thank you. Please listen to this article using the above players. ✖
Want to listen to this article for free?
Complete the form below to unlock access to all audio articles.
New research examines the role of AI in supporting emergency teams
Researchers at Mount Sinai Health Systems conducted one of the biggest prospective assessments of artificial intelligence (AI) in an emergency medical setting. Trained with over 1 million patient records, the machine learning model was designed to identify patients who arrive in the emergency department (EDS) who are likely to require hospitalization.
The aim was to generate within hours (preferably than traditional hospitalization decisions) of patients' arrival so that hospitals could begin planning bed allocation and resource allocation earlier. This study was published in Mayo Clinic Proceedings: Digital Health.
Emergency departments across the nation face sustained challenges with overcrowding and “boarding.” There, hospitalized patients remain in the ED because there is a lack of available patient beds. These delays can increase the waiting time for incoming patients and place an additional burden on clinical staff. Mount Sinai's approach sought to address this issue by providing teams with potential notifications of potential admissions.
Collaboration with clinical teams from seven hospitals
More than 500 ED nurses participated in the survey at seven Mount Sinai Network hospitals. For two months, the AI model generated real-time predictions for approximately 50,000 patient visits. These predictions were compared to nurses' own assessments during the triage process.
Triage is an important stage in emergency care, and determines the urgency of treatment based on the patient's symptoms and condition. In this study, triage nurses were asked to provide an assessment of whether patients needed hospitalization. The researchers then analyzed the accuracy of the AI model independently and in combination with nurse evaluations.
Prediction performance and consistency
Results showed that AI models were consistently implemented across different hospital sites despite differences in patient demographics and case type. Interestingly, when combined with nurse predictions and AI output, the overall accuracy did not increase significantly. This suggests that the model itself is already a strong independent predictor of admission.
Findings suggest that although human expertise is essential for decision-making, AI tools can work effectively as standalone support systems. Its main advantage is that it alerts the care team much earlier in the patient's stay, allowing for faster adjustments in services such as adjusting diagnostic tests and preparing hospital patients.
Potential operational benefits
Although this study did not directly measure operational outcomes, the researchers noted that prior identification of possible hospitalizations would help hospitals address capacity constraints. For example, knowing a few hours in advance that a bed will be needed can help hospital management teams accelerate patient discharge from inpatient units, prepare the necessary equipment and assign staff accordingly.
Overcrowding in the emergency department is a recognized national issue. Previous predictions of admission could improve the patient experience by reducing patient flow bottlenecks, reducing waiting times and reducing pressure on staff.
Next stage of the test
The current study was limited to a single health system and a relatively short assessment period. The authors will test the AI models in a live clinical workflow where predictions are directly integrated into ED operations. Future studies will assess impacts on measurable outcomes, including boarding time, patient throughput, and overall operational efficiency.
Although this technique demonstrated strong predictive capabilities, researchers emphasize that it is intended to complement rather than replace clinical judgment. Nurses' participation in the project (more than 500 people were directly involved) emphasizes the importance of combining technical tools with frontline expertise.
reference: Nover J, Bai M, Tismina P, et al. Comparing machine learning and nurse predictions for hospitalization in a multisite emergency health system. MCP: Digital Health. 2025. doi: 10.1016/j.mcpdig.2025.100249
This content contains text generated with AI support. You can find the AI policy for your technology network here.
