Food entries in a large allergy data repository

Area of Science:

• Medical Informatics
• Allergy and Immunology
• Natural Language Processing

Background:

• Accurate documentation of food adverse sensitivities in electronic health records (EHRs) is critical for patient safety.
• Existing terminologies may not fully capture the breadth of food allergen concepts.

Purpose of the Study:

• To examine, encode, and group foods causing adverse sensitivities using natural language processing (NLP) and standard terminologies.
• To assess the concept coverage of SNOMED-CT and Unique Ingredient Identifiers (UNII) for food allergens.

Main Methods:

• Utilized the Medical Text Extraction, Reasoning, and Mapping System (MTERMS) to process allergy repository data.
• Normalized food allergen terms into concepts, encoded using SNOMED-CT and UNII.
• Categorized allergen concepts, calculated frequencies, and performed external validation.

Main Results:

• Identified 672 food allergen concepts from 158,552 records.
• High-frequency groups included shellfish, fruits/vegetables, and dairy.
• SNOMED-CT demonstrated broader concept coverage than UNII.

Conclusions:

• Food adverse sensitivities are diverse, with existing standard terminologies exhibiting coverage gaps.
• New strategies are required for representing and standardizing food adverse sensitivity concepts in EHRs.