St. Luke’s Health Network uses AI tool to reduce cardiac arrests and ICU transfers

Anna Stone was doing the first rounds of her nursing shift at St. Luke’s Upper Bucks Campus when she noticed a patient’s heart rate was elevated, a sign that they could be at risk of a cardiac emergency.

Before she could look into the patient’s chart and decide whether to call for help, a critical care doctor came rushing to the patient’s bedside.

A drop in the patient’s oxygen levels had been detected by a monitor that uses artificial intelligence to continuously evaluate vital signs. This triggered an automatic alert for the hospital’s critical care team to send help.

The AI tool is intended to help doctors and nurses more quickly identify patients whose condition is deteriorating — often before signs of distress are visible to medical staff — and intervene sooner.

The approach contributed to a 34% decline in cardiac arrests, and a 12% drop in patients crashing so hard and fast that they required rapid response transfers to the ICU between 2022 and 2024, according to St. Luke’s.

Survival rates among cardiac arrest patients rose from 24% to 36%.

St. Luke’s experiment with a program called the Deterioration Index, created by healthcare software giant Epic, is among the latest ways hospitals are bringing artificial intelligence into their patients’ rooms.

In other Philadelphia-area initiatives, Jefferson Health and Penn Medicine recently debuted an ambient listening tool that records conversations between doctors and patients, distilling the critical details into a well-organized visit note.

St. Luke’s has been using its AI monitoring system across all 16 of its campuses, including Quakertown, Upper Bucks, and Grand View, which the health system acquired in July.

The health system’s initiative was recognized by The Hospital and Healthsystem Association of Pennsylvania, the region’s largest industry group for hospitals, with an award honoring safety and quality initiatives that improved patient care while reducing hospital costs.

The monitoring device, which attaches to a patient’s finger, records and continuously updates patients’ electronic medical records with vital metrics such as heart rate, blood pressure, and lab work results.

Using this matrix of data points, it assigns each patient a “deterioration index” — a score between 0 and 100 indicating their overall stability — and automatically alerts critical care when the score rises too high.

It is not intended to replace in-person monitoring, but serves as an extra set of eyes when nurses are away from their bedside.

What’s more, the sophisticated technology is capable of picking up on nuanced changes in a patient’s status before they show physical signs of distress.

“We would ideally like to intervene on these patients before they reach a point where the intervention isn’t that helpful,” said Matthew Zheng, a critical care doctor at St. Luke’s Hospital — Upper Bucks. “Our nurses work very hard, but they can’t be in the same room all the time.”

When a patient’s “deterioration index” rises above 60, the device sends an alert to the hospital’s virtual response center — a remote hub where a nurse monitors three screens showing the status of all patients.

Alerts may also be sent directly to a patient’s care team or the rapid response unit, if the AI monitoring detects that a patient is quickly deteriorating and needs emergency care.

“What that’s allowed is for us to have a proactive response instead of being reactive to patients,” said Charles Sonday, an associate chief medical information officer at St. Luke’s who leads AI initiatives.

Stone, the Quakertown nurse, said having the tool to constantly watch over patients while she’s out of their room is reassuring.

Doctors like that it enables them to quickly get up to speed on the status of a patient they transferred out of the ICU, and respond more immediately to their new medical needs, said Zheng, the critical care doctor.

St. Luke’s plans to continue fine-tuning the technology, and customize it to meet the unique patient profiles of each of its campuses, which span 11 counties and two states, from the Lehigh Valley to New Jersey.

The social and economic factors that affect patient health, such as pollution, and illness rates, vary significantly across the health system’s sprawling network, Sonday said.

The system will also explore customizing the tool for specialty services, such as pediatrics and behavioral health.