Hierarchical Annotation for Building A Suite of Clinical Natural Language Processing Tasks: Progress Note...

Area of Science:

• Natural Language Processing (NLP)
• Clinical Informatics
• Biomedical Data Science

Background:

• Natural language processing (NLP) on electronic health records (EHR) is expanding.
• Current NLP research on EHR data primarily focuses on textual features and relation prediction.
• There is a lack of annotated corpora specifically designed for modeling clinical diagnostic thinking, which involves understanding, abstraction, and reasoning.

Purpose of the Study:

• To introduce a hierarchical annotation schema for clinical text.
• To address clinical text understanding, clinical reasoning, and summarization.
• To create a new suite of tasks, Progress Note Understanding, for NLP model training and evaluation.

Main Methods:

• Developed a hierarchical annotation schema with three stages: text understanding, clinical reasoning, and summarization.
• Created an annotated corpus using publicly available daily progress notes from electronic health records (EHR).
• Defined a new suite of tasks, Progress Note Understanding, aligned with the three annotation stages.

Main Results:

• The study presents a novel hierarchical annotation schema for clinical text.
• An annotated corpus of daily progress notes was created.
• A new set of NLP tasks, Progress Note Understanding, was defined.

Conclusions:

• The developed corpus and tasks address the need for better clinical diagnostic thinking modeling in NLP.
• This work facilitates the training and evaluation of NLP models for clinical text understanding, knowledge representation, inference, and summarization.
• The Progress Note Understanding tasks offer a new benchmark for advancing NLP in clinical settings.