End-to-end pipeline for automated heart failure diagnosis with clinical notes using SNOMED-CT

2026-04-19 · Source: Machine learning : nature.com subject feeds · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Clinical Natural Language Processing · Depth: Expert, extended

Summary

A novel end-to-end pipeline has been developed for automated heart failure diagnosis, utilizing electronic health records (EHR) and German clinical notes from 846 patients. The pipeline integrates abbreviation disambiguation, translation of German notes to English, and medical entity linking to SNOMED-CT, followed by classification. It employs zero-shot learning for disambiguation and entity linking, reducing reliance on extensive training data. The system achieved an abbreviation disambiguation accuracy of up to 96.1% and competitive entity linking performance. For heart failure classification, an SVM approach using SNOMED-CT concepts and EHR data yielded an F1-score of 65.3%, matching a fine-tuned medBERT.de neural baseline. This pipeline demonstrates high potential for real-world clinical use and decision support, particularly in environments with limited language-specific resources.

Key takeaway

For NLP Engineers developing clinical decision support systems, this pipeline offers a robust framework for handling multilingual, unstructured clinical data. You should consider integrating zero-shot learning for abbreviation disambiguation and entity linking to SNOMED-CT, especially when working with resource-scarce languages like German. This approach can achieve diagnostic accuracy comparable to neural baselines while providing greater interpretability and adaptability to varying document lengths, crucial for real-world clinical deployment.

Key insights

An end-to-end pipeline automates heart failure diagnosis from German clinical notes using zero-shot learning and SNOMED-CT.

Principles

• Zero-shot learning reduces training data dependency.
• Standardized terminologies enhance clinical data utility.
• Combining structured and unstructured data improves diagnostic accuracy.

Method

The pipeline involves four steps: abbreviation disambiguation, German-to-English translation, semantic zero-shot entity linking to SNOMED-CT/UMLS, and SVM-based heart failure classification using SNOMED-CT concepts and EHR data.

In practice

• Use BioLORD for English and paraphrase-multilingual-MiniLM-L12-v2 for German semantic similarity tasks.
• Employ TF-IDF weighting for SNOMED-CT concepts in classification.
• Integrate online learning for iterative performance improvements based on physician feedback.

Topics

• Heart Failure Diagnosis
• SNOMED-CT
• Medical Entity Linking
• Zero-Shot Learning
• Clinical Note Processing

Code references

• chanzuckerberg/MedMentions

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

Related on AIssential

• Language Models as Semantic Teachers: Post-Training Alignment for Medical Audio Understanding — Aligning audio encoders with medical LLMs via a "semantic teacher" paradigm significantly enhances clinical diagnostic performance.
• Intelligent Healthcare Imaging Platform: A VLM-Based Framework for Automated Medical Image Analysis and Clinical Report Generation — A VLM-based framework automates medical image analysis and clinical report generation across multiple imaging modalities.
• Using natural language processing to identify emergency department patients with incidental lung nodules requiring follow-up — An NLP pipeline using RoBERTa and spaCy can automate incidental lung nodule follow-up identification from ED CT reports.
• Learning Directional Semantic Transitions for Longitudinal Chest X-ray Analysis — ProTrans models disease progression in CXRs as directional semantic transitions using vision-language pretraining and radiology reports.
• A Multimodal and Explainable Machine Learning Approach to Diagnosing Multi-Class Ejection Fraction from Electrocardiograms — Multimodal machine learning can accurately stratify LVEF from ECG and EHR data, enhancing accessibility.
• CA-GCL: Cross-Anatomy Global-Local Contrastive Learning for Robust 3D Medical Image Understanding — CA-GCL uses global-local contrastive learning and clinical-aware text augmentation to prevent representation collapse in 3D medical FVLP.

Open in AIssential →

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