Predicting AKI in the Intensive Care Unit (ICU): Insights from Machine-Learning Analysis

Résumé fourni par la source

Background: AKI remains a common and serious complication in critically ill patients, with significant impact on morbidity and mortality. Understanding the clinical differences between patients who develop AKI and those who do not is essential for early intervention and improved outcomes in ICU. Methods: We retrospectively analyzed 11,660 eICU patients, comparing those with AKI (N=6,552) and without (N=5,108). AKI was defined as a ≥0.3 mg/dL rise in creatinine within 48 hours and staged 1–3. Admission labs and clinical data were assessed. Logistic regression and neural network models identified key AKI predictors. Results: Patients with AKI were older, had higher BMI and SOFA scores, and exhibited elevated serum creatinine, BUN, lactate, and potassium, along with lower bicarbonate. Comorbidities such as chronic kidney disease, diabetes, and heart failure were more prevalent in the AKI group, who also required vasopressors and mechanical ventilation more frequently. Multivariable logistic regression showed that higher serum creatinine (OR 7.11, 95% CI 5.51–9.18, p<0.001), BUN (OR 1.01, 95% CI 1.00–1.02, p=0.008), lactate (OR 1.13, 95% CI 1.07–1.20, p<0.001), hemoglobin (OR 1.05, 95% CI 1.01–1.10, p=0.027), and vasopressor use (OR 1.28, 95% CI 1.04–1.59, p=0.022) independently predicted AKI, while male gender (OR 0.72, 95% CI 0.58–0.89, p=0.003) and age (OR 0.99, 95% CI 0.98–1.00, p=0.005) were associated with lower risk. Neural network analysis identified initial serum creatinine as the most influential predictor (importance score 0.168, 100% normalized), substantially outperforming lactate (38.4%), AST (34.9%), and total lymphocytes (28.2%). The model achieved 82.7% accuracy in training and 79.1% in testing, with higher predictive accuracy for positive cases. Conclusion: In critically ill patients, higher creatinine, BUN, lactate, hemoglobin, and vasopressor use significantly increased AKI risk, while male gender and older age were protective. The burden of comorbidities and increased need for organ support in AKI patients underscore the need for early risk stratification. Machine learning models, particularly those highlighting serum creatinine, offer robust predictive performance and may enhance clinical decision-making for AKI prevention and management in the ICU.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.