A Pilot Report on Extracting Symptom Onset Date and Time From Clinical Notes in Patients Presenting With Chest Pain

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Data Research

Background:

• Accurate timing of clinical events is crucial for understanding disease progression, treatment, and patient outcomes.
• Electronic health record (EHR) documentation of time information is inconsistent, limiting its research and clinical utility.
• Manual review of EHRs for event timing is labor-intensive, necessitating automated solutions.

Purpose of the Study:

• To evaluate the efficacy of off-the-shelf natural language processing (NLP) pipelines for extracting date and time (DateTime) information from free-text EHR clinical notes.
• To assess the performance of parsedatetime and regular expression (regex) methods in automated DateTime extraction.

Main Methods:

• Pilot testing of two NLP pipelines, parsedatetime and regex, on 71 annotated clinical notes from various EHR sources (History and Physical, Emergency Department Screening, Triage Notes).
• Quantitative evaluation of extracted DateTime information accuracy and performance metrics (F1-score).

Main Results:

• The parsedatetime pipeline achieved a 50.7% accuracy in identifying correct DateTime information, with a low F1-score of 0.31.
• The regex approach failed to produce any accurate outputs.
• Both tested NLP methods demonstrated inadequate performance, requiring substantial customization for improved efficacy.

Conclusions:

• Current off-the-shelf NLP tools are insufficient for reliable automated extraction of DateTime information from EHRs.
• Advanced, rule-based NLP methods are needed to handle complex clinical narratives for accurate DateTime extraction.
• Standardized time documentation practices by clinicians are essential to enhance the usability of EHR data for research and care.