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

Abstract

Electronic health records (EHRs) secondary usage with large language models (LLMs) raise privacy challenges. National regulations like GDPR and HIPAA offer protection frameworks, but specific strategies are needed to mitigate risk in generative AI. Risks can be reduced by using strategies like privacy-preserving locally deployed LLMs, synthetic data generation, differential privacy, and deidentification. Depending on the task, strategies should be employed to increase compliance with patient privacy regulatory frameworks.