Artificial Intelligence-Based Predictive Model for Early Identification of Postoperative Complications Using Clinical and Functional Indicators

This is the authors' abstract. We don't add key points for this paper.

An artificial intelligence-based predictive model was developed for the early identification of postoperative complications using clinical and functional indicators. A total of 10,000 synthetic records from the TERMINET dataset were analyzed using logistic regression, Random Forest, and XGBoost, with stratified data splitting, cross-validation, and discrimination and calibration metrics. The models demonstrated exceptional predictive performance, with perfect classification achieved by logistic regression and Random Forest. SHAP analysis identified sodium, urea nitrogen, creatinine, and INR-related parameters among the main predictors. The findings demonstrate the computational feasibility of the proposed approach; however, the exceptionally high performance warrants cautious interpretation and subsequent external validation using real and independent clinical cohorts.