Work

Prediction of acute kidney injury in traumatic shock patients using machine learning methods

Year: 2026

Type: article

Abstract: Acute kidney injury (AKI) represents a common and potentially fatal complication in patients with traumatic shock. Early and accurate prediction of AKI plays a critical role in guiding individualized therapeutic strategies and improving patient prognosis. The present study aimed to develop an interpretable machine learning model for the early prediction of new-onset AKI within 48 h after admission in this high-risk population. A cohort of 407 patients with traumatic shock was retrospectively enrolled and randomly assigned to training and testing cohorts at a 7:3 ratio. A three-step feature selection strategy—incorporating the Boruta algorithm, the least absolute shrinkage and selection operator (LASSO), and multivariable logistic regression—was employed to identify key predictors. These predictors were then used to construct eight machine learning models, including Logistic Regression (LR), K-Nearest Neighbors (KNN), Naive Bayes (NB), Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF), Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LGBM). Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration plots, and decision curve analysis (DCA). Discrimination was quantified using the mean area under the curve (AUC) with 95% confidence intervals (CIs) derived from 500 bootstrap resamples. Additional metrics including specificity, sensitivity, precision, recall, and F1-score were also calculated. SHapley Additive exPlanations (SHAP) were applied to enhance model interpretability and quantify the contribution of individual predictors. Eight variables, including the Sequential Organ Failure Assessment (SOFA) score, procalcitonin (PCT), lactate (Lac), serum potassium (K), coefficient of variation of red blood cell distribution width (RDW-CV), standard deviation of red blood cell distribution width (RDW-SD), Glasgow Coma Scale (GCS) score, and C-reactive protein (CRP), were identified as key predictors of 48-hour-onset AKI. Among the evaluated models, XGBoost demonstrated relatively balanced predictive performance during internal validation, achieving AUC values of 0.979 in the training cohort and 0.898 in the testing cohort. Calibration analysis demonstrated good agreement between predicted and observed outcomes, while Decision curve analysis suggested a favorable net clinical benefit across a range of clinically relevant threshold probabilities. SHAP analysis identified PCT and the SOFA score as the most influential contributors to model predictions. This study developed an interpretable machine learning model for predicting 48-hour-onset AKI in patients with traumatic shock. The model demonstrated favorable predictive performance during internal validation and may assist in identifying patients at increased risk of AKI. However, prospective multicenter studies with external validation are required before routine clinical implementation. Not applicable.

Source: Scandinavian Journal of Trauma Resuscitation and Emergency Medicine

DOI: 10.1186/s13049-026-01684-4

Authors: Dongfang Wang, Shuaipeng Gu, Yahui Li, Shengying Wang

Institutions: Qingdao Municipal Hospital, Huazhong University of Science and Technology, Tongji Hospital

Language: English

Topic: Acute Kidney Injury Research

Subfield: Nephrology

Field: Medicine

Domain: Health Sciences

SDG: Good health and well-being

Open Access status: gold

Estimated APC paid: $2,590