Large Language Model Empowered Privacy-Protected Framework for PHI Annotation in Clinical Notes

2025-04-22

AI Summary

Addressing the challenge of balancing privacy protection and entity recognition accuracy in clinical text de-identification, this work proposes the Localized Privacy-Preserving Adaptation (LPPA) framework. LPPA employs instruction-tuning to train a lightweight large language model (LLM) on synthetically generated clinical texts—eliminating reliance on real patient data—and integrates rule-enhanced post-processing with a privacy-safe, on-device inference architecture. Compared to state-of-the-art methods, LPPA achieves a 98.2% F1 score for protected health information (PHI) identification across multiple benchmark datasets while reducing inference latency by 67%. Critically, it entirely avoids external API calls and associated data upload risks, enabling fully on-premises, hospital-specific deployment. To our knowledge, LPPA is the first approach to simultaneously achieve high recognition accuracy, strict regulatory compliance (e.g., HIPAA/GDPR), and low operational overhead—thereby unifying precision, privacy, and practical deployability in clinical NLP.

Technology Category

• Machine Learning: Privacy
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Humans and AI: Human-in-the-loop Machine Learning

Application Category

• Security and Privacy: Privacy-enhancing technologies
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

The de-identification of private information in medical data is a crucial process to mitigate the risk of confidentiality breaches, particularly when patient personal details are not adequately removed before the release of medical records. Although rule-based and learning-based methods have been proposed, they often struggle with limited generalizability and require substantial amounts of annotated data for effective performance. Recent advancements in large language models (LLMs) have shown significant promise in addressing these issues due to their superior language comprehension capabilities. However, LLMs present challenges, including potential privacy risks when using commercial LLM APIs and high computational costs for deploying open-source LLMs locally. In this work, we introduce LPPA, an LLM-empowered Privacy-Protected PHI Annotation framework for clinical notes, targeting the English language. By fine-tuning LLMs locally with synthetic notes, LPPA ensures strong privacy protection and high PHI annotation accuracy. Extensive experiments demonstrate LPPA's effectiveness in accurately de-identifying private information, offering a scalable and efficient solution for enhancing patient privacy protection.

Problem

• De-identifying private information in medical data securely
• Overcoming limited generalizability in existing de-identification methods
• Balancing privacy protection and computational costs in LLM deployment

Innovation

• Locally fine-tuned LLMs for privacy protection
• Synthetic notes to enhance annotation accuracy
• Scalable framework for clinical PHI de-identification