Rare Disease Identification from Clinical Notes with Ontologies and Weak Supervision

Area of Science:

• Medical Informatics
• Computational Linguistics

Background:

• Identifying rare diseases in clinical notes is challenging due to limited data for machine learning and the need for expert annotation.
• Existing methods often struggle with the scarcity of rare disease cases in clinical text.

Purpose of the Study:

• To propose and evaluate a novel method for identifying rare diseases from clinical notes using ontologies and weak supervision.
• To improve the accuracy and efficiency of rare disease case detection in large clinical datasets.

Main Methods:

• A two-step approach: Text-to-UMLS linking using a named entity linking tool, customized rules, and BERT contextual representations with weak supervision.
• UMLS-to-ORDO matching to map identified concepts to rare diseases in the Orphanet Rare Disease Ontology (ORDO).
• Utilized MIMIC-III intensive care discharge summaries as a case study, without requiring expert-annotated data.

Main Results:

• Weak supervision significantly improved the Text-to-UMLS linking process.
• The proposed pipeline successfully identified rare disease cases from discharge summaries.
• The method surfaced cases missed by manual International Classification of Diseases (ICD) coding.

Conclusions:

• Ontology-based weak supervision offers a powerful approach to overcome data scarcity in rare disease identification.
• This method enhances the discovery of rare disease cases from unstructured clinical text, improving upon traditional coding systems.