Presentation
HE10 - Human-AI Teams in Critical Care: An IMOI Model for Effective Collaboration
SessionPoster Session 1
Description(AI) systems have emerged as invaluable tools to assist humans in a range of tasks in critical care settings where every second counts . AI has shown promise in enhancing clinical decision-making, improving diagnostic accuracy, and supporting patient monitoring through analyzing vast amounts of data in real-time, reducing clinician burden, and providing predictive insights. This is particularly valuable in high-stakes environments like critical care where timely interventions are essential (AHA, 2023; Fulga et al., 2023; Yu et al., 2022). However, the successful integration of AI into human teams requires careful consideration of human factors, such as trust, communication, and interface design, to ensure seamless collaboration under high-pressure conditions. This presentation explores how AI assists in critical care decisions and how human factors science can foster more effective human-AI teams.
We propose an IMOI (Input-Mediator-Output-Input) model, specifically tailored to human-AI teams in critical care. The IMOI (Input-Mediator-Output-Input) model originated from organizational psychology and has been applied to various team dynamics research to understand how team processes and outcomes evolve. The model has since been adapted in fields like healthcare, offering a way to examine the interaction between team composition, processes such as communication and trust, and the resulting team performance, with feedback loops to inform future improvements (Ilgen et al., 2005; Marks et al., 2001). The model provides a comprehensive framework for understanding how various factors impact team dynamics and performance.
Input. Inputs would include the initial conditions, such as the team's composition (clinicians, nurses, AI systems), task complexity, and patient data inputs that the AI processes (e.g., vital signs, imaging results). Examples include how AI-driven alerts about a patient’s deteriorating condition can be fed into the team’s decision-making process (Zhang et al., 2020).
Mediator. Mediators include team processes such as communication, coordination, and trust between human team members and the AI system. For instance, trust in AI has been shown to influence the extent to which clinicians rely on its suggestions or override them (Lyell & Coiera, 2021). Poor interface design can reduce this trust, leading to hesitation or over-reliance on AI, both of which can be detrimental in critical situations (Endsley, 2017).
Output. Outputs would include team outcomes focused on task performance, including the quality and speed of critical care interventions. Studies indicate that well-designed human-AI teams can improve patient outcomes, especially when the AI is seamlessly integrated into workflows (Yu et al., 2022).
Input (Feedback Loop): The final input refers to the feedback loop where the outcomes of the team’s performance influence future decisions, team training, and AI system improvements (e.g., learning from previous cases or incorporating new AI algorithms).
This topic is particularly important because it is still understudied in both research and practice. Much of the existing literature on human-AI collaboration has focused on domains such as autonomous driving or customer service (Guzman & Lewis, 2020), leaving a gap in understanding how AI affects team processes, emergent states, and performance in critical care settings. Some key takeaways from this presentation will include insights into the role of trust and communication, user interface design, and training in effective human-AI collaboration.
First, trust in AI systems is crucial for effective collaboration, yet it is often overlooked. Research suggests that human teams may either overly rely on or underutilize AI systems based on trust levels (Lyell & Coiera, 2021), an issue that becomes critical in time-sensitive, high-stakes environments like critical care (Endsley, 2017). Second, the presentation of AI in human-AI teams will also emphasize the need for intuitive AI interfaces. Poor design can lead to misinterpretations or delays in decision-making, which can have serious consequences in emergency medical situations (Yu et al., 2022). Finally, proper training is essential to ensure that clinicians and staff understand how to collaborate effectively with AI systems. This includes understanding AI's limitations and strengths to avoid over-reliance or mistrust as current training approaches do not adequately address these needs, making it an area for further development (Guzman & Lewis, 2020). Overall, the findings and model we propose will help pave the way for improved integration of AI into human teams in critical care, enhancing decision-making processes and ultimately improving patient outcomes.
References
AHA. (2023). How AI is improving diagnostics, decision-making, and care. American Hospital Association.
Endsley, M. R. (2017). Autonomous driving systems: Trust, driver behavior, and performance. Human Factors, 59(1), 52-67. https://doi.org/10.1177/0018720816659950
Fulga, A., Neagu, M., & Piraianu, A. I. (2023). Advancing patient care: How artificial intelligence is transforming healthcare. Journal of Personalized Medicine, 13(8), 1214. https://doi.org/10.3390/jpm13081214
Guzman, A. L., & Lewis, S. C. (2020). Artificial intelligence and communication: A review of AI applications in human communication research. Journal of Communication, 70(2), 178-197. https://doi.org/10.1093/joc/jqz028
Ilgen, D. R., Hollenbeck, J. R., Johnson, M., & Jundt, D. (2005). Teams in organizations: From Input-Process-Output Models to IMOI Models. Annual Review of Psychology, 56, 517-543. https://doi.org/10.1146/annurev.psych.56.091103.070250
Lyell, D., & Coiera, E. (2021). Automation bias and verification complexity: A systematic review. Journal of the American Medical Informatics Association, 28(5), 986-995. https://doi.org/10.1093/jamia/ocaa338
Marks, M. A., Mathieu, J. E., & Zaccaro, S. J. (2001). A temporally based framework and taxonomy of team processes. Academy of Management Review, 26(3), 356-376. https://doi.org/10.5465/amr.2001.4845785
Yu, K. H., Beam, A. L., & Kohane, I. S. (2022). Artificial intelligence in healthcare. Nature Biomedical Engineering, 6(3), 133-143. https://doi.org/10.1038/s41551-021-00781-z
We propose an IMOI (Input-Mediator-Output-Input) model, specifically tailored to human-AI teams in critical care. The IMOI (Input-Mediator-Output-Input) model originated from organizational psychology and has been applied to various team dynamics research to understand how team processes and outcomes evolve. The model has since been adapted in fields like healthcare, offering a way to examine the interaction between team composition, processes such as communication and trust, and the resulting team performance, with feedback loops to inform future improvements (Ilgen et al., 2005; Marks et al., 2001). The model provides a comprehensive framework for understanding how various factors impact team dynamics and performance.
Input. Inputs would include the initial conditions, such as the team's composition (clinicians, nurses, AI systems), task complexity, and patient data inputs that the AI processes (e.g., vital signs, imaging results). Examples include how AI-driven alerts about a patient’s deteriorating condition can be fed into the team’s decision-making process (Zhang et al., 2020).
Mediator. Mediators include team processes such as communication, coordination, and trust between human team members and the AI system. For instance, trust in AI has been shown to influence the extent to which clinicians rely on its suggestions or override them (Lyell & Coiera, 2021). Poor interface design can reduce this trust, leading to hesitation or over-reliance on AI, both of which can be detrimental in critical situations (Endsley, 2017).
Output. Outputs would include team outcomes focused on task performance, including the quality and speed of critical care interventions. Studies indicate that well-designed human-AI teams can improve patient outcomes, especially when the AI is seamlessly integrated into workflows (Yu et al., 2022).
Input (Feedback Loop): The final input refers to the feedback loop where the outcomes of the team’s performance influence future decisions, team training, and AI system improvements (e.g., learning from previous cases or incorporating new AI algorithms).
This topic is particularly important because it is still understudied in both research and practice. Much of the existing literature on human-AI collaboration has focused on domains such as autonomous driving or customer service (Guzman & Lewis, 2020), leaving a gap in understanding how AI affects team processes, emergent states, and performance in critical care settings. Some key takeaways from this presentation will include insights into the role of trust and communication, user interface design, and training in effective human-AI collaboration.
First, trust in AI systems is crucial for effective collaboration, yet it is often overlooked. Research suggests that human teams may either overly rely on or underutilize AI systems based on trust levels (Lyell & Coiera, 2021), an issue that becomes critical in time-sensitive, high-stakes environments like critical care (Endsley, 2017). Second, the presentation of AI in human-AI teams will also emphasize the need for intuitive AI interfaces. Poor design can lead to misinterpretations or delays in decision-making, which can have serious consequences in emergency medical situations (Yu et al., 2022). Finally, proper training is essential to ensure that clinicians and staff understand how to collaborate effectively with AI systems. This includes understanding AI's limitations and strengths to avoid over-reliance or mistrust as current training approaches do not adequately address these needs, making it an area for further development (Guzman & Lewis, 2020). Overall, the findings and model we propose will help pave the way for improved integration of AI into human teams in critical care, enhancing decision-making processes and ultimately improving patient outcomes.
References
AHA. (2023). How AI is improving diagnostics, decision-making, and care. American Hospital Association.
Endsley, M. R. (2017). Autonomous driving systems: Trust, driver behavior, and performance. Human Factors, 59(1), 52-67. https://doi.org/10.1177/0018720816659950
Fulga, A., Neagu, M., & Piraianu, A. I. (2023). Advancing patient care: How artificial intelligence is transforming healthcare. Journal of Personalized Medicine, 13(8), 1214. https://doi.org/10.3390/jpm13081214
Guzman, A. L., & Lewis, S. C. (2020). Artificial intelligence and communication: A review of AI applications in human communication research. Journal of Communication, 70(2), 178-197. https://doi.org/10.1093/joc/jqz028
Ilgen, D. R., Hollenbeck, J. R., Johnson, M., & Jundt, D. (2005). Teams in organizations: From Input-Process-Output Models to IMOI Models. Annual Review of Psychology, 56, 517-543. https://doi.org/10.1146/annurev.psych.56.091103.070250
Lyell, D., & Coiera, E. (2021). Automation bias and verification complexity: A systematic review. Journal of the American Medical Informatics Association, 28(5), 986-995. https://doi.org/10.1093/jamia/ocaa338
Marks, M. A., Mathieu, J. E., & Zaccaro, S. J. (2001). A temporally based framework and taxonomy of team processes. Academy of Management Review, 26(3), 356-376. https://doi.org/10.5465/amr.2001.4845785
Yu, K. H., Beam, A. L., & Kohane, I. S. (2022). Artificial intelligence in healthcare. Nature Biomedical Engineering, 6(3), 133-143. https://doi.org/10.1038/s41551-021-00781-z
Event Type
Poster Presentation
TimeMonday, March 314:45pm - 6:15pm EDT
LocationFrontenac Foyer
Digital Health (DH)
Simulation and Education (SE)
Hospital Environments (HE)
Medical and Drug Delivery Devices (MDD)
Patient Safety and Research Initiatives (PS)