate that MuFuse significantly outperforms state-of-the-art methods in both intensive care prediction and chronic disease forecasting tasks, validating the effectiveness and generalizability of multiplicative embedding fusion for clinical time-series modeling.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization

Abstract

Clinical time series derived from electronic health records (EHRs) are inherently irregular, with asynchronous sampling, missing values, and heterogeneous feature dynamics. While numerical laboratory measurements are highly informative, existing embedding strategies usually combine feature identity and value embeddings through additive operations, which constrains their ability to capture value-dependent feature interactions. We propose MedFuse, a framework for irregular clinical time series centered on the MuFuse (Multiplicative Embedding Fusion) module. MuFuse fuses value and feature embeddings through multiplicative modulation, preserving feature-specific information while modeling higher-order dependencies across features. Experiments on three real-world datasets covering both intensive and chronic care show that MedFuse consistently outperforms state-of-the-art baselines on key predictive tasks. Analysis of the learned representations further demonstrates that multiplicative fusion enhances expressiveness and supports cross-dataset pretraining. These results establish MedFuse as a generalizable approach for modeling irregular clinical time series.

Problem

• Handles irregular EHR time series with missing values and asynchronicity
• Models value-dependent feature interactions through multiplicative embedding fusion
• Improves predictive performance on clinical tasks across care settings

Innovation

• Multiplicative fusion replaces additive embedding operations
• Models higher-order dependencies across clinical features
• Handles irregular sampling and missing EHR values