A major shift is occurring in artificial intelligence (AI) in perioperative medicine. Previously, AI was used to process large datasets, including patient records and research data, to identify associations and risk factors within a narrow range of specific requirements. This worked well within defined research parameters and only when the data could be curated. However, this approach proved difficult to integrate into clinical systems with their real-world, real-time demands and workflows.
Now, developments in AI, particularly machine learning (ML), are revealing capabilities that promise to revolutionise preoperative risk prediction, intraoperative monitoring and automation, detection of postoperative complications, and operational and workflow optimisation. ML can analyse vast datasets of high-dimensional data (data with multiple covariates), identify subtle patterns, and interpret nuanced or specialised clinical language within patient records. However, a gap remains between these promises and practical clinical application, and several concerns must still be addressed before these models can become part of routine practice.
In a panel discussion at the Euroanaesthesia Congress, speakers identified key areas in which AI can be implemented in anaesthetic practice and evaluated the supporting evidence. The session also addressed the potential benefits and risks associated with integrating AI into clinical workflows.
Predicting potential risks during preoperative assessment with high reliability supports clinical decision-making, improves patient safety, and contributes to better outcomes. AI now promises to predict personalised risks. Prof Bettina Jungwirth, Chairman of the Department of Anesthesiology and Intensive Care Medicine at the University of Ulm, Germany, discussed the use of AI in risk prediction, clinical decision support, and the translation of AI into perioperative workflows. Identifying personalised risk for each patient, without relying on intraoperative or postoperative data, represents an important development beyond conventional population-level prediction models.
AI and ML also show promise in forecasting critical events during surgery. One area attracting particular attention is predictive support for bleeding and transfusion risk. These predictions can help optimise workflows, estimate the duration of surgery and time spent in PACU, and reduce delays and cancellations. Prof Jens Meier, Chair Professor of Anaesthesiology and Intensive Care at Johannes Kepler University in Linz, Austria, discussed intraoperative AI applications such as predicting depth of anaesthesia, forecasting haemodynamic instability, and automated drug delivery through closed-loop systems. However, these developments are not without challenges. Any interaction between AI and humans in high-risk environments must support concentration and effective decision-making without increasing cognitive load. There are also important regulatory and legal questions surrounding the use of non-human intelligence in high-risk clinical decision-making.
As AI systems become more integrated into anaesthesia, caution remains appropriate. AI systems are capable of hallucination, producing false information with complete confidence. Another important issue is bias. Can AI exhibit bias even though it lacks human qualities such as prejudice or discrimination? Dr Sarah Saxena, Professor of anaesthesiology at the University of Mons, Belgium, said yes during her presentation on whether AI is gender-biased. She demonstrated that AI models exhibited notable biases in gender, race/ethnicity, and age representation, failing to reflect the actual diversity within the anesthesiologist workforce. This evidence of bias highlights the dangers of using AI uncritically. It is necessary to integrate critical AI literacy into medical curricula. Students need training, not just on how to use GenAI tools, but on when not to rely on them.
Source: Euroanaesthesia 2026
Image Credit: iStock
