An integrated clinical and imaging model for predicting post-traumatic nonunion.

Objective: This model aims to support early risk stratification and clinical decision-making. Methods: A retrospective cohort of 343 patients with unilateral closed long bone fractures treated by internal fixation was analyzed. Clinico-radiological variables-including Injury Severity Index, maximum fracture gap width, volume of cystic changes, callus volume growth rate, and Radiographic Union Scale for Tibial fractures (RUST) score-were systematically collected. Predictors were selected using Least Absolute Shrinkage and Selection Operator (LASSO) regression. Multivariable logistic regression and multiple machine learning algorithms (e.g., random forest, gradient boosting machine) were applied. The dataset was split 7:3 into training and validation sets. All feature selection and model hyperparameter tuning were restricted to the training set, information leakage was avoided via 10-fold cross-validation, and a single evaluation was finally conducted on the independent validation set. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. Results: Training and validation cohorts exhibited comparable baseline characteristics. Five independent predictors were identified: Injury Severity Index, maximum fracture gap width, and cystic change volume were risk factors, whereas callus growth rate and RUST score were protective. The gradient boosting machine model achieved the highest predictive accuracy, with an AUC of 0.866 (95% Confidence Interval (CI): 0.783-0.948) in training and 0.858 (95% CI: 0.716-1.000) in validation. Calibration was satisfactory, and decision curve analysis demonstrated a superior net benefit across threshold probabilities. A nomogram was constructed, and SHapley Additive exPlanations (SHAP) analysis improved interpretability. Conclusion: A machine learning model integrating clinical and imaging predictors was successfully developed and validated for predicting post-traumatic nonunion. It exhibits strong discriminative ability and may assist in personalized treatment planning.