GRASCCO - The First Publicly Shareable, Multiply-Alienated German Clinical Text Corpus

Area of Science:

• Medical Informatics
• Computational Linguistics
• Corpus Linguistics

Background:

• Clinical documents contain sensitive patient information, limiting their public use.
• Developing shareable clinical corpora is crucial for research and development in medical informatics.
• Synthetic data generation offers a potential solution for privacy preservation in clinical text data.

Purpose of the Study:

• To introduce GRASCCO, a novel, privacy-preserving, German-language clinical corpus.
• To assess the linguistic representativeness of the synthetic GRASCCO corpus compared to real clinical data.
• To facilitate public access to German clinical text data for research purposes.

Main Methods:

• Creation of GRASCCO through alienation techniques applied to real clinical documents.
• Development of a comparable real clinical corpus for validation purposes.
• Comparative analysis of linguistic features between the synthetic and real corpora.

Main Results:

• GRASCCO comprises approximately 60 clinical documents and over 43,000 tokens.
• The synthetic corpus was generated via alienation steps to obfuscate sensitive information.
• Evidence suggests GRASCCO effectively represents common clinical language use.

Conclusions:

• GRASCCO is a publicly shareable, synthetic German clinical corpus.
• The corpus retains the characteristics of authentic clinical language.
• GRASCCO can be utilized for research without legal restrictions, advancing medical informatics.