Title : Utilizing AI to optimize EMS response to acute mental illness and resulting ER resource allocation
Abstract:
Background: Typically emergency medical technicians(EMTs) are not equipped to handle calls related to mental health and substance abuse. This deficiency often necessitates the transport of these patients to emergency departments, a practice that not only strains public health resources but also potentially diverts ambulances from trauma-related emergencies. This study aims to develop a predictive model capable of accurately categorizing the nature of incoming emergency medical services (EMS) calls, thereby facilitating a more nuanced response strategy. This research also highlights the inappropriate admission of patients with mental health crises to EDs, often at public expense, when mental health treatment is more appropriate. The proposed policy intervention involves the establishment of EMS crew to be staffed with mental health professionals when the predictive algorithm identifies such calls.
Methods: After a retrospective study of AI projects in this space, the model was established using 24 million patient interactions from NYC’s EMS Incident Dispatch Data collected between January 2005 and March 2022. The model was validated using additional data from April-December 2022. Class-imbalance methodology and gradient boosting strategies were implemented to train and test the model.
Results: The model achieved 94.5% predictability, compared to 92.3% accuracy for EMS operators.
Conclusions: Across 9 million NYC 911 calls annually, the algorithm could more effectively allocate resources for 198,000 cases. Staffing EMS crews with mental health professionals could also reduce ER visits and ambulance transfers––with $5.6 billion spent in the U.S. annually on mental and substance use disorder ER visits, a savings of $123 million or more may be possible.
