Discovering body site and severity modifiers in clinical texts

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Medical Informatics

Background:

• Clinical text analysis is crucial for extracting patient information.
• Identifying body site and severity modifiers is challenging but vital for clinical decision-making.

Purpose of the Study:

• To develop and evaluate computational methods for discovering body site and severity modifiers in clinical texts.
• To frame modifier discovery as a relation extraction task within a supervised machine learning context.

Main Methods:

• Utilized a supervised machine learning framework with rich linguistic features.
• Employed a support vector machine (SVM) model for relation classification.
• Evaluated models on two annotated corpora and compared them against rule-based baselines.
• Conducted cross-domain portability and feature ablation experiments.

Main Results:

• Achieved high F1 scores for body site modifiers (0.740-0.908) and severity modifiers (0.905-0.929).
• Demonstrated strong performance on both in-domain and out-of-domain data.
• Identified token and named entity features as most salient for performance.

Conclusions:

• The developed computational methods effectively discover body site and severity modifiers in clinical texts.
• Performance approaches that of human annotators, indicating the robustness of the approach.
• The best performing system is released as open-source software (cTAKES) for broader use.