Identification of Preanesthetic History Elements by a Natural Language Processing Engine

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Natural Language Processing

Background:

• Streamlining preanesthesia evaluations can enhance resource utilization and efficiency.
• Natural language processing (NLP) extracts relevant information from unstructured text.
• A clinical NLP pipeline was developed to identify preoperative medical history from clinical notes.

Purpose of the Study:

• To evaluate the efficacy of a clinical NLP pipeline in identifying preoperative medical history.
• To compare NLP-identified medical conditions with those noted by anesthesiologists.
• To assess the potential of NLP in automating preanesthesia evaluations.

Main Methods:

• Collected patient notes from electronic medical records up to one day before anesthesia clinic appointments.
• Utilized a Named Entity Recognition (NER) pipeline, an NLP model, to process free-text notes.
• Mapped recognized medical concepts to a predefined list of conditions relevant to preanesthesia evaluation.

Main Results:

• The NLP pipeline and anesthesiologists agreed on condition presence/absence in 81.24% of cases.
• NLP identified information missed by anesthesiologists in 16.57% of instances.
• NLP failed to identify conditions noted by anesthesiologists in only 2.19% of instances.

Conclusions:

• NLP successfully identified relevant medical conditions from unstructured preanesthesia notes.
• Automated risk stratification tools can offer clinical decision support.
• Further research is needed to integrate NLP into clinical workflows and validate its effectiveness.