Predicting Postoperative Sepsis Risk in Diabetic Urolithiasis Patients: A Multimodal Clinical Data-Driven Machine Learning Model

Abstract Objective Diabetic patients are more prone to urinary tract infections due to metabolic abnormalities and impaired immune function, which can progress to urosepsis. This study aims to construct and validate an efficient and accurate predictive model, based on multimodal clinical data combined with machine learning techniques, to early assess the risk of postoperative infectious urosepsis in diabetic patients with urinary stones, thereby providing support for clinical decision-making. Methods 532 patients diagnosed with diabetes who underwent surgical treatment for upper urinary tract stones was included. The patients were randomly divided into training (70%) and validation (30%) cohorts. A total of 164 multimodal clinical parameters were included, and those with statistically significant differences ( P < 0.05) were selected. Feature selection was performed using LASSO regression and the Boruta algorithm. Nine machine learning (ML) algorithms were explored to predict the risk of postoperative infectious urosepsis in diabetic patients with urinary stones. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), learning curve, calibration curve, and decision curve analysis (DCA). The contribution of key predictive factors was visualized using the SHAP method. Results Among the 164 multimodal clinical parameters, 59 were significantly associated with the occurrence of postoperative infectious urosepsis. After feature selection, four key parameters were identified: p_IL_6, p_PCTp_ALB, p_T, and p_HR. Among the nine ML algorithms, logistic regression exhibited the best predictive ability. The AUC in the validation cohort was 0.903. The learning curve indicated good and stable model fitting, while the calibration curve demonstrated a high degree of agreement between predicted and actual probabilities. The decision curve analysis revealed that the model provided significant clinical net benefit within a threshold range of 5% to 90%. Conclusion The model constructed using logistic regression performed excellently in predicting the risk of postoperative infectious urosepsis in diabetic patients with urinary stones and can help clinicians better identify high-risk patients. Further prospective validation in multicenter studies is needed to confirm the model's generalizability.