Development and validation of a machine learning–based model for predicting textbook outcome after minimally invasive pancreaticoduodenectomy

Textbook Outcome (TO) has been proposed as a composite indicator reflecting overall surgical quality. With the expanding use of minimally invasive pancreaticoduodenectomy (MIPD), including laparoscopic and robotic-assisted approaches, reliable prediction of TO achievement has become essential. This study aimed to identify independent predictors of TO after MIPD and to develop an interpretable machine learning (ML)–based prediction model. We retrospectively analyzed 411 consecutive patients who underwent MIPD at Zhejiang Provincial People’s Hospital between 2017 and 2023. Clinical, operative, and pathological variables were collected. The Least Absolute Shrinkage and Selection Operator regression was used for feature selection, and ten ML algorithms were developed and compared. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), Brier score, and decision curve analysis. Model interpretability was assessed using SHapley Additive exPlanations (SHAP). Of 411 patients, 263 (63.99%) achieved TO. Eight variables were identified as independent predictors: pancreatic duct diameter > 3 mm, intraoperative blood loss, Nutritional Risk Screening 2002 score, direct bilirubin, triglyceride, carbohydrate antigen 125, male, and body mass index. Among the ten algorithms, the Random Forest model demonstrated the best discrimination (AUC = 0.86; Brier score = 0.15). SHAP analysis confirmed that larger pancreatic duct diameter and reduced blood loss were most strongly associated with TO achievement. An interpretable ML-based model accurately predicted TO after MIPD. The Random Forest algorithm demonstrated superior performance and may assist in individualized perioperative risk stratification and management.