Nueraili Abudurexiti, Bide Liu, Shuheng Wang, Qiang Dong, Maimaitiaili Batuer, Zewei Liu, Xun Li Department of Urology, People’s Hospital
Nueraili Abudurexiti, Bide Liu, Shuheng Wang, Qiang Dong, Maimaitiaili Batuer, Zewei Liu, Xun Li Department of Urology, People’s Hospital of Xinjiang Uygur Autonomous Region, Urumqi, 830001, People’s Republic of ChinaCorrespondence: Xun Li, Department of Urology, People’s Hospital of Xinjiang Uygur Autonomous Region, No. 91, Tian-Chi Road, Tianshan District, Urumqi, Xinjiang, 830001, People’s Republic of China, Email xjmnlixun@163.comObjective: This study aimed to develop and validate a machine learning-based model for predicting systemic inflammatory response syndrome (SIRS) in pediatric patients undergoing percutaneous nephrolithotripsy (PCNL) and to establish a prediction platform specifically tailored for this population.Methods: We retrospectively analyzed clinical data from 410 pediatric patients who underwent PCNL at the People’s Hospital of Xinjiang Uygur Autonomous Region between January 2013 and September 2024. The dataset was split into training and validation sets using a 7:3 ratio based on positive samples. The Synthetic Minority Over-sampling Technique (SMOTE) was applied to overcome class imbalance in the training set, while feature selection was performed using a combination of LASSO regression and Boruta algorithms. Eight advanced machine learning algorithms were employed to construct predictive models. The best-performing model was selected based on multiple performance metrics. Additionally, we validated an existing adult model to assess its effectiveness in the pediatric population and compared it with our model. Shapley Additive Explanations (SHAP) analysis was utilized to determine feature importance and model decision basis. Finally, we developed a prediction platform specifically for pediatric patients.Results: The postoperative SIRS incidence was 20.24%. The LightGBM algorithm demonstrated superior predictive performance, achieving an area under the curve (AUC) of 0.8576 and an F1 score of 0.6154. The existing adult models showed lower predictive accuracy in the pediatric cohort (AUC values of 0.7420 and 0.7053). Analysis of SHAP values indicated that operation time, stone burden, preoperative hemoglobin, preoperative monocyte count, and hydronephrosis were the five most critical features affecting predictions. We established a prediction platform specifically designed for the pediatric population.Conclusion: The LightGBM-based model effectively predicts postoperative SIRS in pediatric PCNL patients, providing a tailored tool for this population. The online prediction platform might be useful to guide clinical decision making.Keywords: pediatric, percutaneous nephrolithotripsy, kidney stones, systemic inflammatory response syndrome, machine learning, clinical prediction platform
