A scoping review of artificial intelligence in acute care surgerypromise, pitfalls, and a path forward

Abstract

BACKGROUND: Acute care surgery (ACS) faces unique challenges due to time-sensitive decisions, high diagnostic variability, and complex patient data. Artificial Intelligence (AI) offers potential solutions, yet the breadth, focus, safety, and translational readiness of current AI applications in ACS remain unclear. METHODS: A scoping review was conducted in PubMed, Scopus, and IEEE Xplore databases for peer-reviewed articles published 2015-2025. Data extraction included model architecture, data source, temporal and functional classifications, performance metrics, external validation, explainability, and risk of bias among others. Descriptive statistics and thematic synthesis were performed. RESULTS: Forty-nine studies describing 341 AI models were included. Most models (69.5%) originated from North America and primarily targeted preoperative prognostic tasks (76.5%). Large-scale registries (46.6%) and single-center studies (33.4%) were the primary data sources, and structured electronic health record data were predominantly used (91.5%), with minimal multimodal data integration. All models underwent internal validation, whereas external validation (20.2%), fairness assessments (3.2%), regulatory approval (0.2%), and adherence to standard reporting guidelines (34.3%) were limited. Model performance varied, with mean±SD (range): sensitivity 71.3±24.2% (10-100%), specificity 81.5±18.7% (5-100%), and AUROC 0.83± 0.11 (0.45-0.99) across studies. CONCLUSION: While AI holds promise for enhancing ACS, current applications are narrowly focused on pre-operative risk prediction using limited data modalities. Future efforts must prioritize prospective validation, real-time dynamic predictions, clinician-centered design, and multimodal data integration to realize AI's potential for improving ACS patient outcomes in time-sensitive, high-acuity surgical settings.

Ähnliche Arbeiten