Impact of an Artificial Intelligence-Powered Clinical Decision Support System for Acute Kidney Injury Prevention in the Intensive Care Unit: Single-Center Uncontrolled Before-and-After Implementation Study.

Background: Acute kidney injury (AKI) is a frequent and serious complication among hospitalized patients, particularly in critical care settings, where its incidence can exceed 50%. AKI is associated with increased mortality, prolonged hospitalization, dialysis dependence, and higher health care costs. Although the KDIGO (Kidney Disease: Improving Global Outcomes) guidelines emphasize supportive care, hemodynamic optimization, and avoidance of nephrotoxins, their implementation remains inconsistent, partly due to the lack of timely risk stratification. Recent advances in artificial intelligence have enhanced early prediction and detection of AKI, offering new opportunities to improve patient outcomes and intensive care unit (ICU) efficiency. The U-Care Renal Platform (UCRP; U-Care Medical S.r.l), a Conformité Européenne (CE)-marked artificial intelligence-powered medical device, integrates directly with the ICU electronic health record to continuously analyze patient data and predict the risk of moderate or severe AKI within 24 hours, providing actionable, guideline-based recommendations. While the predictive performance of UCRP has been validated previously, its real-world impact on clinical and operational outcomes in the ICU remains underexplored. Objective: This single-center uncontrolled before-and-after implementation study aims to evaluate the association between UCRP implementation and selected ICU clinical and operational outcomes in routine practice at SCIAS Hospital, Barcelona. Methods: This study was conducted as a retrospective service evaluation of a workflow-embedded clinical decision support system between March 2023 and March 2025. It included 202 postsurgical adult ICU patients. Outcomes of interest were assessed by comparing preimplementation and postimplementation periods. Months during which the UCRP was inactive were excluded from the analysis (total excluded duration: 10 months; 5 in the preimplementation period and 5 in the postimplementation period). The outcomes included the incidence of moderate-to-severe AKI (KDIGO stages 2 and 3), the use of nephrotoxic medications, the frequency of hypotensive episodes among patients with AKI, and the ICU length of stay. Results: During the postimplementation period, lower rates of moderate-to-severe AKI (9/99, 9.1% vs 12/103, 11.7%), nephrotoxic drug administration, and hypotensive episodes among patients with AKI were observed compared with the preimplementation period. Conclusions: Integration of the UCRP into ICU workflows was associated with differences in selected AKI-related process and intermediate clinical outcomes in this single-center uncontrolled before-and-after implementation study. However, given the study design, causal relationships cannot be established, and the findings should be interpreted as preliminary signals requiring confirmation in larger, controlled, and multicenter studies, including patient-centered outcomes.