Machine Learning and Initial Nursing Assessment-Based Triage System for Emergency Department.

Area of Science:

• Emergency Medicine
• Clinical Informatics
• Machine Learning in Healthcare

Background:

• Emergency department (ED) triage aims to identify patients at high risk for adverse outcomes.
• Existing triage systems, such as the Korea Triage and Acuity Scale (KTAS) and Sequential Organ Failure Assessment (SOFA), have limitations in accurately predicting critical events.
• The integration of machine learning (ML) with initial nursing assessment (INA) offers a potential advancement in predictive triage accuracy.

Purpose of the Study:

• To develop and validate a novel triage system using ML and INA to predict adverse clinical outcomes in the ED.
• To compare the predictive performance of the ML and INA-based system against established triage tools (KTAS and SOFA).

Main Methods:

• A retrospective analysis of 86,304 ED visits between January 2016 and December 2017.
• Development of four classifiers: logistic regression and deep learning on INA and low-dimensional INA (LD INA), and logistic regression on KTAS and SOFA.
• Identification of key predictive variables using random forest information gain for LD modeling.

Main Results:

• The ML and INA-based triage system demonstrated superior predictive accuracy compared to KTAS and SOFA.
• Area Under the Curve (AUC) values for the INA models (logistic regression and deep learning) ranged from 87.2% to 87.6%.
• The low-dimensional INA models achieved AUC values between 80.7% and 81.2%, indicating efficient prediction.

Conclusions:

• A novel ML and INA-based triage system has been successfully developed for emergency departments.
• This new system significantly improves the prediction of clinical outcomes compared to current KTAS and SOFA triage methods.
• The findings suggest a promising new approach for enhancing patient safety and resource allocation in ED settings.