Presentation
No Handoff Left Behind: AI Support for Critical Care Transitions
DescriptionAs artificial intelligence (AI) continues to be developed and tuned, the healthcare industry has been increasingly focused on utilizing AI technologies to improve patient care and the efficiency of clinicians through decision-making support, patient data and diagnostics, and risk classification and prevention (Secinaro et al., 2021; Yu et al., 2018). More recently, researchers have proposed and examined the use of AI to assist in patient handoffs through means of generative reports (Landman et al., 2024; Genes et al., 2025) and clinical complication risk identification and predictions (Abraham et al., 2023; Xue et al., 2021).
A handoff is the transfer of care, responsibility, and liability of a patient or group of patients from one set of care providers to another (Desmedt et al., 2021). Effective handoffs are essential for high-quality patient care (Keebler et al., 2016); however, their effectiveness is limited by unstandardized handoff protocols, information omissions, inaccurate information, skill deficits, communication breakdowns, and inadequate training (Abraham et al., 2014). Despite recent research on AI use for inpatient-to-inpatient or emergency department (ED) to inpatient handoffs, there is currently a lack of literature examining AI in emergency medical service (EMS) to ED or inter-facility patient handoffs. EMS to ED handoffs deal with unique challenges, such as the variable and chaotic environment, stress levels, and interprofessional interactions (Troyer & Brady, 2020). Interfacility patient handoffs are also at greater risk of having poor patient outcomes due to patient status changes during transfer, poor transfer processes, lack of standardized tools between facilities, and interoperability issues (Galatzan et al., 2024). This is beyond the fact that transfers are typically done due to a complication or procedure that the current facility is not equipped to handle.
Various approaches of AI can, in theory, be utilized to support clinical handoffs in general; however, EMS and interfacility handoffs contain unique situations and challenges that may make the use of various models more difficult. For example, the loud environment in an ambulance would interfere with AI effectively performing tasks that rely on efficient speech recognition. Three primary AI model types would be suited for clinical handoff support. First, natural language processing (NLP) can analyze and understand human language in both spoken and written forms and then convert the data into a valuable and structured format (Zhou et al., 2022). Second, generative AI can synthesize information into a summative report that can be used during the handoff (Genes et al., 2025). Lastly, predictive models can be used to diagnose health issues and predict survival rates (Rong et al., 2020) as well as identify and stratify risk factors to aid in prevention for patients (Secinaro et al., 2021; Yu et al., 2018).
Gaps in AI utilization in EMS and interfacility handoffs are evident in both practice and literature. While there has been promise in utilizing AI models for various purposes within healthcare and even in some handoff settings, as previously discussed, this has not been examined in an EMS or interfacility context. Given the unique challenges each of these contexts faces, it is paramount to begin examining these unique AI use cases.
Challenges for the integration of AI systems should also be considered. Aung et al. (2021) detail several categories containing a variety of implementation challenges, including social factors such as trust in AI and misunderstanding, technology/development factors such as bias and data overfitting, data acquisition factors such as data availability and quality, implementation factors such as stakeholder buy-in and the current lack of evidence, and ethical factors such as privacy, safety, and accountability. While many of these factors have been explored in the AI literature across various domains and contexts, it has been underexplored within clinical handoffs, especially in the EMS to ED and interfacility contexts.
Future research on AI-assisted handoffs in the discussed contexts should evaluate whether they improve patient outcomes, such as mortality rates, length of stay, and the incidence of adverse outcomes, by having a higher quality handoff defined by factors including, but not limited to, information omission, inaccurate information, and handoff duration. AI integration challenges previously discussed should also be heavily researched. Although AI integration solutions in the literature would work for these transitions in theory, they must be tested extensively and adapted based on results. Furthermore, AI tools should be co-designed and tested with EMS personnel and frontline clinical staff to ensure a high-quality user experience (UX). Research should also be conducted to examine situational occurrences in these high-risk transitions, such as interoperability issues between organizations. Lastly, research into developing AI technologies should aim for any systems implemented to be that are capable of functioning offline, over extended durations, and in austere environments. While this goes beyond the day-to-day standards necessitated by many urban EMS providers, by ensuring a more robust system, this technology can be introduced into more rural and backcountry settings over time, in addition to being better equipped for mass-casualty, natural disaster, and other extreme urban scenarios where internet, radio, and cell phone signals may be impaired (Freeman et al., 2008).
AI-assisted handoffs may facilitate significantly higher quality handoffs for EMS and interfacility patient handoffs, reducing poor patient outcomes in addition to decreasing the mental workload (MWL) and stress of frontline staff. On the contrary, AI systems can also cause negative outcomes, such as biasing clinical providers and creating complacency for critical care decisions. Furthermore, if the handoffs in these settings remain overlooked in the general and AI literature, these care settings risk falling further behind in the quality of care they provide. By examining the gaps and challenges of AI use in EMS and interfacility handoffs, a high-risk sector, this work lays the groundwork for developing solutions to enhance the quality of care and patient safety in these settings.
A handoff is the transfer of care, responsibility, and liability of a patient or group of patients from one set of care providers to another (Desmedt et al., 2021). Effective handoffs are essential for high-quality patient care (Keebler et al., 2016); however, their effectiveness is limited by unstandardized handoff protocols, information omissions, inaccurate information, skill deficits, communication breakdowns, and inadequate training (Abraham et al., 2014). Despite recent research on AI use for inpatient-to-inpatient or emergency department (ED) to inpatient handoffs, there is currently a lack of literature examining AI in emergency medical service (EMS) to ED or inter-facility patient handoffs. EMS to ED handoffs deal with unique challenges, such as the variable and chaotic environment, stress levels, and interprofessional interactions (Troyer & Brady, 2020). Interfacility patient handoffs are also at greater risk of having poor patient outcomes due to patient status changes during transfer, poor transfer processes, lack of standardized tools between facilities, and interoperability issues (Galatzan et al., 2024). This is beyond the fact that transfers are typically done due to a complication or procedure that the current facility is not equipped to handle.
Various approaches of AI can, in theory, be utilized to support clinical handoffs in general; however, EMS and interfacility handoffs contain unique situations and challenges that may make the use of various models more difficult. For example, the loud environment in an ambulance would interfere with AI effectively performing tasks that rely on efficient speech recognition. Three primary AI model types would be suited for clinical handoff support. First, natural language processing (NLP) can analyze and understand human language in both spoken and written forms and then convert the data into a valuable and structured format (Zhou et al., 2022). Second, generative AI can synthesize information into a summative report that can be used during the handoff (Genes et al., 2025). Lastly, predictive models can be used to diagnose health issues and predict survival rates (Rong et al., 2020) as well as identify and stratify risk factors to aid in prevention for patients (Secinaro et al., 2021; Yu et al., 2018).
Gaps in AI utilization in EMS and interfacility handoffs are evident in both practice and literature. While there has been promise in utilizing AI models for various purposes within healthcare and even in some handoff settings, as previously discussed, this has not been examined in an EMS or interfacility context. Given the unique challenges each of these contexts faces, it is paramount to begin examining these unique AI use cases.
Challenges for the integration of AI systems should also be considered. Aung et al. (2021) detail several categories containing a variety of implementation challenges, including social factors such as trust in AI and misunderstanding, technology/development factors such as bias and data overfitting, data acquisition factors such as data availability and quality, implementation factors such as stakeholder buy-in and the current lack of evidence, and ethical factors such as privacy, safety, and accountability. While many of these factors have been explored in the AI literature across various domains and contexts, it has been underexplored within clinical handoffs, especially in the EMS to ED and interfacility contexts.
Future research on AI-assisted handoffs in the discussed contexts should evaluate whether they improve patient outcomes, such as mortality rates, length of stay, and the incidence of adverse outcomes, by having a higher quality handoff defined by factors including, but not limited to, information omission, inaccurate information, and handoff duration. AI integration challenges previously discussed should also be heavily researched. Although AI integration solutions in the literature would work for these transitions in theory, they must be tested extensively and adapted based on results. Furthermore, AI tools should be co-designed and tested with EMS personnel and frontline clinical staff to ensure a high-quality user experience (UX). Research should also be conducted to examine situational occurrences in these high-risk transitions, such as interoperability issues between organizations. Lastly, research into developing AI technologies should aim for any systems implemented to be that are capable of functioning offline, over extended durations, and in austere environments. While this goes beyond the day-to-day standards necessitated by many urban EMS providers, by ensuring a more robust system, this technology can be introduced into more rural and backcountry settings over time, in addition to being better equipped for mass-casualty, natural disaster, and other extreme urban scenarios where internet, radio, and cell phone signals may be impaired (Freeman et al., 2008).
AI-assisted handoffs may facilitate significantly higher quality handoffs for EMS and interfacility patient handoffs, reducing poor patient outcomes in addition to decreasing the mental workload (MWL) and stress of frontline staff. On the contrary, AI systems can also cause negative outcomes, such as biasing clinical providers and creating complacency for critical care decisions. Furthermore, if the handoffs in these settings remain overlooked in the general and AI literature, these care settings risk falling further behind in the quality of care they provide. By examining the gaps and challenges of AI use in EMS and interfacility handoffs, a high-risk sector, this work lays the groundwork for developing solutions to enhance the quality of care and patient safety in these settings.
Event Type
Oral Presentations
TimeMonday, March 2311:30am - 12:00pm EDT
LocationMurray Hill West
Hospital Environments