Pseudonymization of PHI Items in German Clinical Reports.

Area of Science:

• Health Informatics
• Natural Language Processing
• Clinical Data Management

Background:

• Pseudonymization is crucial for protecting sensitive patient data in clinical research.
• Existing tools often lack the specific adaptations needed for complex clinical text.
• A German email corpus pseudonymization system offers a potential foundation for clinical applications.

Purpose of the Study:

• To adapt a non-clinical pseudonymization system for clinical data.
• To evaluate the efficacy and accuracy of the adapted system in replacing Protected Health Information (PHI).
• To assess the grammatical correctness, semantic, and medical plausibility of generated pseudonyms.

Main Methods:

• The study involved adapting an existing pseudonymization tool designed for a German email corpus.
• The system identifies and replaces Protected Health Information (PHI) items (names, places, organizations) with semantically appropriate, fictitious surrogates.
• Evaluation metrics included grammatical correctness, semantic plausibility, and medical plausibility of the generated substitutes.

Main Results:

• The adapted pseudonymization system demonstrated high performance in clinical data.
• Generated fictitious surrogates were found to be grammatically correct and semantically plausible.
• Error instances, including lack of medical plausibility, were notably low, occurring in less than 1% of cases.

Conclusions:

• The adapted pseudonymization system is suitable for clinical use, enhancing patient data privacy.
• The tool effectively generates plausible and correct substitutes for Protected Health Information (PHI).
• The low error rate confirms the system's reliability for clinical applications.