LLMs outperform traditional systems in de-identifying sensitive health data

Researchers have developed a new method using large language models (LLMs) to identify and recover protected health information (PHI) that traditional de-identification systems often miss. By employing institution-specific prompts, LLMs can better understand context-dependent PHI, such as hospital abbreviations and internal codes, which are crucial for accurate de-identification in electronic health records. This approach significantly outperformed existing systems in a study on pediatric oncology notes, achieving a recall rate of 0.981 and an F1 score of 0.907, while also offering a way to audit the reference standard for de-identification. AI

IMPACT LLMs can enhance the accuracy and efficiency of de-identifying sensitive health data, improving secondary use of EHRs.

RANK_REASON Research paper detailing a new method for de-identifying PHI using LLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Health Insurance Portability and Accountability Act
• Large Language Models
• OpenMed PII
• Stanford TiDE
• Texas Children's Hospital

• paper
• safety
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →