RWJBarnabas Health Reduced Mortality by 18%: How an AI Agent Accelerated Response to Critical Patient Conditions

A multi-hospital study involving 23,132 patients, led by RWJBarnabas Health and Rutgers Robert Wood Johnson Medical School, resulted in the implementation of an AI agent that reduced in-hospital mortality among high-risk patients by 18%. The system now automatically alerts rapid response teams, significantly accelerating intervention and improving patient outcomes.

In hospitals, critical conditions evolve rapidly. Every minute of delay reduces a patient's chances of survival. However, human factors, fatigue, and staff overload often lead to signals of patient deterioration being noticed too late. This results in increased mortality and unnecessary transfers to intensive care. Modern technologies allow for the creation of a system that continuously monitors patient status and instantly notifies relevant specialists, thereby saving lives and optimizing resources.

The Reality of the Problem: Costly Delays and "Alert Fatigue"

In traditional hospital systems, medical staff are relied upon to monitor patient indicators, react to changes, and make decisions about further actions. However, with a large number of patients and the complexity of medical data, this becomes a significant challenge.

Existing early warning systems, based on machine learning algorithms, often suffer from a "deployment gap": they are good at predicting patient deterioration but don't always integrate effectively into real clinical practice. Notifications displayed only on standard dashboards in Electronic Health Records (EHRs) are often missed due to overworked nurses and doctors. By the time a signal is noticed, the patient's physiological state may have already critically worsened.

Furthermore, inaccurate or excessive alerts lead to "alert fatigue," where staff begin to ignore warnings, exacerbating the problem. This can also result in unnecessary patient transfers to Intensive Care Units (ICUs), creating undue strain on these resources.

The Path to the AI Agent: From Passive Monitoring to Active Intervention

Previously, RWJBarnabas Health, like many other hospitals, used standard methods for monitoring patient conditions. Systems provided data, but they lacked the ability to actively and immediately inform staff. Relying solely on passive EHR dashboards was insufficient to prevent rising in-hospital mortality.

The company concluded that a tool was needed that not only displayed data but also acted proactively, automatically initiating the intervention process. This led to the development and implementation of an AI agent capable of real-time data analysis, identifying critical changes, and immediately notifying the relevant teams.

How the AI Agent Was Designed: Dynamic Analysis and Instant Alerts

The AI agent was designed as an integrated system that continuously analyzes data from electronic health records and, if critical risk is identified, automatically sends notifications to rapid response teams. The main elements of the system include:

• Dynamic Recalculation Every 15 Minutes. The AI agent constantly analyzes 31 EHR variables, including vital signs trends, laboratory results, nursing assessments, and patient age. The patient deterioration index is recalculated every 15 minutes, ensuring data timeliness.
• Calibrated Risk Triage Thresholds. Patients are categorized into three risk levels: Green (<30), Yellow (30–59), and Red (>60). A "Red" level indicates a high risk of severe deterioration or death.
• Unified Mobile Push Integration. When a patient's condition reaches the "Red Alert" threshold, the system automatically sends an instant push notification directly to the mobile devices of on-duty Rapid Response Teams, bypassing static desktop interfaces.
• Alert Fatigue Suppression Logic. Built-in rules prevent redundant notifications if the patient is already in the ICU, receiving comfort care, or if a rapid response or sepsis alert has already been triggered within the last 6 hours. This helps reduce unnecessary signals and increases their significance for staff.

The goal was to identify patients in the early stages of deterioration, when intervention could still alter the course of events. Dr. Thomas Nahass, VP of Health Informatics at RWJBarnabas Health, noted that the deterioration index provides an earlier point for intervention, which is critically important for outcomes.

Implementation: From Pilot to System-Wide Solution

The implementation of the AI agent was carried out in stages across 11 acute care hospitals. The process included:

• Integration with Existing EHR. The system was embedded within the Epic Deterioration Index (EDI), allowing for the use of existing data and infrastructure.
• Staff Training. Rapid response teams and other medical personnel were trained on the new alert system and response protocols.
• Gradual Scaling. Starting with pilot projects, coverage was gradually expanded across the entire hospital network, which allowed for process refinement and consideration of each clinic's specific needs.

Crucially, the system not only predicted but also initiated action, which was key to its success.

Results

A key achievement was an 18% reduction in in-hospital mortality among high-risk patients. At the same time, the number of evaluations performed by rapid response teams increased from 25.3% to 37.5%, indicating earlier and more active intervention. Importantly, this increase did not lead to a rise in ICU transfers, demonstrating the effectiveness of early intervention at the general ward level.

How to Replicate This in Your Organization

This case demonstrates that AI agents can significantly improve clinical outcomes and optimize resource utilization in healthcare settings. To replicate this success, consider the following steps:

• Assess Current Processes. Identify where delays in responding to patient deterioration occur in your institution and where human factors are most vulnerable.
• Integrate the AI Agent with Existing Systems. Utilizing existing data from EHRs and other information systems will significantly simplify implementation.
• Develop Clear Response Protocols. The AI agent should be part of a comprehensive process where every step, from alerting to intervention, is clearly defined.
• Focus on Mobile Notifications. Instant push notifications to mobile devices ensure a rapid response, overcoming the problem of "alert fatigue" on stationary screens.
• Implement Logic for Suppressing Redundant Alerts. This is critically important for preventing "alert fatigue" and maintaining staff trust in the system.

If this case sounds like what's happening in your company, our manager can help: he'll analyze your business and niche for free and point out where an AI agent would bring a real result in your case. Message the manager