Implementation of an AI-Triggered Rapid Response - Association with Mortality

Use of a real-time, machine learning-enabled clinical deterioration prediction model combined with an automated rapid response team (RRT) activation could reduce inpatient mortality. The objective was to assess reduction in risk-adjusted inpatient mortality following implementation of the Epic Deterioration Index (EDI) combined with an automated RRT activation in a large regional health system.

We conducted a quasi-experimental, staggered cohort study using a pre- versus postimplementation design. The setting was a regional health system with 11 acute care hospitals including academic, community teaching, and nonteaching facilities. Participants included adult medical-surgical admissions with an EDI greater than or equal to 60 from October 1, 2022, to August 30, 2024. We evaluated the clinical impact of deploying real-time, targeted electronic health record (EHR) alerts and RRT push notifications; the primary outcome was risk-adjusted inhospital mortality.

The study included 23,132 patients. Mean age was 71.9 years, 51.6% were male, and 51.4% were non-Hispanic white. In total, 10,803 patients (46.7%) were preintervention and 12,329 (53.3%) were postintervention; 5746 postintervention encounters (46.6%) generated a push notification to the RRT, although not all resulted in RRT activation. The postintervention group compared with the preintervention group was associated with increased RRT activations (37.5% vs. 25.3%; absolute difference, 12.2 percentage points [95% confidence interval (CI), 11.0 to 13.4]; adjusted odds ratio, 1.74 [95% CI, 1.61 to 1.88]) and decreased unadjusted inhospital mortality (18.6% vs. 23.1%; absolute difference, -4.5 percentage points [95% CI, -5.6 to -3.5]), without a significant increase in escalations of care (1.1% vs. 1.0%; absolute difference, 0.1 percentage points [95% CI, -0.18 to 0.37]; adjusted odds ratio, 1.34 [95% CI, 0.97 to 1.85]). Risk-adjusted odds of inhospital mortality, accounting for age, comorbidities, hospital type, EDI score, and clustering at the hospital level were lower in the intervention group (adjusted odds ratio, 0.82 [95% CI, 0.74 to 0.91]).

Implementation of the EDI combined with RRT activation was associated with a reduction in risk-adjusted inpatient mortality across different types of hospitals in a large health system.