Negation Detection for Clinical Text Mining in Russian.

Area of Science:

• Medical informatics
• Natural Language Processing (NLP)
• Machine Learning

Background:

• Predictive modeling in medicine relies on extracting features from unstructured clinical texts.
• Existing Natural Language Processing (NLP) tools for Russian medical records are limited.
• Negation detection is crucial for accurate interpretation of clinical information.

Purpose of the Study:

• To develop and evaluate a corpus-free machine learning module for negation detection in Russian clinical texts.
• To assess the module's performance in identifying denied, not mentioned, or present diseases.
• To demonstrate the utility of negation detection in improving predictive models for patient outcomes.

Main Methods:

• A corpus-free machine learning approach utilizing a gradient boosting classifier.
• Training and evaluation of a negation detection model on clinical text data.
• Classification of disease mentions as denied, not mentioned, or present.

Main Results:

• The negation detection module achieved an average F-score ranging from 0.81 to 0.93 for five diseases.
• Demonstrated improved prediction of surgery presence in acute coronary syndrome patients using negation detection.
• The model effectively distinguishes between affirmed and negated disease mentions.

Conclusions:

• The developed negation detection module is effective for Russian clinical texts.
• Negation detection significantly enhances the accuracy of medical predictive modeling.
• This tool addresses a critical gap in NLP capabilities for Russian healthcare.