Representation learning to advance multi-institutional studies with electronic health record data from US and France

2026-04-03 · Source: Machine learning : nature.com subject feeds · Field: Health & Wellbeing — Health & Medical Research, Medical Devices & Health Technology, Healthcare Systems & Policy · Depth: Expert, long

Summary

A new graph-based framework addresses the challenge of harmonizing fragmented and heterogeneously coded electronic health record (EHR) data across privacy-siloed institutions. This framework treats data harmonization as a scalable representation learning problem, integrating institution-specific summary statistics from health records, curated biomedical knowledge graphs, and semantic information derived from large language models. The joint learning approach creates a shared semantic space, aligning diverse, site-specific vocabularies while preserving patient privacy. Evaluated across seven institutions and two languages, this framework provides a robust, data-centric foundation for training and deploying clinical models across disparate healthcare systems. The pairwise cosine similarity data derived from the GAME embeddings are publicly available, while institution-level summary data require restricted access via data use agreements.

Key takeaway

For AI Scientists and Machine Learning Engineers developing clinical models across multiple healthcare systems, this framework offers a solution to data heterogeneity and privacy concerns. You should explore integrating this graph-based representation learning approach to harmonize diverse EHR vocabularies, enabling more robust model training and deployment. Consider leveraging the publicly available GAME embeddings for initial investigations into cross-institutional data alignment.

Key insights

A graph-based framework harmonizes heterogeneous EHR data across institutions using representation learning and privacy-preserving methods.

Principles

• Integrate diverse data sources for semantic alignment.
• Preserve patient privacy through collaborative learning.
• Treat data harmonization as a representation learning task.

Method

The framework integrates institution-specific EHR summary statistics, biomedical knowledge graphs, and large language model semantic information to learn a shared semantic space, aligning diverse vocabularies without sharing patient-level data.

In practice

• Access GAME embeddings for public cosine similarity data.
• Request DUAs for institution-level summary data access.

Topics

• Representation Learning
• Electronic Health Records
• Data Harmonization
• Privacy-Preserving Learning
• Biomedical Knowledge Graphs

Code references

• celehs/GAME

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Knowledge-Embedded Latent Projection for Robust Representation Learning — Leveraging semantic side information improves latent space model estimation in high-dimensional, imbalanced data.
• Comparing decentralized machine learning and AI clinical models to local and centralized alternatives: a systematic review — Decentralized learning offers clinically acceptable performance in healthcare, especially for privacy-constrained contexts, outperforming local learning.
• Statistical Embeddings for Similarity, Retrieval, and Interpretable Alignment of Numeric Tabular Datasets — Statistical embeddings enable interpretable similarity and alignment of heterogeneous numeric tabular datasets for LLMs, preserving statistical context.
• Social-JEPA: Emergent Geometric Isomorphism — Independently trained JEPA world models spontaneously develop linearly alignable latent spaces, enabling efficient knowledge transfer.
• Learning Emergent Modular Representations in Multi-modality Medical Vision Foundation Models — Regulating specialization and coordination in modular networks overcomes Non-IID challenges in multi-modality medical vision foundation models.
• ChatHealthAI: Aligning Electronic Health Record Representations with Large Language Models for Grounded Clinical Reasoning — ChatHealthAI integrates EHR representations with LLMs for grounded, interpretable clinical reasoning and accurate patient prediction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.