Annotating temporal information in clinical narratives.

Area of Science:

• Clinical Informatics
• Medical Natural Language Processing (NLP)
• Health Data Science

Background:

• Temporal information in clinical narratives is vital for accurate patient diagnosis, treatment, and prognosis.
• Medical NLP systems require robust temporal information identification and interpretation capabilities.
• Existing resources for temporal analysis in clinical text are limited, hindering research progress.

Purpose of the Study:

• To develop a temporally annotated corpus of clinical narratives to support research in medical NLP.
• To facilitate the accurate identification and interpretation of temporal expressions and relations within clinical documents.
• To promote advancements in understanding patient timelines and clinical event sequencing.

Main Methods:

• Development of detailed annotation guidelines for clinical events, temporal expressions, and temporal relations.
• Creation of a corpus comprising 310 de-identified clinical discharge summaries.
• Systematic annotation process involving trained annotators to ensure corpus quality and consistency.

Main Results:

• A comprehensive corpus of 310 discharge summaries with rich temporal annotations was successfully created.
• Annotations include clinical events, specific temporal expressions (e.g., dates, times), and the relationships between them.
• The corpus provides a valuable resource for training and evaluating medical NLP models for temporal reasoning.

Conclusions:

• The i2b2 temporally annotated corpus is a significant resource for advancing medical NLP research.
• Accurate temporal information extraction from clinical text is essential for improving clinical decision support systems.
• This corpus will enable the development of more sophisticated NLP tools for analyzing patient histories and predicting outcomes.