Detecting temporal expressions in medical narratives

Area of Science:

• Natural Language Processing
• Clinical Informatics
• Computational Linguistics

Background:

• Accurate temporal ordering of medical events is crucial for clinical practice and epidemiological studies.
• Existing tools like the Temporal Awareness and Reasoning Systems for Question Interpretation (TARSQI) Toolkit (TTK) were developed for newspaper articles.
• Adapting TTK for medical notes is necessary to leverage clinical data effectively.

Purpose of the Study:

• To adapt and evaluate the TTK for processing temporal expressions in Veterans Affairs (VA) clinical notes.
• To create an enhanced version, Med-TTK, and compare its performance against the original TTK.
• To assess the accuracy of identifying and classifying temporal expressions (DATE, TIME, DURATION, SET) in medical narratives.

Main Methods:

• Modified and appended rules to TTK's time tagger using a development set of 200 VA clinical notes to create Med-TTK.
• Evaluated TTK and Med-TTK on an independent set of 100 clinical notes.
• Used dual human review to establish a reference standard for temporal expressions and calculated recall and precision.

Main Results:

• The original TTK achieved a recall of 0.15 and precision of 0.27.
• Med-TTK identified 3174 temporal expressions with significantly improved recall of 0.86 and precision of 0.85.
• These results demonstrate a substantial performance increase for temporal expression identification in medical notes.

Conclusions:

• Med-TTK significantly enhances the identification and classification of temporal expressions in medical narratives compared to TTK.
• This natural language processing tool provides a foundation for longitudinal patient history mapping within electronic health records.
• Med-TTK is an open-source tool available for broader application in clinical informatics.