Association Between Whole Blood Cell-Derived Inflammatory Markers and All-Cause Mortality in Patients with Sepsis-Related Acute Kidney Injury and Development of a Machine Learning Prognostic Model

Background Sepsis-associated acute kidney injury (SA-AKI) is a common and severe complication of sepsis, associated with significantly increased patient mortality. Whole blood cell-derived inflammatory markers, such as the neutrophil-to-lymphocyte ratio (NLR) and the systemic immune-inflammation index (SII), are valuable tools for assessing inflammatory status and prognosis due to their ready availability and cost-effectiveness. However, the association between these markers and prognosis in patients with SA-AKI requires further investigation. This study aims to clarify the relationship between these inflammatory markers and all-cause mortality in SA-AKI patients and to develop a prognostic model using machine learning methods, thereby providing new evidence to support clinical decision-making.. Methods This study included patients with SA-AKI, defined by Sepsis-3 and KDIGO criteria, from the MIMIC-IV database (development cohort) and an external cohort from The Second Affiliated Hospital of Nanchang University (validation cohort). We collected demographic data, vital signs, laboratory parameters (including NLR, NPAR, PNR, and SII), comorbidities, and treatment information. The primary outcome was 14-day all-cause mortality following ICU admission.Kaplan-Meier survival analysis and multivariate Cox regression models were used to evaluate the association between inflammatory markers and mortality. For model development, the MIMIC-IV cohort was randomly split into training (70%) and testing (30%) sets. Feature selection was performed using LASSO regression, the Boruta algorithm, and univariate logistic regression. Six machine learning models (Random Forest, Logistic Regression, XGBoost, GAMboost, CatBoost, and GBM) were subsequently developed. Five-fold cross-validation was employed for model tuning, and the SMOTE (Synthetic Minority Over-sampling Technique) algorithm was applied to address class imbalance.Model performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC) and Brier score. Model interpretability was assessed using SHAP (SHapley Additive exPlanations), and clinical utility was determined via Decision Curve Analysis (DCA). The best-performing model was then validated using the external cohort. Results A total of 4,311 patients with SA-AKI were included in the development cohort, of whom 696 (16.1%) died within 14 days. Survival analysis demonstrated that elevated NLR, SII, and NPAR, as well as a low PNR, were significantly associated with 14-day all-cause mortality. Following feature selection, 17 predictors were used to develop six machine learning models. The CatBoost model achieved the best discrimination, with an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.861 on the internal test set and 0.755 on the external validation cohort. Decision curve analysis confirmed its potential clinical utility across a wide range of threshold probabilities. Furthermore, SHAP analysis identified lactate and blood urea nitrogen as the most significant contributors to the model's predictions. Conclusion The predictive model developed using the CatBoost algorithm effectively assesses the 14-day mortality risk in patients with SA-AKI. The model demonstrated robust predictive performance and clinical applicability in both internal and external validation cohorts. This tool holds promise for assisting clinicians with risk stratification and guiding personalized treatment strategies for SA-AKI patients.