A De-identification method for bilingual clinical texts of various note types

Area of Science:

• Medical Informatics
• Natural Language Processing
• Health Data Privacy

Background:

• De-identification of personal health information is crucial for patient privacy and enables retrospective research without explicit consent.
• Existing de-identification methods primarily focus on English clinical narrative text, limiting their applicability to multilingual healthcare data.

Purpose of the Study:

• To develop and validate a novel regular expression-based de-identification method for bilingual (Korean and English) clinical records.
• To enhance the efficiency and scope of retrospective medical research by addressing limitations in current de-identification techniques.

Main Methods:

• A regular expression-based approach was developed using a training dataset of 6,039 clinical notes across 20 types.
• The developed rules were validated on a separate dataset of 5,000 notes covering 33 types.
• Fifteen regular expression rules were constructed and refined based on performance metrics.

Main Results:

• The developed de-identification method achieved high performance on the validation dataset, with a precision of 99.87% and a recall of 96.25%.
• The method demonstrated successful removal of identifiers across diverse types of bilingual clinical narrative texts.
• The approach proved effective in handling both Korean and English personal health information within clinical notes.

Conclusions:

• The proposed regular expression-based de-identification method is effective for bilingual clinical records.
• This technique facilitates easier and more comprehensive retrospective research by ensuring robust data privacy.
• The study contributes a valuable tool for managing and utilizing multilingual clinical data securely.