West Orange, N.J.-based RWJBarnabas Health is working to harness the power of predictive modeling to identify early signs of patient deterioration, enabling clinical teams across its 12 hospitals to intervene more quickly and save lives.
Epic's Deteriorization Index examines data including vital signs, demographic information and lab results to identify inpatients at risk of deterioration or death. The tool assigns patients a score of 0 to 100, with a higher number indicating an elevated risk of deterioration. Since the model scans data from the entire patient profile, versus current physiological information only, it is able to flag patients who might not show obvious health problems, such as low blood pressure, a fast heart rate or low urine output. It also refreshes about every 20 minutes, providing ongoing patient assessments in the background.
"It's like the machine is rounding on you constantly," Andy Anderson, MD, executive vice president and chief medical and quality officer of RWJBarnabas Health, told Becker's. "We've been able to identify patients, intervene early and save lives. We've seen some really significant results and patients who might otherwise have died because we were able to detect them earlier."
Since the tool went live systemwide in March, RWJBarnabas estimates it has saved about 100 lives.
Dr. Anderson recently spoke with Becker's about the tool's rollout and early successes, offering a glimpse into how predictive modeling is shaping care delivery at RWJBarnabas Health.
The rollout
In February 2023, the system created an interdisciplinary committee to validate the tool at Robert Wood Johnson University Hospital, its flagship facility in New Brunswick, N.J. This effort came as the system approached the final stages of its Epic rollout, which is set to be completed in October 2024.
During the deterioration index's validation phase, leaders sought to understand how often the alerts fired and whether they fired for the appropriate patients. After making adjustments for oversensitivity, the team developed a three-tiered alert system with the following metrics:
- A green alert for patients with a risk score of 0 to 30
- A yellow alert for those scoring 31 to 59
- A red alert for patients scoring 60 or higher
The group determined that sending alerts to the hospital's rapid response team, rather than the primary attending or bedside nurse, would minimize administrative burden and prevent potential delays in care. Red alerts are sent directly as text messages to the rapid response team members' cellphones, detailing the primary reason for the alert.
In May 2023, the system launched a pilot at Robert Wood Johnson University Hospital. Later that year, leaders tested the model at three additional hospitals with varying rapid response team compositions.
"A big factor in success here was a lot of communication and input from front-line clinicians who were going to be using this tool," Dr. Anderson said. "We phased it in, fine-tuned it and learned a lot along the way before rolling it out to the whole system."
For example, after receiving feedback from physicians and some leaders, the health system worked with its legal team to develop formal guidelines for using the deterioration index that align with the World Health Organization's AI guidance.
The guideline underscores that AI tools are meant to augment, not replace, physicians' individual assessments and clinical decision-making, ensuring that human expertise remains at the heart of patient care.
Strong communication and education, coupled with early results from the tool, proved powerful in securing buy-in from both leadership and front-line clinicians, according to Dr. Anderson.
"The fact that we've got results and that people are being saved, I think that's a pretty compelling reason for people to embrace it," he said.
Early results
Six months into the systemwide rollout of the tool, RWJBarnabas Health reports a 15% relative reduction in inpatient mortality. Based on 2023 data, this improvement translates to an estimated 100 lives saved. During the initial pilot phase, the mortality rate for patients transferred to the intensive care unit also fell 27%.
Although the data has not yet undergone peer review, RWJ Barnabas is analyzing it for statistical significance. A team presented early results from the initiative Aug. 21 at Epic's annual Users Group Meeting.
With early success evident, the potential for predictive tools to transform patient care and outcomes is becoming increasingly clear, according to Dr. Anderson.
"It's one of those things where I don't see us ever going back," he said. "Why didn't we always have this? There's going to be other opportunities for us to use this type of AI tool, and we're doing some of that already with readmissions. It's just another example of how these predictive tools can be very powerful."