Artificial Intelligence in Anaesthesia: A Systematic Review and Meta-Analysis of Clinical Outcomes

Abstract

Background: The integration of Artificial Intelligence (AI) and Machine Learning (ML) into anaesthesiology presents unprecedented opportunities to enhance perioperative care, optimize drug delivery, and accurately predict clinical outcomes. Current clinical paradigms rely heavily on generalized, subjective scoring systems that fail to capture the high-dimensional, non-linear physiologic changes occurring during surgery. This systematic review and meta-analysis aimed to evaluate the clinical efficacy of AI/ML interventions across diverse anaesthesia sub-specialties. Methods: Following PRISMA 2020 guidelines, a systematic literature search was conducted across PubMed/MEDLINE, Embase, Scopus, Web of Science, and Cochrane CENTRAL for studies published between 1 January 2000 and 1 May 2026. Primary clinical studies using AI/ML models for perioperative predictions and randomized controlled trials (RCTs) evaluating automated closed-loop delivery systems were included. Risk of bias was assessed using the PROBAST tool for predictive machine learning models, the Cochrane RoB 2.0 tool for interventional RCTs, and AMSTAR-2 for previously published systematic reviews. Quantitative meta-analysis was performed on overlapping continuous clinical outcomes using DerSimonian-Laird random-effects models. Results: From an initial pool of 19,300 articles, 85 primary studies were included in the qualitative synthesis, and 18 were included in the quantitative meta-analysis. In preoperative assessment, machine learning and deep learning models predicted difficult intubation with accuracies of up to 91.9%, outperforming standard clinical scoring systems. For intraoperative management, automated closed-loop delivery systems significantly reduced propofol induction doses (Mean Difference [MD] = -0.37 mg/kg; 95% CI [-0.57, -0.17]; P < 0.00001) and shortened recovery times (MD = -1.62 min; 95% CI [-2.64, -0.60]; P < 0.0001) compared to manual administration. In perioperative risk stratification, ML algorithms applied to electronic health records (EHR) accurately forecasted 1-to-24-month mortality and major complications such as acute kidney injury (AUC ranging from 0.73 to 0.89). Secondary outcomes included the prediction of acute hypotensive events minutes before their onset, objective pain assessment using facial expression analysis, and a reduced incidence of paresthesia during ultrasound-guided regional anaesthesia. Conclusions: AI algorithms offer robust predictive capabilities and improved precision in anaesthetic drug delivery, frequently outperforming traditional human-derived risk-scoring models and manual drug titration. However, widespread clinical implementation remains hindered by a lack of prospective, multi-center external validation and the inherent "black box" nature of deep neural networks. PROSPERO registration number: CRD420261329257.

Keywords: Artificial intelligence, Machine learning, Deep learning, Anaesthesia, Postoperative complications, Perioperative care, Closed-loop systems, Precision medicine.