Summary
This technical report outlines a project focused on the de-identification of clinical notes using artificial intelligence (AI) to address privacy and governance concerns. The organization faced challenges in sharing free-text notes that contained valuable scientific and operational insights due to risks associated with protected health information (PHI). To resolve this, a note de-identification pipeline was developed, combining Epic knowledge with AI technology. In-house clinicians annotated the notes to enhance precision, while legal and compliance teams established an expert determination framework that successfully passed audit reviews. The resulting solution allows the client to process approximately 10.5 million notes annually, facilitating safer sharing of de-identified data with trusted partners. This initiative not only supports various downstream applications, such as clinical decision support and trial matching, but also generates revenue by transforming sensitive data into a valuable asset for research and development. The report emphasizes the collaborative effort required across multiple teams to achieve a compliant and effective de-identification process.
