A software tool for removing patient identifying information from clinical documents.

Area of Science:

• Medical Informatics
• Health Data Security
• Clinical Data Management

Background:

• Patient privacy is paramount in healthcare research.
• Accurate de-identification of clinical data is crucial for compliance with regulations like HIPAA.
• Existing de-identification methods may struggle with diverse clinical document types.

Purpose of the Study:

• To develop and evaluate a software tool, the Medical De-identification System (MeDS), for accurate de-identification of clinical documents.
• To assess the effectiveness of MeDS in removing patient-identifying information from various data sources.
• To evaluate the impact of de-identification on the readability and interpretability of clinical reports.

Main Methods:

• Developed the Medical De-identification System (MeDS) software.
• Conducted two evaluations using Health Level Seven (HL7) messages and pathology reports.
• Compared MeDS de-identification results against a human-reviewed gold standard.
• Calculated metrics including successful scrubs, missed identifiers, and over-scrubs.
• Assessed the readability and interpretability of de-identified documents.

Main Results:

• In the first evaluation, MeDS achieved high scrub rates: 99.06% for HIPAA-specified identifiers and 98.26% for non-HIPAA identifiers.
• Following modifications, the second evaluation showed improved scrub rates: 99.47% for HIPAA-specified and 96.93% for non-HIPAA identifiers.
• Approximately 95% of the de-identified messages remained readable and interpretable.

Conclusions:

• MeDS demonstrates high accuracy in de-identifying diverse clinical documents from multiple sources.
• The system effectively preserves the readability and interpretability of de-identified reports.
• MeDS is a valuable tool for maintaining the utility of clinical data for research while protecting patient privacy.