A Machine Learning Model For Predicting Laparoscopic Surgery Time And Early Postoperative Complications Of Congenital Choledochal Cyst

Part Ⅰ:A machine learning model for predicting surgical time of laparoscopic choledochal cystectomy and Roux-Y hepaticojejunostomyObjective:In this study,we use machine learning methods to research the factors affecting the surgical time of laparoscopic choledochal cystectomy and Roux-Y hepaticojej unostomy(RYHJ)in children with congenital choledochal cyst(CCC),and the result can be used to guide preoperative preparation and select the appropriate time for surgery.Methods:The preoperative data of children who underwent laparoscopic choledochal cystectomy and RYHJ in our hospital from October 2016 to October 2022 were retrospectively analyzed,and the surgery time of all the 120 children was recorded.The prolonged surgery time was defined as exceeding the 75th percentile of all,and all children were divided into the normal surgery time group and the prolonged surgical time group.In our machine learning model,all children were randomly assigned to the training and testing datasets in a 6:4 ratio,and three algorithms were used in the model:support vector machine(SVM),logistic regression(LR),and extreme gradient boosting(XGBoost).The receiver operating characteristic curve(ROC)was used to compare these three models,and the forecasting capabilities of the three models were evaluated by area under the curve(AUC)and other metrics.Finally,the model with the best prediction performance was selected.Then,we use SHAP(SHapley Additive exPlanations)to demonstrate post-hoc analysis of each model,evaluate the importance of the variables in the model,and the top 20 feature variables were selected for visualization.Results:XGBoost showed good predictive performance in terms of AUC,sensitivity,specificity and accuracy,which has an AUC value of 0.916 and it’s an ideal model for predicting the surgical time of laparoscopic choledochal cystectomy and RYHJ.The factors affecting the prolongation of laparoscopic surgery in children with CCC are mainly cyst maximum diameter,albumin,cholecystitis,and biliary stones.Among them,the maximum diameter of the cyst is the most important factor in predicting the prolongation of the operation time.Conclusion:This study suggests that machine learning models can be used to predict the relevant factors of laparoscopic surgery time in children with CCC,and the XGBoost model shows good predictive performance,which is a more ideal model for predicting the surgical time of laparoscopic choledochal cystectomy and RYHJ.This study may provide a clinical basis for selecting the appropriate timing of surgery and predicting the difficulty of surgery.Part Ⅱ:A machine learning model for predicting postoperative complications of laparoscopic choledochal cystectomy and Roux-Y hepaticojejunostomyObjective:In this study,we use machine learning methods to research the perioperative factors affecting the early complications of children who underwent choledochal cystectomy and Roux-Y hepaticojejunostomy(RYHJ)during hospitalization.The study aims to improve perioperative management and prognosis of children with congenital choledochal cyst(CCC).Methods:The preoperative and intraoperative data of children who underwent laparoscopic choledochal cystectomy and RYHJ from October 2016 to October 2022 in our hospital were retrospectively analyzed,and early postoperative complications of all children during hospitalization were collected.Then,they were evaluated according to the Clavien-Dindo grading system,and no complications and Clavien-Dindo grade Ⅰ were classified as the group without serious complications,and those with Clavien-Dindo gradeⅡ or higher grades were classified as the group with serious complications.All children were randomly divided into training and testing dataset at the ratio of 6:4,and three algorithms were used in the model:logistic regression(LR),support vector machine(SVM),and extreme gradient boosting(XGBoost).The receiver operating characteristic curve(ROC)was used to compare the models,and the area under the curve(AUC)and other metrics were used to evaluate the prediction performance of the three models separately,and the model with the best prediction performance was finally selected.Then,we use SHAP(SHapley Additive exPlanations)to demonstrate post-hoc analysis of each model,evaluate the importance of the variables in the model,and the top 20 feature variables were selected for visualization.Results:XGBoost is an ideal model to identify serious postoperative complications of laparoscopic choledochal cystectomy and RYHJ.Maximum cyst diameter,pancreaticobiliary maljunction,and protein plugs were important risk factors for predicting the occurrence of postoperative complications.Conclusion:This study suggests that machine learning models can be used to predict the influencing factors of early postoperative complications in children with CCC during hospitalization,and the XGBoost model shows good predictive performance,which is an ideal model for predicting postoperative complications in children undergoing laparoscopic choledochal cystectomy and RYHJ.Therefore,it can provide a basis for the reduction of postoperative complications and improving the prognosis of children with CCC.