An Iterative Composite Method for the De-Identification of Medical Notes.

Artificial intelligence and large language models hold significant potential to revolutionize healthcare, driving an increasing need for access to deidentified medical text extracted from electronic medical records. This study proposes an iterative method for the de-identification of medical notes, merging multiple techniques such as pattern matching, dictionary-based identification, contextual token analysis, and iterative review to address potential residual identifiers. Applied to a diverse dataset comprising 350,000 medical notes, the method demonstrated robust performance, achieving a high F1 score of 0.992 in preliminary evaluations.