Can We Be Outsmarted By AI?
In a given 9-1-1 system, the prehospital phase of medical care includes a structured sequence of events beginning with intake at dispatch call centers, directed vehicular responses, subsequent on-scene procedures, treatment considerations, and disposition decision-making, which can include the mode and type of transport to the appropriate facility for the condition at hand.1,2
For the past six decades, EMS personnel have navigated each of these steps and challenges based on education, training, and certain types of targeted technology (e.g., computerized monitor-defibrillators, feedback ventilatory devices, point of care ultrasound, etc). Today, however, very sophisticated artificial intelligence (AI) applications are being explored for use at almost every stage of these traditional aspects of management.2-5 While hurdles still exist with respect to inevitable on-going adaptations and modifications, integration of AI across the EMS ecosystem has the potential for creating a more cohesive and efficient prehospital workflow. Improvements in dispatch functions may expedite triage and time-critical responses to applicable patients, while more accurate on-scene prognostication can better inform allocation of resources and transport planning.
Together, inter-related enhancements may reinforce and augment one another and, in turn, ultimately contribute to improved patient outcomes. Therefore, the purpose of this review is to explore current proposed applications of AI in prehospital EMS, and the potential roles of AI in enhancing triage and prognostication, clinical decision support, and operational management.
Triage and Prognostication at the Dispatch Office
Rapid identification of severe conditions at public safety answering points is paramount to expedite time-dependent interventions including during the dispatcher phase of care.3 In some settings, AI has been demonstrated to identify 36% of out-of-hospital cardiac arrest calls within the first minute, compared to 25% identification with human dispatchers.6 While the specifics of the large language models are complex and nuanced, word selection, pattern recognition, and potentially even caller tone of voice can play a role in call identification.
More broadly, research has demonstrated that AI may outperform conventional triage tools, clinical decision aids, and human comparators. One recent review found that AI was superior to traditional dispatch procedures in nearly 80% of related studies. Another narrative review demonstrated that this advantage extended across the spectrum of trauma, stroke, and cardiac emergencies.4,7 In trauma patients, AI has been used to triage calls to determine level of severity of injury or predict the level of care needed, including the number of apparatus, supervisors, or crews to send.2,4
On-Scene Clinical Decision Support
During active prehospital patient care, an AI-directed review of continuous data stream can allow for real-time information about the patient’s condition and, most importantly, it can alert care providers to condition changes that would require rapid or eventual intervention.1,4 In South Korea, comprehensive technical-support using AI (CONNECT-AI) was developed for EMS to provide live patient data and video streaming for emergency care centers, while also providing guidance for prehospital patient management.8 Moreover, this real-time information sharing can allow collaboration from medical providers, which can be particularly helpful in times of prolonged transport or otherwise limited resource settings.1,8
AI models have even been used to predict outcomes after defibrillation, which, in turn, could indicate either futility or the need to continue resuscitative efforts.4 Ultimately, AI may better predict the likelihood of many time-sensitive conditions being still salvageable or, in contrast, determine a clear likelihood of a poor outcome. This prognostic information may not only help to direct triage and destination dispositions, but it could also support the decision to terminate further interventional efforts in applicable cases.
Further integration of prehospital patient care records into AI models (that are coupled with pre-established clinical guidelines and patient care pathways) may help to bolster an EMS clinician’s ability to make time-critical decisions and/or on-scene resource allocation.1 Appropriate testing, critique and external validation will still be required to fully determine the feasibility, accuracy and value in various prehospital settings, but such scrutiny will also lead to improvements in AI and help to direct appropriate evolution in terms of relevant sensitivity and specificity of application functions.
Operational Management
AI has also been implemented to assist in operational management of prehospital care including route optimization and strategic placement of response units. Much akin to the familiar automobile navigation tools used to guide one’s travel direction, efficiency of routes traveled by EMS has been managed by AI through the assessment of traffic patterns, but also temporal and spatial demands.1,2 From a predictive standpoint, AI has the power to incorporate large databases of past call volumes and patterns to predict the likelihood of similar instances occurring at various times and locations. Implementation of this type of technology has been shown to reduce response intervals (as much as 16%) through resulting optimization of resources.3,4 This predictive feature of AI has potential for application particularly in mass-casualty events, mass gathering situations, or during natural disasters such floods and dangerous tornado debris in which there are significant disruptions of roadways that may hinder rapid patient transport and optimal access to definitive care of critical conditions.
Limitations of AI Implementation
While the benefits of AI across dispatch functions, on-scene clinical decision support, and overall operational management are promising, there are several limitations that must be addressed before AI can be broadly adopted and implemented in prehospital care. For EMS personnel to accept and trust AI outputs, its recommendations must be explained within the context of the individual patient’s presentation, rather than a black box suggestion.4 Perfunctory adherence to AI-generated predictions of futility could potentially result in a type of self-fulfilling prophecy.
Another critical barrier is the concern over data privacy, regulatory considerations, and medical-legal protection. AI models often rely on sensitive patient information, audio recordings, and free-text data, all of which may be subject to privacy policies and thus can limit their availability and appropriateness.3 Adding to difficulties, the current heterogeneity of the different EMS datasets and patient care records amongst various EMS systems and health systems would require extensive data linkage and more comprehensive management to effectively train and incorporate optimal AI modeling.3
For AI models to be as nuanced as a human dispatcher or a well-trained medical professional, they would require a large, interoperable and more standardized dataset that would seamlessly link EMS and outcome data across any system of care. That ultimate achievement will remain difficult within the current state of data fragmentation, privacy constraints, and cost barriers.3 However, such barriers should be overcome as AI datasets advance and can perform more complex data linking. Finally, many of the studies in the literature rely on retrospective information for their creation and testing of AI.3 In the future, prospective (real-time) studies will be required to further establish the robust evidence needed to facilitate a trusted integration of AI into real life clinical settings.
In summary, based on evolving experience, AI has the potential to be integrated into nearly every stage of EMS functions, from dispatch triage and outcome prognostication to on-scene decision support and real-time operational management of EMS systems. In the future, such initiatives could also be extended to fleet management, logistical support, and even vehicle safety monitoring for individual drivers, as well as quality assurance activities extending from patients’ complaints to medical care performance.
While the human factor remains critical, initial studies already suggest that AI can match or outperform experienced dispatchers in identifying life-threatening conditions, while also improving resource allocation and transport logistics. Challenges remain in terms of transparency, data privacy, infrastructure and regulatory considerations as well as the obvious need for validation and applicability across different systems of care. Moreover, the perspective of patients and the public-at-large regarding AI-directed prehospital care remains largely unknown. In the meantime, even as corroborative evidence grows, AI may be best positioned as a supportive tool for EMS personnel and not a proposed replacement for their clinical judgment.
About the Authors
Megan Amarbayan, MD, MPP was senior medical student at the Medical College of Wisconsin and a project manager for the Chief Health Policy Advisor to Milwaukee County government at the time of this writing. She has since graduated and started internal medicine residency at the University of Colorado.
Paul E. Pepe, MD, MPH is coordinator of the Metropolitan EMS Medical Directors (aka “Eagles”) Global Alliance and Public Safety/EMS Medical Director for Dallas County , Texas.
Benjamin W. Weston, MD, MPH is a professor at the Medical College of Wisconsin and chief medical director for the Milwaukee County Office of Emergency Management.
REFERENCES
1. Al Badawi AK, Nashwan AJ. The future of prehospital emergency care: Embracing AI applications in ambulance services. International Emergency Nursing. 2024;72doi:10.1016/j.ienj.2023.101385
2. Alrawashdeh A, Alqahtani S, Alkhatib ZI, et al. Applications and Performance of Machine Learning Algorithms in Emergency Medical Services: A Scoping Review. Prehosp Disaster Med. Oct 2024;39(5):368-378. doi:10.1017/S1049023X24000414
3. Elfahim O, Edjinedja KL, Cossus J, Youssfi M, Barakat O, Desmettre T. A Systematic Literature Review of Artificial Intelligence in Prehospital Emergency Care. Big Data and Cognitive Computing. 2025;9(9)doi:10.3390/bdcc9090219
4. Chee ML, Chee ML, Huang H, et al. Artificial intelligence and machine learning in prehospital emergency care: A scoping review. iScience. Aug 18 2023;26(8):107407. doi:10.1016/j.isci.2023.107407
5. Di Nardo Di Maio N, Carducci A, Dante A. Mapping artificial intelligence applications for clinical and operational decision support in prehospital emergency medical services: A scoping review. Int Emerg Nurs. Aug 6 2026;88:101887. doi:10.1016/j.ienj.2026.101887
6. Byrsell F, Claesson A, Ringh M, et al. Machine learning can support dispatchers to better and faster recognize out-of-hospital cardiac arrest during emergency calls: A retrospective study. Resuscitation. May 2021;162:218-226. doi:10.1016/j.resuscitation.2021.02.041
7. Zarei R, Downs MC, Torgerson L. Artificial Intelligence in Prehospital Emergency Care: Advancing Triage and Destination Decisions for Time-Critical Conditions. Cureus. 2025;doi:10.7759/cureus.91542
8. Kim JH, Kim MJ, Kim HC, Kim HY, Sung JM, Chang HJ. A Novel Artificial Intelligence-Enhanced Digital Network for Prehospital Emergency Support: Community Intervention Study. J Med Internet Res. Jan 23 2025;27:e58177. doi:10.2196/58177
