Repurposing the clinical record: can an existing natural language processing system de-identify clinical notes?

Area of Science:

• Medical Informatics
• Natural Language Processing

Background:

• Electronic clinical documentation aids public health surveillance, quality improvement, and research.
• Current de-identification methods may not adequately protect patient data.

Purpose of the Study:

• To evaluate the MedLEE natural language processor's effectiveness in de-identifying electronic clinical notes.
• To assess MedLEE's capability to remove protected health information (PHI) while retaining medical concepts.

Main Methods:

• The MedLEE system was used to process 100 outpatient clinical notes without modification.
• The output was compared to the original notes to identify instances of residual PHI.

Main Results:

• MedLEE successfully processed clinical notes, retaining medical concepts.
• Only 3.2% (26 of 809) of PHI instances were detected in the output due to processing or identification errors.
• Detected PHI was highly transformed into normalized medical terms, potentially hindering re-identification.

Conclusions:

• MedLEE demonstrates potential as a secondary de-identification tool for electronic health records.
• The system effectively removes PHI and provides structured, coded data, enhancing privacy and data utility.