Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Guanchen Wu, Linzhi Zheng, Han Xie, Zhen Xiang, Jiaying Lu, Darren Liu, Delgersuren Bold, Bo Li, Xiao Hu, Carl Yang

TL;DR

LPPA tackles PHI de-identification in clinical notes by combining synthetic data generation with instruction-tuning of a locally hosted LLM. It introduces two synthetic-data pipelines (AEG and SPI) and a hybrid mixing strategy, followed by instruction-tuning with LoRA to produce a privacy-preserving PHI tagger that operates locally. Empirical results show competitive performance on real clinical notes and strong gains on synthetic data, with significant reductions in required annotated data and no reliance on external APIs. The framework offers scalable privacy-preserving de-identification with potential applicability to other privacy-sensitive domains.

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.

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

TL;DR

LPPA tackles PHI de-identification in clinical notes by combining synthetic data generation with instruction-tuning of a locally hosted LLM. It introduces two synthetic-data pipelines (AEG and SPI) and a hybrid mixing strategy, followed by instruction-tuning with LoRA to produce a privacy-preserving PHI tagger that operates locally. Empirical results show competitive performance on real clinical notes and strong gains on synthetic data, with significant reductions in required annotated data and no reliance on external APIs. The framework offers scalable privacy-preserving de-identification with potential applicability to other privacy-sensitive domains.

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.

Paper Structure

This paper contains 20 sections, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Problem Formulation
  5. Framework Overview
  6. LLM-based Synthetic Data Generation
  7. Instruction Tuning
  8. Experiments
  9. Real-World Datasets
  10. Experiment Setup
  11. Experimental Implementation
  12. Results
  13. Synthetic Data Evaluation
  14. Case Studies
  15. Computational and Financial Costs
  16. ...and 5 more sections

Figures (6)

  • Figure 1: Illustration of existing de-identification methods and our proposed de-identification framework.
  • Figure 2: LPPA Framework Overview. This framework leverages LLMs to generate synthetic training data using two distinct approaches, which are then combined with a task-oriented prompt to fine-tune a base model. The fine-tuned model processes real clinical notes to output a PHI dictionary identifying sensitive information. A subsequent programming technique is applied to the identified PHI to generate de-identified clinical notes, ensuring privacy while preserving the note's structure.
  • Figure 3: Case Study
  • Figure 4: Cost Estimation
  • Figure 5: GPU Utilization During Fine-Tuning
  • ...and 1 more figures