De-Identifying GRASCCO - A Pilot Study for the De-Identification of the German Medical Text Project (GeMTeX) Corpus

Area of Science:

• Medical Informatics
• Natural Language Processing
• Data Privacy

Background:

• The German Medical Text Project (GeMTeX) is a major initiative for processing German clinical documents.
• De-identification of clinical text is crucial for data privacy and secondary use of medical records.

Purpose of the Study:

• To introduce the architecture of the de-identification pipeline developed for the GeMTeX project.
• To create a robust and efficient system for removing protected health information (PHI) from German clinical texts.

Main Methods:

• A multi-stage pipeline involving data export, import into the INCEpTION platform, automated PHI pre-tagging using Averbis Health Discovery, manual curation, and automated PHI replacement.
• Implementation and pilot testing at Data Integration Centers in Leipzig and Erlangen with annotators and curators.

Main Results:

• Successful de-identification of clinical documents.
• High inter-annotator agreement (Krippendorff's α ≈ 0.97) achieved during the annotation campaign on the GRASSCO corpus.
• Creation of 1.4 K curated PHI annotations.

Conclusions:

• The developed pipeline effectively de-identifies German clinical documents.
• The curated PHI annotations are released as open-source data.
• This constitutes the first publicly available German clinical language text corpus with PHI metadata.