Abstract
Purpose: A prediction model for systemic inflammatory response syndrome (SIRS) following percutaneous nephrolithotomy (PCNL) was created using a machine learning (ML) algorithm. The model's diagnostic prediction ability and clinical utility for SIRS were examined and analyzed in order to give clinicians a foundation for diagnosing and treating patients' conditions.
Methods: 444 individuals with upper urinary tract calculi who had PCNL were included in this study. Depending on whether SIRS developed after PCNL, the patients were divided into SIRIS and non-SIRS groups, 68 clinical variables were examined. 131 of the 444 patients experienced SIRS. The traditional binary logistic regression (LR) was utilized to create the prediction model after the clinical data from the two groups were compared to assess the risk variables, and the optimal ML algorithm was chosen to create the SIRS prediction model. The ML prediction models were assessed for the prediction model's performance by drawing the receiver operating characteristic curve (ROC) and calculating the area under the curve (AUC).
Results: 24 characteristics showed statistical significance in univariate analysis. By using multivariate analysis, 8 independent risk variables were found, including preoperative nitrite (OR=7.453, P<0.001), history of hypertension (OR=1.93, P=0.021), postoperative urinary white blood cells (OR=1.001, P<0.034), postoperative nitrite (OR=6.775, P<0.001), postoperative interleukin-6 (OR = 1.001, P = 0.028), postoperative C-reactive protein (OR=1.014, P= 0.027) and postoperative nephrostomy (OR=3.004, P<0.001). After determining eight independent risk variables, a binary LR prediction model was created, and its AUC was 0.827. The XGBoost has built a ML prediction model with an AUC of 0.941. The ML model's strong therapeutic advantages are confirmed by the decision analysis curve (DCA).
Conclusion: The ML predictive model is more credible and offers better therapeutic advantages than the traditional LR prediction model, with a higher AUC. In certain instances, ML predictive model might offer additional foundation for clinicians to make timely and precise decisions about the early detection and diagnosis of SIRS.