A distributable German clinical corpus containing cardiovascular clinical routine doctor's letters

Area of Science:

• Medical Informatics
• Natural Language Processing

Background:

• Clinical text data is crucial for developing advanced NLP models.
• Large, annotated clinical corpora are scarce, especially in German and for specific domains like cardiology.

Purpose of the Study:

• To introduce CARDIO:DE, the first freely available German clinical corpus focused on the cardiovascular domain.
• To provide a high-quality, annotated dataset for research in clinical NLP.

Main Methods:

• Compiled 500 German clinical routine doctor's letters from Heidelberg University Hospital.
• Manually de-identified all letters while preserving temporal information.
• Added manual annotation layers for medication information and CDA-compliant section classes.

Main Results:

• Successfully created CARDIO:DE, a large, manually annotated, de-identified German clinical corpus.
• The corpus maintains the original structure of clinical documents and preserves temporal data.
• Includes two high-quality annotation layers: medication and section classes.

Conclusions:

• CARDIO:DE is the first freely distributable German clinical corpus in the cardiovascular domain.
• The corpus facilitates reproducible research on NLP models for German clinical texts.
• Offers unique opportunities for advancing clinical NLP through collaborative research.