AI Model Boosts Early Delirium Detection for Improving Health Outcomes of Hospitalized Patients

Delirium, a sudden onset of severe confusion, poses serious life-threatening risks and affects up to one-third of patients in hospitals, often going unnoticed. Without intervention, it can lengthen hospital stays, increase the risk of mortality, and lead to poorer long-term outcomes. Despite the efforts of artificial intelligence (AI)-based models to predict delirium, previous attempts have not resulted in significant improvements in patient care. However, a new AI model has successfully enhanced patient outcomes by increasing the detection and treatment of delirium fourfold. The model identifies high-risk patients, notifying a specialized team to assess and, if necessary, initiate a treatment plan.

Developed by researchers at the Icahn School of Medicine at Mount Sinai (New York, NY, USA), the model has been integrated into hospital operations, assisting health care professionals in recognizing and managing delirium, a condition affecting a large proportion of hospitalized patients. Published in JAMA Network Open, the study represents the first successful application of an AI-powered delirium risk model in real-world clinical practice, showing benefits beyond controlled laboratory environments. Unlike previous methods, the research team collaborated closely with Mount Sinai clinicians and staff throughout the development process. This "vertical integration" approach allowed the team to refine the model in real time, ensuring it was practical and effective for use in clinical settings.

The study involved over 32,000 patients at The Mount Sinai Hospital, where the AI model analyzed both structured data and clinicians' notes from electronic health records. It utilized machine learning to detect patterns in chart data linked to a high risk of delirium and incorporated natural language processing to identify cues from the language used in the hospital staff's notes. This technique captures subtle signs of mental status changes, often observed by staff without them realizing the impact their observations have on improving the AI model's accuracy.

The model was applied in a highly diverse group of patients, encompassing a wide variety of medical and surgical conditions, much broader than those typically included in machine learning studies focused on delirium prediction. The tool led to a dramatic increase in monthly delirium detection rates—from 4.4% to 17.2%—allowing for earlier intervention. Moreover, patients identified by the model received lower doses of sedative medications, reducing side effects and enhancing overall care. Although the model has shown strong results at The Mount Sinai Hospital and is being tested at other Mount Sinai locations, further validation at different hospital systems is necessary to assess its performance in varied settings and to make any required adjustments.

“Current AI-based delirium prediction models haven’t yet shown real-world benefits for patient care. We wanted to change that by creating a model that accurately calculates delirium risk in real time and integrates smoothly into clinical workflows, helping hospital staff catch and treat more patients with delirium who might otherwise be overlooked," said senior corresponding study author Joseph Friedman, MD. “Our model isn’t about replacing doctors—it’s about giving them a powerful tool to streamline their work. By doing the heavy lifting of analyzing vast amounts of patient data, our machine learning approach allows health care providers to focus their expertise on diagnosing and treating patients more effectively and with greater precision.”