Clinical concept normalization with a hybrid natural language processing system combining multilevel matching and...

Area of Science:

• Medical Informatics
• Computational Linguistics
• Natural Language Processing

Background:

• Clinical concept normalization is challenging due to complex terminology in medical records.
• Accurate mapping of clinical mentions to standardized terminologies is crucial for data analysis.
• The 2019 National NLP Clinical Challenges (n2c2) focused on clinical concept normalization.

Purpose of the Study:

• To develop and evaluate a clinical NLP system for automatic normalization of clinical mentions.
• To map clinical terms to concept unique identifiers in the Unified Medical Language System (UMLS).
• To participate in and assess performance within the 2019 n2c2 Shared Task.

Main Methods:

• Developed a hybrid clinical NLP system combining multilevel matching and machine learning.
• Explored two machine learning ranking systems: ensemble of similarity features and Siamese attention network.
• Evaluated a general-purpose clinical NLP system based on Unstructured Information Management Architecture.

Main Results:

• The best system achieved an accuracy of 0.8101, ranking fifth in the 2019 n2c2 challenge.
• An improved system using a Siamese attention network reached an accuracy of 0.8209.
• Demonstrated the effectiveness of combining multilevel matching with machine learning for normalization.

Conclusions:

• The proposed approach combining multilevel matching and machine learning ranking is effective and interpretable.
• Results highlight the system's capability in clinical concept normalization.
• Opportunities exist for further performance improvement by integrating general clinical NLP systems.