Term2Note: Synthesising Differentially Private Clinical Notes from Medical Terms

2025-09-13

AI Summary

In healthcare, real-world clinical texts are rarely accessible for model training due to stringent privacy regulations. To address this, we propose a differential privacy (DP)-enabled synthetic data generation framework tailored for long-form clinical records. Our method decouples content—comprising medical terminology—from structural elements—such as section-level organization—and applies independent DP constraints to each component. We further introduce a staged generation strategy coupled with a DP-aware quality selector to jointly optimize utility and privacy, guaranteeing strict ε-differential privacy. Experimental results demonstrate that the synthetic clinical notes closely approximate real data in both statistical distributions and semantic fidelity. When used to train multi-label classification models, they achieve 96.2% of the performance attained using real clinical notes—substantially outperforming existing DP-compliant text generation approaches.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Privacy
• Computer Vision: Diffusion Models for Vision

Application Category

• Security and Privacy: Data transparency and provenance
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems

Abstract

Training data is fundamental to the success of modern machine learning models, yet in high-stakes domains such as healthcare, the use of real-world training data is severely constrained by concerns over privacy leakage. A promising solution to this challenge is the use of differentially private (DP) synthetic data, which offers formal privacy guarantees while maintaining data utility. However, striking the right balance between privacy protection and utility remains challenging in clinical note synthesis, given its domain specificity and the complexity of long-form text generation. In this paper, we present Term2Note, a methodology to synthesise long clinical notes under strong DP constraints. By structurally separating content and form, Term2Note generates section-wise note content conditioned on DP medical terms, with each governed by separate DP constraints. A DP quality maximiser further enhances synthetic notes by selecting high-quality outputs. Experimental results show that Term2Note produces synthetic notes with statistical properties closely aligned with real clinical notes, demonstrating strong fidelity. In addition, multi-label classification models trained on these synthetic notes perform comparably to those trained on real data, confirming their high utility. Compared to existing DP text generation baselines, Term2Note achieves substantial improvements in both fidelity and utility while operating under fewer assumptions, suggesting its potential as a viable privacy-preserving alternative to using sensitive clinical notes.

Problem

• Synthesizing clinical notes with differential privacy
• Balancing privacy and utility in medical text generation
• Generating realistic synthetic healthcare data securely

Innovation

• DP synthetic clinical notes generation
• Separates content and form structurally
• DP quality maximiser enhances output