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

INTRODUCTION: Demand for emergency ambulances is increasing, therefore it is important that ambulance dispatch is prioritised appropriately. This means accurately identifying which incidents require a lights and sirens (L&S) response and those that do not. For traffic crashes, it can be difficult to identify the needs of patients based on bystander reports during the emergency phone call; as traffic crashes are complex events, often with multiple patients at the same crash with varying medical needs. This study aims to determine how well the text sent to paramedics en-route to the traffic crash scene by the emergency medical dispatcher (EMD), in combination with dispatch codes, can predict the need for a L&S ambulance response to traffic crashes. METHODS: A retrospective cohort study was conducted using data from 2014 to 2016 traffic crashes attended by emergency ambulances in Perth, Western Australia. Machine learning algorithms were used to predict the need for a L&S response or not. The features were the Medical Priority Dispatch System (MPDS) determinant codes and EMD text. EMD text was converted for computation using natural language processing (Bag of Words approach). Machine learning algorithms were used to predict the need for a L&S response, defined as where one or more patients (a) died before hospital admission, (b) received L&S transport to hospital, or (c) had one or more high-acuity indicators (based on an a priori list of medications, interventions or observations. RESULTS: There were 11,971 traffic crashes attended by ambulances during the study period, of which 22.3 % were retrospectively determined to have required a L&S response. The model with the highest accuracy was using an Ensemble machine learning algorithm with a score of 0.980 (95 % CI 0.976-0.984). This model predicted the need for an L&S response using both MPDS determinant codes and EMD text. DISCUSSION: We found that a combination of EMD text and MPDS determinate codes can predict which traffic crashes do and do not require a lights and sirens ambulance response to the scene with a high degree of accuracy. Emergency medical services could deploy machine learning algorithms to improve the accuracy of dispatch to traffic crashes, which has the potential to result in improved system efficiency.

A predictive ambulance dispatch algorithm to the scene of a motor vehicle crash: the search for optimal over and under triage rates.

BACKGROUND: Calls for emergency medical assistance at the scene of a motor vehicle crash (MVC) substantially contribute to the demand on ambulance services. Triage by emergency medical dispatch systems is therefore important, to ensure the right care is provided to the right patient, in the right amount of time. A lights and sirens (L&S) response is the highest priority ambulance response, also known as a priority one or hot response. In this context, over triage is defined as dispatching an ambulance with lights and sirens (L&S) to a low acuity MVC and under triage is not dispatching an ambulance with L&S to those who require urgent medical care. We explored the potential for crash characteristics to be used during emergency ambulance calls to identify those MVCs that required a L&S response. METHODS: We conducted a retrospective cohort study using ambulance and police data from 2014 to 2016. The predictor variables were crash characteristics (e.g. road surface), and Medical Priority Dispatch System (MPDS) dispatch codes. The outcome variable was the need for a L&S ambulance response. A Chi-square Automatic Interaction Detector technique was used to develop decision trees, with over/under triage rates determined for each tree. The model with an under/over triage rate closest to that prescribed by the American College of Surgeons Committee on Trauma (ACS COT) will be deemed to be the best model (under triage rate of ≤ 5% and over triage rate of between 25-35%. RESULTS: The decision tree with a 2.7% under triage rate was closest to that specified by the ACS COT, had as predictors-MPDS codes, trapped, vulnerable road user, anyone aged 75 + , day of the week, single versus multiple vehicles, airbag deployment, atmosphere, surface, lighting and accident type. This model had an over triage rate of 84.8%. CONCLUSIONS: We were able to derive a model with a reasonable under triage rate, however this model also had a high over triage rate. Individual EMS may apply the findings here to their own jurisdictions when dispatching to the scene of a MVC.

Motor Vehicle Crash Characteristics That Are Predictive of High Acuity Patients: An Analysis of Linked Ambulance and Crash Data.

BACKGROUND: Motor vehicle crashes (MVCs) comprise a significant component of emergency medical service workload. Due to the potential for life-threatening injuries, ambulances are often dispatched at the highest priority to MVCs. However, previous research has shown that only a small proportion of high-priority ambulance responses to MVCs encounter high acuity patients. Alternative methods for triaging patients over the phone are required to reduce the burden of over-triage. One method is to use information readily available at the scene (e.g. whether a person was a motorcyclist, ejection status or whether an airbag deployed) as potential predictors of high acuity. Methods: A retrospective cohort study was conducted of all MVC patients in Perth attended by St John Western Australia between 2014 and 2016. Ambulance data was linked with Police crash data. The outcome variable of interest was patient acuity, where high acuity was defined as where a patient (1) died on-scene or (2) was transported by ambulance on priority one (lights & sirens) from the scene to hospital. Crash characteristics that are predictive of high acuity patients were identified by estimating crude odds ratios and 95% confidence intervals. Results: Of the 18,917 MVC patients attended by SJ-WA paramedics, 6.4% were classified as high acuity patients. The odds of being a high acuity patient was greater for vulnerable road users (motorcyclists, pedestrians and cyclists) than for motor vehicle occupants (OR 3.19, 95% CI, 2.80-3.64). A 'not ambulant patient' (one identified by paramedics as unable to walk or having an injury incompatible with being able to walk) had 15 times the odds of being high acuity than ambulant patients (OR 15.34, 95% CI, 11.48-20.49). Those who were trapped in a vehicle compared to those not trapped (OR 4.68, 95% CI, 3.95-5.54); and those who were ejected (both partial and full) from the vehicle compared to those not ejected (OR 6.49, 95% CI, 4.62-9.12) had higher odds of being high acuity patients. Discussion: There were two important findings from this study: (1) few MVC patients were deemed to be high acuity; and (2) several crash scene characteristics were strong predictors of high acuity patients.