Artificial intelligence in preoperative assessment and optimization: a narrative review

Abstract

Background: Preoperative assessment and optimization are pivotal for complex surgical patients, yet traditional scores and clinician-driven workflows may miss subtle risk interactions and modifiable factors. Artificial intelligence (AI), especially machine learning (ML), can leverage large-scale electronic health record (EHR) data to improve risk stratification and support targeted optimization strategies. Methods: We conducted a narrative review focusing on practical, clinically actionable AI applications in preoperative assessment and optimization, emphasizing recent advances and implementation considerations for high-risk surgical patients. Results: ML-based preoperative risk models can outperform conventional calculators, exemplified by a gradient-boosted decision-tree model trained on >1.4 million surgical cases that achieved an AUROC of 0.95 for 30-day mortality and exceeded NSQIP performance, with prospective real-world deployment. Automated frailty phenotyping from structured preoperative EHR data (demographics, ASA/acuity, ICD-10/CCS diagnoses, and routine labs) has also been externally validated in older adults and shows strong stepwise associations with adverse outcomes. AI additionally supports preoperative optimization by identifying actionable targets such as anemia risk (supporting early iron/EPO pathways), penicillin allergy delabeling opportunities, and improved detection of risky alcohol use via natural language processing of clinical notes. Successful clinical impact depends on workflow integration, interpretability, and attention to privacy, bias, regulation, and prospective evidence. Conclusions: AI-enabled preoperative assessment can enhance identification of high-risk patients and systematically surface modifiable factors for optimization, with early evidence of improved predictive performance and feasible integration into point-of-care workflows. Future work should prioritize robust external validation across diverse populations, implementation studies demonstrating patient-centered benefit, and governance frameworks ensuring safety, fairness, and clinician trust.

Ähnliche Arbeiten