Advanced natural language processing technique to predict patient disposition based on emergency triage notes.

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Informatics
• Natural Language Processing

Background:

• Emergency department (ED) patient disposition prediction is crucial for resource management.
• Triage notes contain valuable information for predicting patient flow.
• Current methods may not fully leverage unstructured clinical text data.

Purpose of the Study:

• To assess the efficacy of machine learning (ML) and NLP in predicting patient disposition using ED triage notes.
• To develop and validate a deep-learning algorithm for this predictive task.
• To demonstrate the potential of AI in supporting clinical decision-making in the ED.

Main Methods:

• A retrospective cohort of ED triage notes was analyzed.
• A deep-learning model utilizing Bidirectional Encoder Representations from Transformers (BERT) was developed.
• The dataset was split into 80% for training and 20% for testing the algorithm.

Main Results:

• The developed algorithm achieved 83% accuracy and an 0.88 area under the curve (AUC).
• Key performance metrics included 72% sensitivity, 86% specificity, 56% precision, and 63% F1-score.
• The model demonstrated strong predictive capabilities for patient disposition.

Conclusions:

• ML and NLP can accurately predict patient disposition from ED triage notes.
• The algorithm can aid clinicians in early identification of patients requiring admission.
• This approach has the potential to optimize ED resource allocation and reduce cognitive load on staff.