This study aimed to establish a predictive model to evaluate the surgical difficulty of laparoscopic cholecystectomy (LC) after percutaneous transhepatic gallbladder drainage (PTGBD) in patients with acute cholecystitis (AC) using preoperative parameters. Clinical data from 580 patients who underwent LCs at three tertiary hospitals between January 2013 and June 2024 were retrospectively analyzed. Predictors were selected through LASSO and Logistic analyses. Eight machine learning algorithms were used to create prediction models, evaluated by ten-fold cross-validation, ROC curves, DCA, calibration curves, precision-recall curves, confusion matrices, and DeLong’s test. SHAP analysis enhanced model interpretability. The intervals between onset and PTGBD, upper abdominal surgery histories, WBC, gallstones, gallbladder wall thickness, neutrophil percentage (Neut%), and time until PTGBD were identified as predictors. The Naive Bayes (NB) model demonstrated outperformed other models in clinical decision-making and prediction. SHAP analysis identified Neut% as the most critical prediction model feature, followed by the time until surgery after PTGBD. A online calculator was developed using the NB model ( https://zw17786325639.shinyapps.io/difficulty/ ). This study developed and verified a network calculator based on the NB model to predict LC surgical difficulty after PTGBD, aiding surgeons in preoperative risk assessment and timing selection.
Yang et al. (Fri,) studied this question.