Prediction of all-cause in-hospital mortality after ICU admission and 1-year mortality after discharge of patients with acute kidney injury

Acute kidney injury (AKI) is a common comorbidity for in-hospital patients. This study aimed to develop dual prediction models for in-hospital and 1-year post-discharge mortality in patients with AKI, with a focus on identifying novel immunological risk factors. 7,937 patients (1,567 in-hospital deaths; 6,370 discharged) were selected from the Medical Information Mart for Intensive Care (MIMIC-IV) database based on serum creatinine and urine output criteria. Forty-two features were selected using least absolute shrinkage and selection operator (LASSO) regression and were used to construct prediction models. eXtreme Gradient Boosting (XGBoost) showed the best performance for in-hospital mortality prediction (training area under the curve (AUC) = 0.833; test AUC = 0.755). Kaplan-Meier curve and log-rank test revealed significant differences in 1-year mortality rates between patients without AKI and patients with AKI stage greater than 2 and those patients who did not recover from AKI at discharge showing a higher 1-year mortality risk (Hazard Ratio (HR) = 1.2, 95% Confidence Interval (CI) = 1.1-1.2). Among the five independent risk factors associated with 1-year mortality identified using univariate Cox regression, two were immune-related cells: neutrophils (HR = 1.01, 95% CI: 1.00-1.01), and neutrophil-to-lymphocyte ratio (HR = 1.01, 95% CI: 1.00-1.02). In conclusion, XGBoost demonstrated the highest predictive performance for in-hospital mortality, with simplified acute physiology score II emerging as the most significant variable.