Extraction of Temporal Structures for Clinical Events in Unlabeled Free-Text Electronic Health Records in Russian

Area of Science:

• Medical Informatics
• Natural Language Processing (NLP)
• Computational Linguistics

Background:

• Patient medical histories are often unstructured free-form text.
• Extracting and normalizing temporal information from clinical notes is challenging.
• Existing NLP tools have limitations, especially for non-English languages.

Purpose of the Study:

• To propose and evaluate a hybrid approach for normalizing temporal expressions in patient medical history text.
• To assess the uncertainty of events based on their remoteness.
• To demonstrate the approach's effectiveness for non-English languages.

Main Methods:

• A hybrid approach combining rule-based and syntactical analysis was developed.
• A dataset of 500 manually labelled sentences was used for evaluation.
• Accuracy metrics were calculated for temporal expression extraction, normalization, and event extraction.

Main Results:

• Temporal expression extraction accuracy reached 95.5%.
• Temporal expression normalization accuracy was 94%.
• Event extraction accuracy was 74.80%.

Conclusions:

• The hybrid approach demonstrates high accuracy in normalizing temporal expressions from medical text.
• The method is particularly advantageous for non-English languages with limited NLP resources.
• This work contributes to better information extraction from clinical narratives.