Construction of a Predictive Model of Gangrenous Cholecystitis Based on Machine Learning

This study reviewed the clinical data of 2023 cholecystectomy patients admitted to our center between January 1, 2015, and May 31, 2015, it includes demographic, clinical features, laboratory and imaging indexes, and constructs five commonly used Decision Tree, SVM, Random Forest, XGBoost, AdaBoost models, feature subsets are selected by Recursive Feature Elimination with Cross-Validation and the importance of variables in each model, model performance is evaluated by Balanced accuracy, Recall, Precision, F1score, and the Precision-Recall(PR) curve, and the final results are verified by independent external validation sets.

Inclusion criteria

• patients diagnosed with acute cholecystitis or acute exacerbation of chronic cholecystitis in our hospital and receiving complete clinical treatment in our hospital;
• performing cholecystectomy;
• having complete and searchable clinical data, such as patient's age, surgical records, and hospitalization days.

Exclusion criteria

• previous diagnosis of chronic cholecystitis, this time for elective surgical treatment;
• previous diagnosis of acute cholecystitis, ultrasound-guided cholecystectomy after elective laparoscopic cholecystectomy;
• concomitant with other acute biliary and pancreatic system-related diseases, such as obstructive jaundice caused by choledochal stones, acute cholangitis, acute pancreatitis, etc.;
• exclude patients who combined with other surgery patients such as choledochotomy and lithotripsy, choledochoscopic exploration and lithotripsy, bile-intestinal anastomosis, appendectomy, etc;
• those with incomplete data