Cohort Identification from Free-Text Clinical Notes Using SNOMED CT's Hierarchical Semantic Relations

Area of Science:

• Medical Informatics
• Natural Language Processing
• Biomedical Ontologies

Background:

• Supervised machine learning approaches for cohort identification have limitations.
• Accurate patient cohort identification is crucial for clinical research.

Purpose of the Study:

• To propose a new cohort identification system leveraging SNOMED CT's semantic hierarchy.
• To overcome limitations of existing supervised machine learning methods.
• To evaluate the system's performance on real-world clinical data.

Main Methods:

• Processed eligibility criteria and clinical notes from the 2018 National NLP Clinical Challenge (n2c2).
• Mapped clinical data to SNOMED CT concepts.
• Measured semantic similarity between eligibility criteria and patients using SNOMED CT.
• Determined patient eligibility based on a similarity score threshold.

Main Results:

• The system achieved an average F1 score of 0.933 for three eligibility criteria.
• Performance surpassed previously reported results from the 2018 n2c2.
• Demonstrated SNOMED CT's capability for cohort identification without external training data.

Conclusions:

• SNOMED CT alone is sufficient for effective cohort identification.
• The proposed system offers a robust alternative to traditional supervised methods.
• This approach enhances the efficiency and accuracy of clinical trial recruitment.