Current 9-1-1 call-taking practice shows poor performance for predicting clinical outcomes in two large emergency medical services datasets

Blanchard, IE; Hao, S; Verma, A; Williamson, T; Lang, ES; Doig, CJ; de Montigny, L.

Introduction: Triage of calls to 9-1-1 for medical emergencies determines the level of EMS response. Undertriage can lead to missed emergencies and poor outcomes; overtriage can lead to unnecessary lights-and-siren driving and diminished efficiency. There have been relatively few assessments of the accuracy with which high-acuity patients can be differentiated from low-acuity patients at the time of the 9-1-1 call.

Objective: To evaluate the performance of current practice in Alberta (AB) and Quebec (QC) at predicting patient acuity based on a range of clinical outcomes.

Methods: We analyzed datasets from two large urban EMS systems accounting for 611 125 adult patient-encounters linked to outcomes. Using the area under the ROC curve (AUC), we assessed the performance of the priority assigned at call-taking (dichotomized as high vs low) in predicting poor status on scene (Critical Illness Prediction score [CIP]; high vs low/moderate), paramedic intervention (time-sensitive vs non-time-sensitive), and post-transport morbidity (admission to ICU) and mortality (survival to discharge).

Results: Current practice predicted CIP with an AUC=0.71 in AB and QC; time-sensitive interventions with AB AUC=0.60 and QC AUC=0.66; admission to ICU with AB and QC AUC=0.65; and survival to discharge with AB AUC=0.63 and QC AUC=0.62. The performance was driven by low sensitivity (41–57%) and acceptable specificity (70–74%). Based on the average of the clinical outcomes, for every 1000 high-priority dispatches in each system, 936 AB patients and 961 QC patients did not show signs of high acuity, whereas for every 1000 low-priority ambulance dispatches in each system, 28 AB patients and 15 QC patients did show signs of high acuity.

Conclusion: We found evidence that current practice in both provinces poorly predicted prehospital patient acuity. These results suggest opportunities to improve medical 9-1-1 call-taking, for example through secondary triage by medically trained professionals or machine learning models trained on a caller’s past medical history and healthcare utilization.