Study: AI could help emergency rooms predict admissions and guide care

Emergency rooms run on adrenaline and split-second decisions, but one thing they've never had is foresight. A new multi-hospital study suggests that artificial intelligence (AI) might change that, helping emergency departments spot which patients will likely need a hospital bed — hours before doctors would normally know. As Jonathan Nover, Vice President of Nursing and Emergency Services at Mount Sinai, put it: "Emergency department overcrowding and boarding have become a national crisis, affecting everything from patient outcomes to financial performance."

The findings hint at a future where chaos at the ER doors could be tamed with a little digital clairvoyance.

How did it work?

The study gathered input from more than 500 nurses across Mount Sinai's seven hospitals. Together, they tested a machine learning model trained on more than a million past patient visits. Over two months, the AI churned through nearly 50,000 new patient arrivals, spitting out predictions about who would be admitted.

• The system forecasted admissions needs before an order was even placed.
• Nurses' triage calls were compared to AI predictions.
• Surprisingly, the AI held its own — it didn't need human predictions layered on top to be accurate.

Dr. Eyal Klang, who leads generative AI research at Mount Sinai's Icahn School of Medicine, explained: "By training the algorithm on more than a million patient visits, we aimed to capture meaningful patterns that could help anticipate admissions earlier than traditional methods."

The result? A tool that transforms messy, complex data into real-time, actionable insights, freeing clinicians from logistical guesswork.

Why does it matter?

Emergency departments are crowded, frantic places where every decision ricochets down the line. When a patient needs to be admitted but no bed is ready — a situation known as "boarding" — everyone suffers.

AI predictions could:

• Give clinicians a head start on bed management.
• Reduce overcrowding and boarding delays.
• Improve patient flow and the overall care experience.
• Let staff spend more time on compassionate care rather than logistics.

Robbie Freeman, Mount Sinai's Chief Digital Transformation Officer, summed it up neatly: "This tool isn't about replacing clinicians; it's about supporting them. By predicting admissions earlier, we can give care teams the time they need to plan, coordinate, and ultimately provide better, more compassionate care."

The context

This wasn't a backroom computer experiment. It was a prospective, real-world trial — one of the largest of its kind — spanning urban and suburban hospitals. Nearly 50,000 patients were part of the evaluation. The study found that AI alone could reliably predict admissions, and crucially, do it early enough to matter.

Of course, there are caveats. The trial was limited to a single health system and ran only for two months. The next step is to fold this AI into live clinical workflows, then measure whether it actually cuts boarding times and improves outcomes.

Still, the vision is striking. As Nover quipped: "Industries like airlines and hotels use bookings to forecast demand and plan. In the ED, we don't have reservations. Could you imagine airlines and hotels without reservations, solely forecasting and planning from historical trends? Welcome to health care."

AI, in this sense, offers hospitals something they've never had before: a reservation system of sorts for patient care. And if it works, it could be the difference between waiting on a gurney in the hallway and getting timely, dignified treatment.