Building a Machine Learning-based Ambulance Dispatch Triage Model for Emergency Medical Services.

Area of Science:

• Emergency medicine
• Artificial intelligence
• Data science

Background:

• Emergency Medical Services (EMS) call center specialists face challenges in accurately assessing emergency case acuity within limited timeframes.
• Current protocols for dispatch decision-making may exhibit limitations in sensitivity and specificity.
• Machine learning (ML) models offer potential for capturing complex patterns and providing rapid, accurate predictions.

Purpose of the Study:

• To develop and evaluate a proof-of-concept machine learning model for enhanced prediction of emergency case acuity.
• To improve the accuracy of triage decisions made by EMS call center specialists.

Main Methods:

• Utilized a dataset of over 360,000 structured emergency call center records from Singapore (2018-2020).
• Engineered features from call records to train multiple machine learning models.
• Employed a Random Forest model as the primary predictive algorithm.

Main Results:

• The Random Forest model demonstrated superior performance in predicting case acuity.
• Achieved a 15% absolute reduction in the over-triage rate compared to human dispatchers.
• Maintained a comparable under-triage rate to that of experienced call center specialists.

Conclusions:

• The developed ML model shows significant potential as a decision support tool for EMS dispatchers.
• Integration of this tool alongside existing protocols can optimize ambulance dispatch triage.
• The model can contribute to more efficient utilization of emergency ambulance resources.