Application Of Improved SOFA Score After Percutaneous Nephrolithotomy And Establishment Of Sepsis Prediction Model Through Artificial Intelligence

ObjectiveTo study the effect of gender factors on the SOFA score after percutaneous nephrolithotomy(PCNL),and to improve the SOFA score and evaluate its predictive efficacy.Meanwhile,the prediction model of PCNL postoperative sepsis was established by artificial intelligence,providing a reference for urologists in the early recognition of sepsis.MethodsPart Ⅰ612 PCNL patients(356 males and 256 females)were retrospectively observed in our center.Look for differences in SOFA scores between the sexes,and whether these differences affect the predictive accuracy of septic shock.The new SOFA score was constructed by replacing the liver and kidney function sub-score in the original SOFA score with the changed values of bilirubin and creatinine in the perioperative period.The new SOFA score was eventually used to eliminate the effect of gender on the prediction of septic shock.Part ⅡThe perioperative data of 694 patients with PCNL were collected,and a univariate analysis of 35 variables was performed.Then,the prediction model of sepsis was constructed with 35 variables through the Random Forest(RF)algorithm,and the final prediction efficiency of the prediction model was measured by the area under the curve(AUC).ResultsPart ⅠSeptic shock occurred in 21 patients(5 males and 16 females).In both gender groups,the median SOFA score was 1(IQR: 0-2).In patients under 40 years of age,women’s SOFA score is higher than men’s(p = 0.048).There was gender difference in SOFA score after PCNL surgery,and the majority of patients with high scores were women(p = 0.011).Due to the high preoperative bilirubin and creatinine levels in male patients,the liver and kidney function sub-score in SOFA score was higher than that in female patients(p<0.05).The new SOFA score used to predict septic shock after PCNL was comparable to the original SOFA score(AUC: 0.987 vs.0.985,p = 0.932).However,while ensuring 100%sensitivity,the new SOFA score reduced the predicted false positive rate of septic shock by43.7% compared to the original SOFA score.Part ⅡUnivariate analysis results showed that the differences in height,weight,creatinine,urinary white blood cells,and maximum cumulative stone diameter of continuous variables and gender,urinary nitrite,solitary kidney,history of urinary surgery on the affected side,creatinine,urinary leukocytes,culture of midstream urinary bacteria,staghorn stones,and renal empyema of categorical variables were statistically significant between the two groups(P < 0.05).After evaluation,it was found that the average AUC of the sepsis prediction model was 0.73,indicating that the prediction ability of the model was good.In terms of the importance of variables,the importance of continuous variables is generally higher than that of categorical variables.The importance of preoperative urinary white blood cells,maximum cumulative stone diameter,preoperative creatinine,type of midstream urinary bacterial culture,white blood cells,age,preoperative antibiotic use time,bilirubin,weight and operation time were ranked in the top 10.ConclusionDifferences in SOFA scores can be caused by gender.This discrepancy between genders might not only underestimate the actual illness severity in women with septic shock but also reduce the SOFA predictive specificity for septic shock in men.Using the perioperative changed values of bilirubin and creatinine as sub-scores of hepatic and renal systems,the new SOFA scoring system eliminated the negative effects from gender difference of serum bilirubin and creatinine levels.In addition,based on the patient’s preoperative and intraoperative clinical data,the PCNL postoperative sepsis prediction model established by machine learning has good predictive power and can provide a reference for early identification of sepsis.