A web-based interpretable machine learning model for immediate postoperative delirium risk stratification in elderly patients undergoing colorectal cancer surgery

Postoperative delirium (POD) is a serious complication among elderly patients undergoing colorectal cancer surgery, yet accurate and clinically interpretable risk stratification remains challenging. This study aimed to develop and validate an interpretable machine learning (ML) model for immediate postoperative POD risk stratification in this specific population. This prospective observational study included consecutive patients aged ≥ 65 years who underwent surgical resection for colorectal cancer at a single medical center between March 2021 and May 2025. The dataset comprised a development cohort ( n = 922) used for model training and internal testing, and a subsequent same-center temporal validation cohort ( n = 394). Eight ML algorithms were developed and evaluated using the area under the receiver operating characteristic curve (AUC), calibration measures, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) analysis was employed to interpret feature importance. Among the eight models, the neural network (NNet) algorithm showed the strongest overall balance of predictive performance, with AUCs of 0.861 in the training set, 0.841 in the internal testing set, and 0.888 in the temporal validation cohort. In temporal validation, the model achieved an accuracy of 0.827, an F1-score of 0.714, and a Brier score of 0.130, with good overall calibration and clinical net benefit. SHAP analysis identified cognitive impairment, age, frailty, intensive care unit (ICU) admission, and the prognostic nutritional index (PNI) as the most influential predictors. A web-based calculator was developed to provide individualized POD risk estimates and risk classifications immediately after surgery. This prospective study developed and temporally validated an interpretable NNet model for immediate postoperative POD risk stratification in older patients undergoing colorectal cancer surgery. The web-based tool may support early identification, intensified surveillance, and timely preventive management for high-risk patients. Multicenter external validation is required before broader clinical implementation.