Not What the Doctor Ordered: Surveying LLM-based De-identification and Quantifying Clinical Information Loss

2025-09-17

AI Summary

This study identifies three critical issues in current LLM-driven medical text de-identification: (1) lack of standardized evaluation metrics; (2) inadequacy of conventional classification metrics (e.g., F1-score) in detecting clinically harmful over-redaction—such as erroneous removal of essential diagnostic or medication information; and (3) absence of human expert validation for novel error types. To address these, we propose a hybrid evaluation framework integrating automated error detection with clinical expert review, systematically benchmarking leading LLMs on standard medical de-identification datasets. We introduce a fine-grained, clinically grounded error annotation schema and quantitative metrics explicitly designed to assess clinical information preservation. Results reveal pervasive over-redaction across state-of-the-art methods, with conventional automatic metrics substantially overestimating performance. Clinical expert validation confirms severe limitations in real-world clinical utility. This work establishes a new evaluation paradigm for medical de-identification that rigorously balances privacy protection and clinical usability.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-in-the-loop Machine Learning

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

De-identification in the healthcare setting is an application of NLP where automated algorithms are used to remove personally identifying information of patients (and, sometimes, providers). With the recent rise of generative large language models (LLMs), there has been a corresponding rise in the number of papers that apply LLMs to de-identification. Although these approaches often report near-perfect results, significant challenges concerning reproducibility and utility of the research papers persist. This paper identifies three key limitations in the current literature: inconsistent reporting metrics hindering direct comparisons, the inadequacy of traditional classification metrics in capturing errors which LLMs may be more prone to (i.e., altering clinically relevant information), and lack of manual validation of automated metrics which aim to quantify these errors. To address these issues, we first present a survey of LLM-based de-identification research, highlighting the heterogeneity in reporting standards. Second, we evaluated a diverse set of models to quantify the extent of inappropriate removal of clinical information. Next, we conduct a manual validation of an existing evaluation metric to measure the removal of clinical information, employing clinical experts to assess their efficacy. We highlight poor performance and describe the inherent limitations of such metrics in identifying clinically significant changes. Lastly, we propose a novel methodology for the detection of clinically relevant information removal.

Problem

• Evaluating LLM-based de-identification's clinical information loss
• Addressing inconsistent metrics and reproducibility challenges
• Quantifying inappropriate removal of clinically relevant data

Innovation

• Surveyed LLM-based de-identification research heterogeneity
• Conducted manual validation with clinical experts
• Proposed novel methodology detecting clinical information removal