A clinical narrative corpus on nut allergy: annotation schema, guidelines and use case

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Allergology

Background:

• Limited availability of Spanish clinical texts hinders the development of information extraction systems.
• A collaborative effort between HUFA's Allergology Unit and IT departments was initiated to address this gap.
• The need for specialized corpora for training and evaluating allergy-related NLP models is critical.

Purpose of the Study:

• To describe and release a novel dataset of Spanish clinical notes focused on nut allergy.
• To provide a valuable resource for training and testing information extraction and named entity recognition (NER) systems.
• To validate the corpus through supervised NER experiments.

Main Methods:

• Compilation of 828 Spanish clinical notes from HUFA.
• Expert annotation of entities into medical semantic groups related to allergies.
• Triple annotation on 8% of texts for inter-annotator agreement assessment.
• Supervised NER task using fine-tuned encoder-based transformers.

Main Results:

• An average F-measure of 86.2% was achieved in the supervised NER experiments.
• The developed corpus demonstrated suitability for training and testing NER models in allergology.
• Guidelines for corpus creation were documented and provided.

Conclusions:

• The newly created nut allergy dataset is a valuable resource for advancing Spanish clinical NLP.
• The corpus facilitates the development and evaluation of advanced NER techniques in allergology.
• The successful application in NER tasks validates the corpus's utility and quality.