Effectiveness of Methods for Predicting Postoperative Complications: A Literature Review

Background: Postoperative complications remain a major challenge in modern surgery, contributing to prolonged hospitalization, increased healthcare costs, morbidity, and mortality. To review the effectiveness of traditional and modern approaches for predicting postoperative complications and to assess the potential of artificial intelligence (AI) and machine learning technologies in surgical risk stratification. A review of recent literature was performed, focusing on clinical and laboratory predictors, conventional risk scores, machine learning algorithms, electronic health records, and dynamic perioperative data. Traditional systems such as ASA, POSSUM, P-POSSUM, and ACS NSQIP remain clinically useful but have limitations in capturing complex and nonlinear relationships among multiple risk factors. Machine learning models, including Random Forest, XGBoost, neural networks, and other algorithms, demonstrated promising predictive performance across various surgical outcomes. Models such as MySurgeryRisk and POTTER showed high discriminatory ability for postoperative complications and mortality. Integration of laboratory, physiological, intraoperative, and electronic health record data may further improve individualized risk prediction. However, substantial heterogeneity among studies, limited external validation, and the predominance of retrospective single-center studies remain important limitations. Additional concerns include data quality, class imbalance, algorithmic bias, model interpretability, and integration into clinical workflows. AI-based prediction models have considerable potential to improve early identification of high-risk surgical patients and optimize perioperative management. Nevertheless, prospective multicenter studies, external validation, standardized reporting, and evaluation of real-world clinical effectiveness are required before widespread implementation. AI should currently be considered a clinical decision-support tool that complements, rather than replaces, professional clinical judgment.