Protected Health Information filter (Philter): accurately and securely de-identifying free-text clinical notes

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Research Data

Background:

• Structured electronic health records lack comprehensive patient data.
• Clinical notes contain rich, detailed patient information crucial for research.
• Protected Health Information (PHI) in notes hinders their use in research.

Purpose of the Study:

• To develop an effective tool for de-identifying clinical notes.
• To enable the use of valuable clinical note data for research.
• To improve patient phenotyping, outcome detection, and surveillance.

Main Methods:

• Created the largest manually annotated clinical note corpus for PHI.
• Developed Philter, a customizable, open-source de-identification software.
• Evaluated Philter's performance against previous de-identification methods.

Main Results:

• Philter demonstrates substantial real-world improvements over prior de-identification techniques.
• The developed software offers enhanced accuracy and usability for clinical note de-identification.
• The large annotated corpus facilitates more robust model training.

Conclusions:

• Philter provides a significant advancement in de-identifying clinical notes for research.
• The tool facilitates the utilization of previously inaccessible clinical data.
• This work supports more comprehensive patient surveillance and medical research.