Clinical Notes with Explicit and Implicit Causal Relations

The MIMICause dataset contains 2,714 samples of clinical notes annotated for causal relations. It includes both explicit and implicit causality between entities within the same or different sentences. The data defines nine directed semantic causal relation types, such as Cause, Enable, Prevent, and Hinder.

Use Cases

• Train a sequence labeling model to classify the nine semantic causal relations (e.g., Cause, Enable, Prevent) between entity pairs in clinical text.
• Develop a relation extraction pipeline to identify both explicit and implicit causality links between entities E1 and E2 in clinical notes.
• Analyze the distribution and directionality of causal relation types (Cause, Enable, Prevent, Hinder, Other) across the 2,714 annotated samples.
• Build a model to distinguish between causality expressed within a single sentence versus across different sentences in clinical narratives.

Strengths

• Contains 2,714 annotated samples of clinical text for causal relation analysis.
• Defines nine specific, directed semantic causal relation types (Cause, Enable, Prevent, Hinder, Other).
• Includes annotations for both explicit and implicit causality, covering entities in the same or different sentences.

Limitations

• The sample size of 2,714 is relatively small for training complex deep learning models without augmentation or transfer learning.
• The dataset's scope is limited to the specific nine causal relation types defined, which may not cover all causal nuances in clinical language.
• No information is provided on the source or temporal range of the underlying clinical notes, limiting assessment of data representativeness.

Provenance

Source

Author 'pensieves' on Hugging Face.

Collection Method

Annotated dataset derived from clinical notes for causal relation extraction.

Freshness

Last updated on 2022-03-29.