Interpretable machine learning for predicting optimal surgical timing in polytrauma patients with TBI and fractures to reduce postoperative infection risk.

This retrospective study leverages machine learning to determine the optimal timing for fracture reconstruction surgery in polytrauma patients, focusing on those with concomitant traumatic brain injury. The analysis included 218 patients admitted to Qilu Hospital of Shandong University from July 2011 to April 2024. Demographic data, physiological status, and non-invasive test indicators were collected. Feature selection via the Boruta and LASSO algorithms preceded the construction of predictive models using Random Forest, Decision Tree, K-Nearest Neighbors, Support Vector Machine, LightGBM, and XGBoost. The Random Forest model excelled in the training set, with an AUC-ROC of 0.828 and accuracy of 0.745, and sustained high performance in the validation set (AUC-ROC: 0.840; Accuracy: 0.813). The final model was informed by eight critical factors, including the Glasgow Coma Scale score, calcium levels, D-dimer, hemoglobin, platelet count, LDL-cholesterol, prothrombin time-international normalized ratio, and prior surgeries. SHAP and LIME algorithms were utilized for model interpretation, elucidating the importance and predictive thresholds of the variables. The application of machine learning in this study provided precise predictions for optimal surgical conditions and timing in polytrauma patients with traumatic brain injury and fractures. This study's findings provide a foundation for personalized surgical planning, potentially reducing postoperative infections and improving patient prognoses.