AI-Generated Real-Time Alerts for Declining Health Speeds Up Treatment and Reduces Hospital Deaths

By HospiMedica International staff writers
Posted on 17 Jun 2024

A fundamental objective of inpatient care is the timely intervention to prevent or manage clinical deterioration, which often leads to escalated care associated with poorer outcomes and increased use of resources. Historically, clinicians have utilized traditional manual methods like the Modified Early Warning Score (MEWS) to predict clinical deterioration. While these scores have shown good performance in retrospective assessments, their prospective validation has been more limited. Recent advancements have seen machine learning (ML) models, trained on extensive electronic health record (EHR) data, outperforming these older methods. These ML approaches generally have retrospective designs, although a few studies have explored the real-world application of ML models, noting improvements in mortality rates. However, solid data on these models are still lacking. Now, new research has found that hospitalized patients were 43% more likely to receive escalated care and significantly less likely to die if their healthcare team received AI-generated alerts about adverse changes in their health status.

The study conducted by researchers at the Icahn School of Medicine at Mount Sinai (New York, NY, USA;) aimed to assess whether rapid AI and machine learning-generated alerts, trained on diverse patient data, could reduce the need for intensive care and mortality rates. This prospective, non-randomized study involved 2,740 adult patients across four medical-surgical units at Mount Sinai Hospital, divided into two groups: one received real-time alerts on potential deterioration directly to their care teams, and the other had alerts generated but not delivered.


Image: Flow chart showing the study notification protocol (Photo courtesy of Mount Sinai Health System)

In the units where alerts were not delivered, patients meeting standard deterioration criteria received immediate intervention from a rapid response team. Further results from the intervention group showed that these patients were more likely to receive cardiovascular support medications, suggesting proactive measures by physicians; they also exhibited a reduced mortality rate within 30 days. The algorithm has since been implemented across all stepdown units at Mount Sinai Hospital, with a streamlined workflow. A team of intensive care physicians reviews the 15 highest-scoring patients daily, providing treatment recommendations to the attending doctors and nurses. As the algorithm is continuously retrained with data from an increasing number of patients, assessments by the intensive care team act as the benchmark for accuracy, further enhancing the algorithm's precision through reinforcement learning.

"Our research shows that real-time alerts using machine learning can substantially improve patient outcomes," said senior study author David L. Reich, MD, President of The Mount Sinai Hospital and Mount Sinai Queens, the Horace W. Goldsmith Professor of Anesthesiology, and Professor of Artificial Intelligence and Human Health at Icahn Mount Sinai. "These models are accurate and timely aids to clinical decision-making that help us bring the right team to the right patient at the right time. We think of these as ‘augmented intelligence’ tools that speed in-person clinical evaluations by our physicians and nurses and prompt the treatments that keep our patients safer. These are key steps toward the goal of becoming a learning health system."

Related Links:
Icahn School of Medicine at Mount Sinai


Latest Critical Care News