The Kaiser Permanente Northern California Advance Alert Monitor Program: An Automated Early Warning System for Adults at Risk for In-Hospital Clinical Deterioration.

BACKGROUND: In-hospital deterioration among ward patients is associated with substantially increased adverse outcome rates. In 2013 Kaiser Permanente Northern California (KPNC) developed and implemented a predictive analytics-driven program, Advance Alert Monitor (AAM), to improve early detection and intervention for in-hospital deterioration. The AAM predictive model is designed to give clinicians 12 hours of lead time before clinical deterioration, permitting early detection and a patient goals-concordant response to prevent worsening. DESIGN OF THE AAM INTERVENTION: Across the 21 hospitals of the KPNC integrated health care delivery system, AAM analyzes electronic health record (EHR) data for patients in medical/surgical and telemetry units 24 hours a day, 7 days a week. Patients identified as high risk by the AAM algorithm trigger an alert for a regional team of experienced critical care virtual quality nurse consultants (VQNCs), who then cascade validated, actionable information to rapid response team (RRT) nurses at local hospitals. RRT nurses conduct bedside assessments of at-risk patients and formulate interdisciplinary clinical responses with hospital-based physicians, bedside nurses, and supportive care teams to ensure a well-defined escalation plan that includes clarification of the patients' goals of care. SUCCESS OF THE INTERVENTION: Since 2019 the AAM program has been implemented at all 21 KPNC hospitals. The use of predictive modeling embedded within the EHR to identify high-risk patients has produced the standardization of monitoring workflows, clinical rescue protocols, and coordination to ensure that care is consistent with patients' individual goals of care. An evaluation of the program, using a staggered deployment sequence over 19 hospitals, demonstrates that the AAM program is associated with statistically significant decreases in mortality (9.8% vs. 14.4%), hospital length of stay, and ICU length of stay. Statistical analyses estimated that more than 500 deaths were prevented each year with the AAM program. LESSONS LEARNED: Unlocking the potential of predictive modeling in the EHR is the first step toward realizing the promise of artificial intelligence/machine learning (AI/ML) to improve health outcomes. The AAM program leveraged predictive analytics to produce highly reliable care by identifying at-risk patients, preventing deterioration, and reducing adverse outcomes and can be used as a model for how clinical decision support and inpatient population management can effectively improve care.

A successful program preventing nonventilator hospital-acquired pneumonia in a large hospital system.

OBJECTIVE: To develop and evaluate a program to presvent hospital-acquired pneumonia (HAP). DESIGN: Prospective, observational, surveillance program to identify HAP before and after 7 interventions. An order set automatically triggered in programmatically identified high-risk patients. SETTING: All 21 hospitals of an integrated healthcare system with 4.4 million members. PATIENTS: All hospitalized patients. INTERVENTIONS: Interventions for high-risk patients included mobilization, upright feeding, swallowing evaluation, sedation restrictions, elevated head of bed, oral care and tube care. RESULTS: HAP rates decreased between 2012 and 2018: from 5.92 to 1.79 per 1,000 admissions (P = .0031) and from 24.57 to 6.49 per 100,000 members (P = .0014). HAP mortality decreased from 1.05 to 0.34 per 1,000 admissions and from 4.37 to 1.24 per 100,000 members. Concomitant antibiotic utilization demonstrated reductions of broad-spectrum antibiotics. Antibiotic therapy per 100,000 members was measured as follows: carbapenem days (694 to 463; P = .0020), aminoglycoside days (154 to 61; P = .0165), vancomycin days (2,087 to 1,783; P = .002), and quinolone days (2,162 to 1,287; P < .0001). Only cephalosporin use increased, driven by ceftriaxone days (264 to 460; P = .0009). Benzodiazepine use decreased between 2014 to 2016: 10.4% to 8.8% of inpatient days. Mortality for patients with HAP was 18% in 2012% and 19% in 2016 (P = .439). CONCLUSION: HAP rates, mortality, and broad-spectrum antibiotic use were all reduced significantly following these interventions, despite the absence of strong supportive literature for guidance. Most interventions augmented basic nursing care. None had risks of adverse consequences. These results support the need to examine practices to improve care despite limited literature and the need to further study these difficult areas of care.