Use of natural language processing method to identify regional anesthesia from clinical notes

Area of Science:

• Anesthesiology
• Medical Informatics
• Data Science

Background:

• Inconsistent documentation of regional anesthesia hinders research and quality improvement.
• Existing data capture methods for regional anesthesia are often incomplete and inaccurate.
• Standardized guidance for documenting regional anesthesia is lacking.

Purpose of the Study:

• To evaluate a natural language processing (NLP)-based algorithm for identifying regional anesthesia in unstructured clinical notes.
• To compare the performance of the NLP algorithm against structured data from the Corporate Data Warehouse.
• To assess the completeness and accuracy of regional anesthesia documentation.

Main Methods:

• A cross-sectional study was conducted using postoperative clinical notes from six Veterans Health Administration hospitals.
• An NLP algorithm was developed and executed to identify regional anesthesia procedures.
• Algorithm performance was compared to structured data (Corporate Data Warehouse) using measures of agreement.

Main Results:

• The NLP algorithm identified 96.6% of regional anesthesia cases present in the referent data with high accuracy (82.5%) and a low false negative rate (0.8%).
• The algorithm identified over twice the number of regional anesthesia cases (9154) compared to the structured referent data (4606).
• This suggests significant underdocumentation in administrative and clinical databases.

Conclusions:

• Natural language processing (NLP) is a promising tool for extracting accurate clinical information from unstructured notes when existing databases are incomplete.
• The findings raise concerns about the reliability of current administrative and clinical databases for capturing regional anesthesia data.
• Improved documentation practices and advanced NLP methods are needed to ensure accurate data capture for research and quality improvement.