Machine learning models predicting undertriage in telephone triage.

Area of Science:

• Emergency medicine
• Health informatics
• Machine learning applications

Background:

• Undertriage in emergency settings leads to poorer patient outcomes.
• Machine learning (ML) shows promise in improving triage accuracy compared to conventional methods.
• No ML-based undertriage prediction models currently exist for prehospital telephone triage.

Purpose of the Study:

• To develop and validate ML models for predicting undertriage in prehospital telephone triage.
• To identify key predictors and risk factors associated with prehospital undertriage.

Main Methods:

• Retrospective cohort study using the largest after-hour house-call (AHHC) service dataset in Japan (Nov 2018 - Jan 2021).
• Developed five ML models: Support Vector Machine (SVM), Lasso Regression (LR), Random Forest (RF), Gradient-Boosted Decision Tree (XGB), and Deep Neural Network (DNN).
• Utilized telephone triage level and routinely available data (age, sex, chief complaints, comorbidities) to predict undertriage.

Main Results:

• 15,442 patients were analyzed, with 1.9% (298) identified as undertriaged.
• RF and XGB models demonstrated superior performance in predicting undertriage.
• Area Under the Receiver Operating Characteristic Curve (AUROC) values for RF and XGB were 0.81 (0.76-0.86) and 0.80 (0.75-0.84), respectively, outperforming other models.

Conclusions:

• ML models, particularly RF and XGB, can effectively predict undertriage in prehospital telephone triage.
• Early detection of undertriage through ML can lead to timely interventions and improved patient outcomes.
• Identifying undertriage risk factors can inform revisions to conventional telephone triage protocols.