EP 214 Artificial Intelligence and Machine Learning–Based Decision Support in General Surgery: A Scoping Review
S Klyde Alistar, Yuan Zher Yap, Yasmina Telek, Evripidis TokidisAbstract
Aims
Artificial intelligence (AI) and Machine Learning (ML) models could potentially support clinical decision making in General Surgery in various domains. However, the scope, performance reporting, and validation of AI/ML-based decision support tools across the specialty remain unclear. This review aims to summarise AI/ML applications in General Surgery and describe model performance reporting.
Methods
The study was registered on Open Science Framework. A literature search was conducted in PubMed, MEDLINE, Scopus, and Embase. Studies applying AI or ML to support decision making within General Surgery were included. Extracted data comprised study characteristics, subspecialty, clinical application, model type, validation strategy, and reported performance metrics. Studies were grouped by General Surgery subspecialty and clinical application.
Results
Of 1,282 identified articles, 164 met inclusion criteria; colorectal (n=75), hepatopancreatobiliary (n=48), upper gastrointestinal surgery (n=28), appendicectomy (n=4), laparoscopic unclassified procedures (n=6) and multiple subspecialties (n=3). AUC was the most commonly reported performance metric, although reporting was inconsistent and often did not distinguish between development and validation datasets. Reported AUC ranges varied widely across subspecialties (upper gastrointestinal: 0.64–0.99; colorectal: 0.49–0.99; hepatopancreatobiliary: 0.60–0.97). Few studies reported AUC for appendectomy or laparoscopic unclassified procedures, and calibration, clinical thresholds, and external validation were infrequently described.
Conclusion
ML and AI show potential to support decision making in General Surgery; however, the evidence base is heterogeneous and largely retrospective, with performance frequently reported using AUC alone, limiting assessment of clinical validity and applicability.