Emergency Medical Services in New York and the Potential of Leveraging Artificial Intelligence
- Sally Dreslin
- Apr 6
- 22 min read
Issue Brief by Sally Dreslin
PDF Available:
Introduction
The use of artificial intelligence (AI) has rapidly expanded across industries in recent years, transforming operations, decision-making processes, and user experiences. From automation and predictive analytics to generative models produced by OpenAI, Google, and Anthropic, among others, enterprises across sectors are increasingly utilizing AI to gain efficiency and pursue innovation. In healthcare, some examples of areas in which AI is being applied include diagnostic imaging, clinical decision-making, patient medical records, robotics, supply chain management, and administrative processes, including clinical documentation and prior authorization. The largest electronic health record (EHR) vendors, Epic and Oracle, are incorporating AI into their products to offer their healthcare customers the opportunity to seamlessly access AI tools. Despite these developments, the integration of AI in healthcare also presents challenges and risks. Concerns surrounding patient privacy, data security, algorithmic bias, and potential errors in automated clinical decision-making underscore the need for careful oversight, transparent practices, and ongoing validation to ensure patient safety and to maintain trust in the tools. As AI and its uses continue to evolve, navigating these complexities will be essential to realize its transformative potential.
In healthcare, a significant challenge for many hospital systems, nursing homes, community clinics, emergency medical services (EMS) agencies, and other providers in adopting AI tools is the information technology infrastructure that they have quilted together over time. Apart from the most well-resourced and sophisticated healthcare delivery systems, many healthcare providers work with a patchwork data and information infrastructure composed of siloed and legacy systems that are uncoordinated, contain large amounts of unstructured data, and often rely on one or two staff who are the only people familiar with the underlying structure. Many providers lack the financial and human resources to develop and maintain a modern health information technology infrastructure with responsible governance and cybersecurity. With this foundation, it can be difficult and costly to leverage new tools such as AI, but by taking an incremental and shared approach, progress in data and infrastructure readiness can be achieved.
The Step Two Policy Project has previously discussed the challenges of democratizing access to AI and the importance of preventing a widening chasm between the “haves” and the “have-nots” within New York’s healthcare delivery system. In our Issue Brief called Creating a Healthcare-Focused Artificial Intelligence Public-Private Platform for New York State, we assert that the potential benefits from the evolution of AI tools in healthcare will be constrained if important parts of the healthcare delivery system are left out because they lack financial resources and/or management capacity. This constraint will disproportionately affect smaller healthcare organizations, including safety-net and rural hospitals, community clinics, EMS, and others that may lack the resources to implement secure AI functionality or benefit from AI-enabled electronic health records used by larger, better-resourced institutions. There are numerous HIPAA-compliant, healthcare-focused applications and platforms available, but they can be costly to purchase, implement, and monitor.
The purpose of this Issue Brief is to review the current and potential uses of AI in healthcare. Rather than examining this in the hospital setting, which is already the focus of significant attention, this Brief will identify some ways in which AI tools might be helpful in the prehospital care delivery system. We will examine EMS use cases such as:
Dispatch, triage, and clinical decision support
Emergency supply delivery using drones
Quality improvement
Documentation
Personnel training
Pre-positioning of emergency equipment in the community
Improving prehospital stroke recognition
New York’s Emergency Medical Services System
Emergency medical services in New York State are currently delivered through a system made up of public, private, nonprofit, municipal, career, volunteer, and mixed agencies. According to data from 2022, there were approximately 2.9 million EMS 911 responses across the state, with approximately 2 million responses resulting in transport. The NYS Comptroller, in the report issued from his office in 2024 called The Growing Role of Counties in Emergency Medical Services, notes that EMS is coordinated at the local, regional, and state levels. Response times across the state have gotten longer, as evidenced by data from a variety of regions[1],[2] but there is little statewide data related to EMS in general. The system lacks consistent information related to response times, ambulance availability, staffing levels, and mutual aid-dependence is often unavailable or difficult to obtain, making it hard for local officials to assess whether service is adequate.
The EMS system in New York is a system under increasing stress. According to the 2024 Update on the EMS Workforce Shortage report from the New York State Emergency Medical Services Council (SEMSCO),[3] the number of active certified EMS responders in New York fell by 17.5 percent from 2019 to 2022, and both career and volunteer agencies are struggling to recruit and retain enough responders to provide around-the-clock coverage. The same report also shows that the number of ambulance agencies in New York declined by nine percent over the last decade, from 1,078 to 982, with 54 percent not-for-profit, 37 percent municipal, and nine percent for-profit.[4]
A decline in volunteers and low wages relative to the requirements for and responsibilities of EMS jobs are significant factors in the challenges of EMS personnel recruitment and retention.[5],[6] For example, based on a SEMSCO survey conducted for the 2024 Report, 51.6 percent of EMTs earn less than $49,000/year, with 24.5 percent earning less than $39,000/year, and 34.6 percent of paramedics earn less than $69,000/year, with 18 percent earning less than $59,000/year. In New York City specifically, the Daily News reports that “one-third of NYC EMS workers plan to leave their job” due to low pay. This is partly a function of the pay disparity between firefighters and EMS responders within the NYC Fire Department (FDNY). A FDNY EMT starts at a salary of $39,386, and after five years, the salary increases to $45,196. This compares to FDNY firefighters’ starting salary of $45,196 and a wage of $110,000 after five years.
Funding challenges for EMS agencies contribute to the current strain on the system in New York. The Rural Ambulance Services Task Force found that EMS reliability is at risk because of insufficient funding for escalating operating costs, insurance reimbursement, and “readiness,” or standby costs, i.e., the cost of maintaining staff and ambulances available between calls when no transport revenue is being generated. The Task Force further concluded that the state’s reliance on a hodgepodge of local approaches (including local and siloed 911 dispatch centers) has produced fragmented service delivery, with some municipalities contracting for coverage, some relying on volunteers, and others lacking formal coverage arrangements altogether.
Governor Hochul has made several efforts to expand the scope of practice and increase the flexibility of the EMS workforce with mixed results, increase the range of reimbursable services— such as reimbursement for treating in place without a transfer to a hospital, or transfers to alternative healthcare settings other than hospitals— and establish EMS as an “essential service” in New York, as police and fire departments are. The effort to establish EMS services as an “essential service,” which would have required counties and municipalities to guarantee EMS services are available to residents through a variety of strategies, was not fully successful. However, the Governor signed legislation and chapter amendments this past year to require New York counties, in coordination with regional EMS councils and local governments, to develop and maintain comprehensive county emergency medical system plans. The goal is to better coordinate and prepare EMS coverage to ensure that reliable emergency medical and ambulance services are available throughout the state.
Current Uses of AI in Emergency Medical Services
Despite recent growth in attention and acceleration in the use of AI in healthcare, including in EMS, AI has played a part in healthcare delivery for decades. Research over the past several years has examined the scope and effectiveness of using AI in emergency medical services. Areas of interest and evaluation include: dispatch, triage, and prehospital decision support; prehospital documentation; scene surveillance and delivery of emergency supplies; and quality improvement.
For an overview, M.L. Chee and colleagues from Australia and Singapore provide a scoping review of artificial intelligence and machine learning in prehospital emergency care. They found that the primary applications of AI in prehospital emergency care include diagnostic and predictive models, dispatch optimization, and resource allocation. AI models, particularly neural networks,[7] have demonstrated strong predictive performance, aiding rapid, accurate triage decisions and identification of high-risk patients in trauma and cardiac arrest cases. For instance, one of the studies reviewed demonstrated the effectiveness of AI-based predictive models utilizing heart rate variability and vital signs to identify trauma patients who may require additional interventions. Similarly, AI has been used to predict outcomes after out-of-hospital cardiac arrest to guide early intervention and improve survival prospects.
This graphic from their 2023 study shows the AI use cases examined in the publications included in the scoping review:
Despite the benefits of AI technology in prehospital emergency care, many challenges exist. These include concerns related to privacy and data security, particularly in the context of personal health information, and complexity related to varying types of data sources such as ECGs, ultrasound images, and body-worn cameras. For example, one of the publications reviewed demonstrated improved accuracy in predicting the outcomes of defibrillation using the integration of exhaled CO2 and ECG data. However, integrating these data modalities on a broad scale remains challenging due to technical limitations, underscoring the need for continued advancements in interoperability to realize greater utility in prehospital care.
A more recent scoping review published by Mallon and colleagues in 2025 identifies three broad categories of AI use cases in EMS. These include studies that focus on the use of AI: 1) to categorize different records or calls, 2) for dispatch forecasting and coordination, and 3) to predict a specific illness or injury. Their paper provides an interesting graphic that visually represents how AI tools can fit into the prehospital emergency care system, based on studies they reviewed. An important topic for further consideration is facilitating better integration of prehospital patient care data and information with hospital-based data and information. In the graphic below, this would mean bringing some of the orange lines across the vertical dashed line that separates the prehospital environment from the facility environment.
Another overview, albeit on a more focused aspect of EMS, is provided by J. Hsueh and colleagues in the United States and the United Kingdom. They provide a review of literature focused on AI applications in helicopter emergency medical services (HEMS) to identify the current state of “the HEMS AI evidence base.” This scoping review included 21 studies published between 2006 and 2023, covering four basic AI use cases – clinical focus, HEMS, landing zone, and vital signs processing.
For example, the Ateyo studies from 2006 and 2008 used AI tools to assess clinical and logistics/dispatch factors to assist HEMS crews in making mission “go/no-go” decisions. The authors concluded that “intelligent systems” were especially useful in processing the many factors involved in calculating missions’ clinical benefits versus the aviation risks.[8],[9]
Additional clinical use cases in the review included AI tools for the assessment of vital signs to predict major hemorrhage, the need for specific emergency clinical interventions, and the mortality risk of patients being transported between facilities, as well as to determine when patients should not be considered for HEMS transfer.
Non-clinical AI use cases discussed in Hsueh et al.’s paper involve tools and technology used in identifying helicopter landing zones in wilderness emergency response, and identifying landing zones such as soccer fields, in populated areas, assisting in the forward deployment of supplies and aircraft, and in HEMS pilot evaluation and assistance.
As discussed above, the HEMS scoping review discusses several general limitations to applying AI in prehospital emergency care. These include limited availability of clinical data needed to power AI models beyond the current non-AI approaches; difficulty accessing patients' electronic health records (EHRs) across different settings, which is important for developing more effective AI models; and the lack of research on non-trauma transfer patients, which is necessary for large AI models to learn differences in outcomes based on the type and timing of transfer.
Dispatch, Triage, and Prehospital Decision Support
Considerations when using AI tools in EMS dispatch, triage, and prehospital decision support include the unpredictability of cellular networks, especially in rural areas; integration with existing dispatch and response systems; challenges with coordination and communication of responses, particularly with agencies that have volunteers responding from their homes; and maintaining the ongoing functioning of the AI system, particularly for smaller agencies.[10]
In 2023, the Artificial Intelligence-Facilitated Emergency Medical Services Call Center Software Market Survey Report, the National Urban Security Technology Laboratory, in conjunction with the Operational Experimentation Program for the U.S. Department of Homeland Security, Science and Technology Directorate, described some approaches and the usefulness of incorporating AI tools into EMS call centers. Their report explains that AI-enabled software for EMS call centers can offer decision support to help staff assess a caller’s condition and make real-time recommendations about prehospital care and field response. It describes how the software integrates and analyzes data from live calls, including speech and background noise, and compares those inputs with a large set of previous emergency call information to identify likely needs and suggest relevant actions. They emphasize that the technology is designed to improve speed, accuracy, and efficiency in emergency response coordination by supporting staff, rather than replacing human dispatch and triage personnel.
Aligning with this approach of using AI tools in EMS to support, rather than replace, practitioners, the University of Pittsburgh Medical Center EMS Department developed a machine learning tool that analyzes hundreds of features of ECGs to “help EMS teams identify conditions like cardiac ischemia or blockages in the blood vessels.” They use the AI tool to support human interpretation of ECGs, since the algorithms can “analyze and interpret a larger number of data features, including those that may not be observable with the naked eye.” They plan to create a dashboard to gather and share the ECG information that the algorithm is identifying.
Emergency Supply Delivery Using Drones
As we see with many AI applications in emergency medical services, the possibilities are evolving more quickly than the capabilities. Roberts and colleagues explain in the review excerpted below,
“Drones offer an exciting possibility to reduce the time-to-intervention and improve patient outcomes for time-dependent medical emergencies that are beyond the rapid arrival of traditional EMS. However, significant research remains before drones are likely to realize their full potential and achieve widespread adoption, including further research demonstrating functionality in real-world scenarios and guidance on how to operationalize broad integration into the EMS system.”
The following is an example of an AI use case in the context of emergency supply delivery:
“Medical indications for using unmanned aerial vehicles, or drones, are rapidly expanding, including the delivery of time-sensitive medical supplies. To date, the drone-based delivery of a variety of time-critical medical supplies has been evaluated, generating promising data suggesting that drones can improve the time interval to intervention through the rapid delivery of automatic external defibrillators (AEDs), naloxone, antiepileptics, and blood products. Furthermore, the improvement in the time until intervention offered by drones in out-of-hospital emergencies is likely to improve patient outcomes in time-dependent medical emergencies. However, barriers and knowledge gaps remain that must be addressed. Further research demonstrating functionality in real-world scenarios, as well as research that integrates drones into the existing EMS structure will be necessary before drones can reach their full potential. The primary aim of this review is to summarize the current evidence in drone-based Emergency Medical Services Care to help identify future research directions. …
“Admittedly, there are a variety of non-medical barriers to EMS drone implementation whose full analysis is beyond the scope of this review. Within the United States, Federal Aviation Administration (FAA) regulations, security concerns, funding streams, and costs and requirements related to licensure, training, and insurance, will need to be addressed prior to widespread adaptation into existing EMS systems.”
Quality improvement
AI tools may offer opportunities to identify patterns, reduce errors, and accelerate quality management and improvement activities. However, alongside these potential benefits, agencies must be cognizant of AI tools’ limitations in navigating unstructured or non-standard data. The following excerpt is from 2024, and given the significant advances in the functionality of LLMs over the past two years, it does not seem unreasonable to believe that these enhancements have made their use considerably more valuable in the context of quality improvement and improvement activities:
“This study assesses the feasibility, inter-rater reliability, and accuracy of using OpenAI’s ChatGPT-4 and Google’s Gemini Ultra large language models (LLMs), for Emergency Medical Services (EMS) quality assurance. The implementation of these LLMs for EMS quality assurance has the potential to significantly reduce the workload on medical directors and quality assurance staff by automating aspects of the processing and review of patient care reports. This offers the potential for more efficient and accurate identification of areas requiring improvement, thereby potentially enhancing patient care outcomes. …
“Large language models demonstrate potential in supporting quality assurance by effectively and objectively extracting data elements. However, their accuracy in interpreting non-standardized and time-sensitive details remains inferior to human evaluators. Our findings suggest that current LLMs may best offer supplemental support to the human review processes, but their current value remains limited. Enhancements in LLM training and integration are recommended for improved and more reliable performance in the quality assurance processes.”
Opportunities for the Use of AI in Emergency Medical Services Going Forward
There are numerous emerging use cases for AI tools in emergency medical services that will depend on the reliable evolution of AI technology and improvements in data coordination and integration. A few emerging use cases for emergency medical services are discussed below.
Documentation
Increasingly, AI tools are being incorporated into electronic health record systems, such as Epic and Oracle ,[11] that are used in hospitals. We are also seeing an acceleration in the uptake in the hospital setting of AI-enabled, ambient scribes that leverage natural language processing to assist in drafting real-time clinical notes when providers are interacting with patients. The impact of using these tools is reported to include a reduction in the amount of time providers spend charting after the fact, more complete and accurate clinical documentation, improved patient and provider experience, and increased efficiency in workflow.
In the prehospital environment, there are applications that assist EMS responders in generating electronic patient care reports (ePCR) to document their findings and interventions, which are shared with hospital emergency department staff. These are essentially digitized versions of the paper process. As occurs in the hospital environment, when paramedics are treating a critical patient or are in a chaotic situation, charting is sometimes completed after the patient encounter, which can lead to incomplete or inaccurate after-the-fact documentation. Hedderson and colleagues published a scoping review in 2025 to “determine paramedic perceptions and user requirements for speech recognition documentation technology.” The authors note that the findings related to “speech recognition” documentation in the hospital setting may not be fully transferable to the prehospital care environment. They explain that “challenges such as ambient noise, uncontrolled care environments, and inclement weather mean that a technology suitable for” a hospital setting may not function well in the prehospital setting. They do state, however, that the advancements in “natural language processing, deep neural networks, speech synthesis, and pre-processing to improve word recognition rates and reduce noise interference” could address many of these prehospital challenges.
Hedderson et al.’s scoping review found limited use of speech recognition documentation technology in prehospital environments, but the majority of recommendations from the studies reviewed were to move forward with “live environment testing.”
Also in 2025, Bai and colleagues published a study to investigate whether “recent” LLMs (i.e., OpenAI’s GPT 4o, Google’s Gemini, Meta’s LLaMA, and Anthropic’s Claude 3.5) could process a transcript of an EMS simulation recording to extract information. The goal for the LLM was to identify clinical information in the transcript that could then be used to produce automated EHR documentation in the prehospital environment. Their findings identified several challenges and limitations related to accuracy but confirmed the potential benefits of “automating EMS EHR documentation” in enhancing the completeness of prehospital documentation and improving the continuity of care. Importantly, their work included mapping the EMS data to Fast Healthcare Interoperability Resources (FHIR), the current standard framework for exchanging electronic health information, thus laying the foundation for better integration between prehospital documentation and hospital-based EHRs.
Related to prehospital documentation, it is important to note that the HIPAA Journal, in a post from January 2025, stated that, “organizations that are required to comply with the Health Insurance Portability and Accountability Act (HIPAA) are not permitted to use these tools [generative AI] in connection with any ePHI [electronic personal health information] unless the tools have undergone a security review and there is a signed, HIPAA-compliant business associate agreement in place with the provider of the tool.”
EMS Personnel Training
One future area for the application of AI is for training EMS personnel. Training scenarios using generative AI to simulate real-life scenarios can provide “opportunities to cultivate critical thinking skills, refine decision-making abilities, and enhance clinical competencies through experiential learning.”[12] Ideally, an AI-facilitated training simulation could create scenarios, adjust the simulation as the trainee delivers interventions, and provide feedback and guidance in real-time. While this is an exciting future opportunity, there continues to be concern related to AI inaccuracies. As discussed above, there is insufficient clinical training data for EMS AI applications to learn from, and this can lead to inaccurate and biased outputs from AI models.
Improving Access to AEDs for Out-of-Hospital Cardiac Arrest
As we see from the discussion above, Chee et al.’s scoping review of AI in prehospital emergency care reviewed several studies focused on automatic external defibrillator (AED) optimization and positioning. An emerging use case for AI in EMS is optimizing the positioning of AED drones to facilitate rapid defibrillation for patients experiencing an out-of-hospital cardiac arrest (OHCA).[13] The purpose of a study by Mackle and colleagues was to “… link real-world heterogeneous datasets to build a system to determine the difference in emergency response times when having aerial ambulance drones available compared to response times when depending solely on traditional ambulance services and lay rescuers who would use nearby publicly accessible defibrillators to treat OHCA victims.”
The Delft University of Technology in the Netherlands created a prototype “ambulance drone” that carries a lightweight AED and includes a built-in camera and a speaker to enable communication between an emergency operator and a person at the site of care. Mackle et al. developed a computer simulation using real-world data of the occurrence of OHCAs in Northern Ireland to identify optimal locations to base AED drones. They then compared response times between AED drones, ambulances, and bystanders with a public AED. The results of the simulation showed, “… significant improvements in response times for OHCA incidents when implementing a drone network. The drones also offer a dedicated response to these incidents and a shorter response time for rural incidents, which is likely to improve the chances of survival for many cases.”
"Health and social care trusts" are the coordinated, regional providers of health and social services in Northern Ireland.
There are several real-life, non-simulation implementation issues to consider, e.g., the cost of the drones is significant, the AED drone technology is still developing, there are no-fly zones to take into account, battery time for the drones has limits, nearby landing zones may not always be available, and weather will impact flying ability. As the authors explain, “benefiting from UAVs [drones] in OHCA occurrences, apart from technology, requires community or governmental ownership and consequently laws and regulations, training protocols, licensing, insurance and social awareness.”
Improving Prehospital Stroke Recognition
Scholz and colleagues explored whether an automatic speech recognition (ASR) software could improve the recognition of stroke by the Emergency Medical Dispatcher (EMD) and by EMS Copenhagen, in Denmark, and subsequently improve stroke treatment. They reviewed data from the Danish Stroke Registry for patients within the Copenhagen area, from 2016-2018. The stroke registry data were joined with EMS stroke data and with the individual Danish citizen ID data from the EMS database. After several layers of categorization and analyses, they concluded, based on the retrospective data, that ASR technology could improve the recognition of stroke for EMS Copenhagen, “… specifically among females, younger stroke patients, within the 1813-Medical Helpline, and on weekends. Under consideration of the beforenamed conditions and limitations, an ASR could have a positive effect on stroke detection, and thereafter on stroke treatment, specifically on thrombolysis.”[14]
Wolcott and English, in their paper Artificial Intelligence to Enhance Prehospital Stroke Diagnosis and Triage: A perspective, created the figure below to identify applications of several different types of AI technologies that could potentially improve prehospital stroke identification and triage:
Considerations with the Use of AI in Emergency Medical Services
The integration of artificial intelligence technology into emergency medical services presents significant opportunities to enhance patient care, but it also raises important concerns regarding privacy, security, accuracy, bias, and transparency that must be thoughtfully considered.
In March 2025, the NYS Office of Information Technology Services (ITS) updated its policy on the Acceptable Use of Artificial Intelligence Technologies. The policy outlines guidelines for the use of AI systems by NYS entities; it does not address private or non-state, municipal EMS agencies. AI systems are defined broadly in the policy as machine-based systems capable of making predictions, recommendations, or decisions that directly impact the public. The NYS ITS policy emphasizes human oversight, requiring that final decisions affecting the public always be made by appropriate staff members. While the policy is intended for state entities, it offers important insights for other entities, such as EMS providers, regarding responsible AI implementation and oversight.
State entities must ensure robust human oversight when AI systems influence decisions that impact the public; completely automated final decision-making systems are explicitly prohibited. The NYS ITS policy also requires regular assessments of outcomes and methodologies, along with periodic evaluations for reliability, safety, and fairness. Additionally, the policy emphasizes fairness and equity, in accordance with applicable State and Federal laws, rules, and regulations. Entities must actively monitor for and address any systemic, computational, or human biases in AI systems.
State entities are required to have policies and controls in place to ensure appropriate privacy protections, especially when the AI system is processing personally identifiable, confidential, or sensitive information. Policies and controls may include privacy impact assessments, adherence to “data minimization” practices,[15] appropriate data retention and disposal practices, confirmation of the accuracy of data inputs and AI outputs, and transparency related to data processing. In all cases, as applicable, the policies and controls should align with standards outlined in frameworks such as the NIST AI Risk Management Framework.
In 2024, the Cybersecurity & Infrastructure Security Agency (CISA) of the U.S. Department of Homeland Security released a brief entitled Artificial Intelligence and the Emergency Services Sector – Benefits and Challenges. The brief identifies several limitations and challenges in the use of AI in the emergency services sector, including “… hiring trained professionals, gathering high-quality data, conducting training sessions with the data, and adapting the obtained results.” Additionally, the document outlines considerations involving the quantity and quality of data needed to make accurate predictions, costs of implementation and support of AI technology, integration of AI tools with existing systems, the responsibility and accountability related to decisions produced by AI algorithms, and ethical and legal considerations, including privacy, data protection, and liability.
Best Practices in Developing a Policy on the Use of AI in EMS
There are several components that should be included when developing policies to guide the use of AI tools in healthcare in general, and in the context of EMS specifically. As discussed in the NYS ITS Acceptable Use of Artificial Intelligence Technologies policy above, concerns related to human oversight, bias, accuracy, ongoing monitoring, and transparency of process are all essential elements to address in an AI use policy. Also critical are issues related to privacy and security. Given the current state of rapidly expanding access to generative AI tools such as Claude, ChatGPT, and others, it is important that policies related to the use of AI in healthcare and EMS include a discussion of publicly accessible open AI tools, in addition to the healthcare-specific HIPAA-compliant tools that are also available. Specifically, “… policies for publicly accessible generative AI … might prioritize data de-identification to protect individual privacy, while policies focused on addressing the use of the AI systems with robust safeguards in place, may not require the same measure.”[16]
To develop robust and implementable policies, it is essential to include a broad range of stakeholders in the process. Recommended policymaking stakeholders include EMS medical directors and field personnel; emergency dispatch personnel; healthcare quality professionals; AI technologists; members of the public; public health experts; and experts in healthcare legal, regulatory, and ethical considerations, among others.
The Kansas Health Institute released a resource for public health organizations to assist in the development of AI policies. The document is divided into sections that address:
Data Privacy
Bias Mitigation
Human Oversight
Transparency
Community Engagement
Training and Capacity Building
Each section includes the purpose of the topic, hypothetical vignettes, an overview of sample policy provisions, and sample provisions that can be used to draft specific policies. This is a visual representation of the document’s structure:
In addition to the topic-specific sections, this resource assists organizations in identifying important considerations prior to developing an AI policy, crafting a purpose statement, establishing key principles, and determining the scope of their AI policy.
Conclusion
The availability and adoption of AI tools have grown rapidly across a wide range of sectors, reshaping operations, decision-making, and user engagement and experience. However, the integration of AI into healthcare can bring significant challenges. Concerns regarding patient privacy, data security and governance, algorithmic bias, and the risk of errors in automated decision-making highlight the need for strong oversight, transparency, and ongoing validation of accuracy to safeguard patient safety and uphold trust. Implementing strong governance and risk management are key elements. As AI technologies continue to evolve, it will be critical to navigate these complexities in order to fully realize their potential benefits.
Artificial intelligence offers new tools that can be applied to many existing challenges. The potential for AI to strengthen EMS in NYS is significant, but its broader application is constrained by the current limitations of the EMS data environment. Currently, there is no publicly available, standardized, and comprehensive statewide dataset on EMS operations and patient outcomes, and there is limited ability to track patients longitudinally from prehospital care through hospital discharge. The Step Two Policy Project has often referred to the importance of “democratizing the availability of health data, information, and analysis.” Publicly available data, appropriately curated, is an essential component in achieving this goal. More fundamentally, EMS remains insufficiently integrated into the State’s broader healthcare delivery system. While some well-resourced hospital systems have developed robust data capabilities for their own patient populations, potentially enabling emergency personnel to access important information about a patient at the time of an emergency, the statewide picture remains fragmented. This limits the availability of the data needed to support many of the most promising AI applications.
As the illustration on the first page of this Issue Brief portrays, maximizing the potential of AI tools depends heavily on the data and infrastructure that support them. In the current context, certain uses—such as dispatch, emergency supply delivery, and clinical functions like ECG interpretation—may be well-positioned to benefit from AI assistance. Ambient documentation tools are increasingly being adopted in hospital and outpatient settings and are producing benefits, but as we see from the research, development lags in EMS due to insufficient training data and the unpredictable environment of care. Most emergency and prehospital clinicians (or other clinicians, for that matter) do not go into the field to spend their time on documentation, either at the time of care or after the fact. Easing the burden of documentation in EMS will enhance both patients’ and providers’ experiences. More advanced AI use cases, including clinical predictive analytics, triage decision-making, quality improvement, and personnel training, will require New York State to build a stronger foundation of standardized clinical and operational data that is better coordinated with the rest of the healthcare system.
NYS can be a leader in enabling the effective use of AI in healthcare while at the same time protecting patients and the healthcare delivery system. By remaining focused on patient-centered goals when pursuing AI functionality, creating clear but adaptable (as evolution occurs) guardrails, and collaborating with public and private stakeholders, NYS can build a solid framework for the responsible use of AI in healthcare.
Endnotes
[1] Critical condition: Why are NYC ambulance response times getting longer every day? 2025. AMNY Newsletter.
[2] The growing role of counties in emergency medical services. 2024. NYS Comptroller Thomas P. DiNapoli.
[3] Disclosure: I am a member of SEMSCO, in the “General Public” seat. I was not involved in the drafting of the workforce report. This Issue Brief does not represent the views of SEMSCO or the NYS DOH. It is written from the perspective of my role with the Step Two Policy Project.
[4] Ambulance offload 2019-2022 data. 2023. Bureau of Emergency Medical Services and Trauma Systems, NYS Department of Health.
[5] Report on rural ambulance services. 2025. The New York State Rural Ambulance Services Task Force.
[6] 2024 Update on the EMS Workforce Shortage. 2024. NYS Department of Health State Emergency Medical Services Council.
[8] Atyeo, S., Sinha, A, Young, K. A Decision Support System for Helicopter EMS Operations. September3-8. In: Poster presented at: International Congress of the Aeronautical Sciences. Hamburg, Germany; 2006.
[9] Atyeo, S. An Intelligent System for the Pre-Mission Analysis of Helicopter EMS Operations. Melbourne, Australia: Royal Melbourne Institute of Technology; 2008 [PhD dissertation].
[11] A look at Epic and Oracle Health’s push to reinvent the EHR, 2025. Becker’s Health IT.
[12] Artificial intelligence chatbots and emergency medical services: Perspectives on the implications of generative AI in prehospital care, 2023. Open Access Emerg Med.
[13] A data-driven simulator for the strategic positioning of aerial ambulance drones reaching out-of-hospital cardiac arrests: A genetic algorithmic approach, 2020. IEEE Journal of Translational Engineering in Health and Medicine.
[14] Artificial intelligence in emergency medical services dispatching: Assessing the potential impact of an automatic speech recognition software on stroke detection taking the capital region of Denmark as case in point, 2022. Scand J Trauma Resusc Emerg Med.
[15] “Data Minimization is a key privacy principle to respecting individual privacy and reducing risks associated with a data breach. Under this principle, only personal information that is directly relevant and necessary to a specified purpose is collected and kept for only as long as needed for that purpose.” Data Minimization, 2022. WaTech.
