Artificial Intelligence in Perioperative Risk Prediction: A Systematic Review of Clinical, Laboratory and Anesthesiology Applications

Abstract

The use of artificial intelligence (AI) in perioperative medicine is rapidly growing, with several applications in the field of risk stratification, clinical decision making and patient management, tailored to each stage of the perioperative journey, from pre- to intra- to post-operation. This systematic review explores the latest evidence surrounding the use of AI in perioperative risk prediction, specifically focused on perioperative clinical, laboratory, and anesthesiology applications. The review summarizes the evidence from studies that examined the use of machine learning (ML), deep learning (DL), natural language processing (NLP) and other AI methods to predict postoperative complications, mortality, hemodynamic instability, difficult airway, prolonged hospitalization, intensive care admission, and other clinically relevant outcomes. Special focus is placed on the incorporation of demographic data, comorbidities, physiological parameters, laboratory markers, electronic health records, imaging data, and intraoperative monitoring data into predictive models. The literature reviewed suggests that AI models can detect patterns and complex, nonlinear relationships that could supplement traditional models used in the perioperative setting. Laboratory-based and multimodal models can also be used to help identify patients at higher risk of adverse outcomes in the early stages, and anesthesiology-related applications have the potential to predict intraoperative hypotension, postoperative respiratory complications, and other anesthesia-related complications. But there's lots of variability in datasets, outcome definitions, model architectures, validation strategies and reporting. Routine use is hindered by issues of external validity, data quality, interpretability of results, algorithmic bias, privacy, workflow integration and prospective clinical testing. In summary, AI is a promising addition to contemporary perioperative risk stratification, however, its clinical utility will ultimately rely on thorough external validation, reporting, multi-disciplinary oversight, and thoughtful application alongside the clinical acumen of the provider, rather than an autonomous decision-making tool.