Research On The Application Of Machine Learning Model In The Preoperative Diagnosis Of Calculous Pyonephrosis And The Prediction Of SIRS After PCNL

Part Ⅰ.Application of Machine Learning-assisted Models in Predicting Calculous PyonephrosisObjective The purpose of this study was to evaluate the risk factors for obstructive pyonephrosis in patients with upper urinary stones and construct a machine-learning assisted model(MLSM)capable of detecting early preoperative calculous pyonephrosis by integrating a number of clinical parameters with predictive value.Methods The clinical data of 322 patients with upper urinary stones and obstructive hydronephrosis were retrospectively analyzed.Patients were divided into two groups:pyonephrosis group and non-pyonephrosis group.Univariate and multivariate logistic regression(LR)analyses were performed for preoperative variables.Furthermore,the cohort study was randomly divided into training and test data sets respectively at a ratio of 70:30 and constructed different MLSMs from 22 clinical indicators.The area under ROC curve(AUC)was used to determine the model with the highest recognition.Decision curve analysis(DCA)was used to analyze the clinical net benefit of the prediction model.Results Single-factor LR analysis revealed statistically significant differences in gender,UTI history,serum creatinine,degree of hydronephrosis,CT value of hydronephrosis,and SHACK(p<0.001).Multi-factor LR analysis showed that the degree of hydronephrosis(p=0.003),CT value of hydronephrosis(p=0.001),urinary leucocyte(p=0.005),and positive urine culture flora(p=0.033)were independent risk factors for pyonephrosis.In the subsequent MLSM,the eXtreme Gradient Boosting(XGBoost)model showed good predictive power;The AUC and accuracy were 0.981 and 0.982,respectively,followed by SVM,Lasso-LR,LR,and random forest.Verification of the model showed that the SVM model had the highest AUC(0.977,95%CI,0.952-1.000),followed by Lasso-LR,XGBoost,LR and random forest.Conclusion Our present MLSM demonstrates a high degree of disparity between hydronephrosis and obstructive pyonephrosis.Moreover,their advantages and predictive capabilities also differed depending on the model of choice.Therefore,using these models would greatly aid urologists in planning perioperative procedures as well as making treatment decisions.Part Ⅱ.Application of machine learning model in predicting SIRS after calculous pyonephrosisObjective This study aimed to evaluate the risk factors and risk classification of systemic response inflammatory syndrome(SIRS)in calculous pyonephrosis patients following percutaneous nephrolithotomy(PCNL)and to develop a risk grading system and a machinelearning assisted model(MLSM)that could predict the occurrence of SIRS.Methods The clinical data of 158 patients with upper urinary tract stones and pyonephrosis who underwent PCNL surgery were retrospectively analyzed.The patients were divided into SIRS group(32 cases)and non-SIRS group(126 cases).Univariate and multivariate logistic regression analyses were used to evaluate the risk factors of SIRS after PCNL,and a risk grading scale was constructed for predicting the risk of SIRS.Furthermore,analysis of the prediction performance of different MLSMs was performed,and the area under the ROC curve(AUC)was used to determine the model with the highest recognition performance.Results In this study,we observed statistically significant differences in the stone load(p<0.001),preoperative blood creatinine value(p=0.026),percutaneous renal puncture sheath size(p=0.009),PCNL intraoperative perfusion fluid flow rate(p<0.001),preoperative CRP(p=0.001)between groups.Single factor logistic regression analysis(LR)showed that the stone load(>300 mm2)(p=0.001),the size of the puncture sheath(p=0.011),and the preoperative CRP(>10 mg/L)(p=0.001),serum creatinine(p=0.015),PCNL intraoperative perfusion fluid flow rate(>450 ml/min)(p=0.001)were risk factors for postoperative SIRS;Multivariate LR found that stone load(>300 mm2)(p=0.046),the preoperative CRP(>10 mg/L)(p=0.004),and PCNL intraoperative perfusion fluid flow rate(>450 ml/min)(p=0.021)were independent risk factors for SIRS after PCNL.Compared with low-risk and medium-risk groups,the risk of SIRS in high-risk groups was significantly increased(p=0.0007).Among the ML-based models,the XGBoost and SVM models have the strongest discriminative power,with AUC of 0.989,followed by RF(AUC 0.969),Lasso-LR(AUC 0.820),LR(AUC 0.789).Conclusion The preoperative CRP,stone load,and the flow rate of the perfusion fluid during the PCNL may affect the risk of SIRS following PCNL coupled with pyonephrosis.A reasonably early risk grading system might aid in the clarification of the risks of postoperative SIRS following PCNL and guide early clinical treatment.Furthermore,XGBoost and SVM models demonstrated the most potent discriminative power among the MLSMs.Therefore,the application of the MLSM mentioned above can significantly improve the predictive ability of SIRS.