Predictive Models and Artificial Intelligence for Risk Stratification in Emergency Abdominal Surgery: A Contemporary Review

Background/Objectives: Emergency abdominal surgery is associated with substantial morbidity and mortality due to disease severity, physiological instability, and limited opportunities for preoperative optimization. Accurate risk stratification is therefore essential for guiding clinical decision-making and resource allocation. This review aims to provide a contemporary overview of predictive models and artificial intelligence (AI) applications for risk stratification in emergency abdominal surgery. Methods: A narrative review of the literature was conducted using PubMed/MEDLINE, Scopus, and Web of Science databases. Studies addressing conventional risk assessment tools, predictive modeling, machine learning, deep learning, radiomics, explainable AI, and clinical implementation in emergency abdominal surgery were evaluated. Relevant publications concerning acute appendicitis, acute cholecystitis, intestinal obstruction, emergency laparotomy, abdominal sepsis, and acute mesenteric ischemia were included. Results: Conventional risk assessment systems, including ASA, APACHE II, SOFA, POSSUM, SORT, and NELA, remain widely used but are limited by static risk calculations and restricted adaptability. Recent advances in artificial intelligence have enabled machine learning and deep learning models capable of integrating complex clinical, laboratory, and imaging data to improve prediction of disease severity, postoperative complications, and mortality. The strongest evidence has been reported for acute appendicitis, acute cholecystitis, intestinal obstruction, abdominal sepsis, and emergency laparotomy, where several studies demonstrated excellent discriminative performance, frequently reporting area under the receiver operating characteristic curve (AUC) values exceeding 0.85. Emerging applications of radiomics, computer vision, and explainable artificial intelligence further enhance predictive capability and model interpretability. However, most available models remain retrospective, single-center developments with limited external validation and inconsistent reporting of calibration, highlighting the need for prospective multicenter evaluation before routine clinical implementation. Conclusions: Artificial intelligence has the potential to significantly enhance risk stratification in emergency abdominal surgery by enabling more precise and individualized prediction of adverse outcomes. Although current evidence is encouraging, robust prospective validation and responsible clinical implementation are required before AI-driven predictive models can be routinely incorporated into emergency surgical practice.