Ambulance dispatch prioritisation for traffic crashes using machine learning: A natural language approach.

Area of Science:

• Emergency Medical Services
• Data Science
• Traffic Safety

Background:

• Increasing demand for emergency ambulances necessitates appropriate dispatch prioritization.
• Traffic crashes present complex scenarios with varying patient needs, making accurate assessment difficult.
• Bystander reports during emergency calls may not fully capture patient acuity.

Purpose of the Study:

• To evaluate the predictive accuracy of emergency medical dispatcher (EMD) text and dispatch codes for lights and sirens (L&S) responses to traffic crashes.
• To determine if machine learning can improve the identification of high-acuity traffic crash incidents.

Main Methods:

• Retrospective cohort study of 11,971 traffic crashes from 2014-2016 in Perth, Western Australia.
• Utilized Medical Priority Dispatch System (MPDS) determinant codes and EMD-generated text as features.
• Employed machine learning algorithms, including an Ensemble model, with natural language processing (Bag of Words) for text analysis.

Main Results:

• An Ensemble machine learning model achieved high accuracy (0.980) in predicting the need for L&S responses.
• The model accurately combined MPDS codes and EMD text to differentiate between incidents requiring L&S and non-L&S responses.
• 22.3% of traffic crashes were retrospectively identified as requiring an L&S response.

Conclusions:

• A combination of EMD text and MPDS codes can accurately predict the need for L&S ambulance responses to traffic crashes.
• Implementing machine learning algorithms can enhance dispatch accuracy for traffic incidents.
• Improved dispatch accuracy has the potential to increase the efficiency of emergency medical services.