Privacy preserving strategies for electronic health records in the era of large language models

The secondary use of electronic health records (EHRs) involves the utilization of EHR data for various purposes other than their original intent in clinical operations, such as clinical research, health systems and services research, patient registries, quality improvement, disease surveillance, and other areas beyond direct patient care 1 . This secondary use of electronic health records (EHRs) has significantly accelerated in the past decade with advances in artificial intelligence (AI), especially in large language models (LLMs) 2 . EHRs are increasingly used with LLMs for primary uses, such as clinical documentation, generation and summarization, as well as for secondary uses, such as information extraction, retrieval, identification of eligible patients for research, medical education, and outcome reporting. Safeguarding both the structured and unstructured sensitive health information (SHI) of patients from EHRs is essential, particularly when LLMs are used for secondary uses 3 , 4 . Most countries across the world have stipulated regulatory privacy acts, guidelines and frameworks to achieve this goal.