Related Experiment Video
Updated: Jan 15, 2026
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Construct validation of machine learning models for predicting surgical site infection risk following ankle fracture
Qinyang Zhang1, Guolin Chen2, Chengqiang Zhou1
1Department of Orthopaedic Surgery, Chongqing Municipal Health Commission Key Laboratory of Musculoskeletal Regeneration and Translational Medicine/Orthopaedic Research Laboratory, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China.
Background:
Surgical site infection (SSI) is a prevalent and severe complication following internal fixation of ankle fractures. The occurrence of SSI not only increases patient healthcare costs but also significantly contributes to elevated morbidity and mortality rates. Assessing preoperative risk factors is crucial for improving risk stratification, which, in turn, enables more accurate selection and implementation of surgical protocols. This study aims to utilize advanced machine learning (ML) techniques to develop and validate a predictive classification model for SSI risk.
Materials And Methods:
Clinical data were collected from patients who underwent open reduction and internal fixation of ankle fractures at the Taizhou People's Hospital, affiliated with Nanjing Medical University Ethics Committee, between January 2023 and December 2024. We compared eight ML algorithms, including Logistic Regression, Support Vector Machine, Gradient Boosting Machine (GBM), Neural Networks, Extreme Gradient Boosting, K-Nearest Neighbors, AdaBoost, and CatBoost. Model performance was evaluated using multiple metrics, including the area under the curve (AUC), decision curve analysis, calibration curve, accuracy, sensitivity, specificity, precision, F1 score, and the Youden Index.
Results:
Eight ML models were developed and evaluated for predicting the incidence of SSI. The results demonstrated that the GBM model exhibited the best performance across all metrics, including an AUC of 0.919, accuracy of 0.82, sensitivity of 0.93, specificity of 0.81, precision of 0.35, F1 score of 0.51, and Youden Index of 0.74. In contrast, the other seven models showed slightly inferior predictive performance. Further analysis identified preoperative albumin levels, type of trauma, history of diabetes, history of hypertension, and surgery duration as the most significant factors influencing the development of SSI.
Conclusion:
The findings of this study demonstrate that modern ML methods offer high predictive accuracy for SSI risk, with the GBM model outperforming the traditional logistic regression model. This study provides strong evidence for the clinical application of ML-based risk assessment, suggesting that integrating preoperative risk factors with ML techniques can enable more precise, individualized treatment strategies for the prevention and management of SSI.
