Automation in perioperative medicine: perceptions, requirements and boundaries - a mixed methods study.

Automation in perioperative medicine: perceptions, requirements and boundaries - a mixed methods study - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Eur J Anaesthesiol Intensive Care . 2026 Apr 21;5(3):1-9. doi: 10.1097/EA9.0000000000000107 Automation in perioperative medicine: perceptions, requirements and boundaries - a mixed methods study Jeffrey David Iqbal 1 , Maximilian Lindholz 1 , Stefan J Schaller 1 , Martin Schlaepfer 1 Author information Article notes Copyright and License information 1 From the Faculty of Medicine, University of Zurich, Zurich, Switzerland (JDI, MS), Berlin Institute of Health at Charité (BIH), Berlin, Germany (ML), Charité – Universitätsmedizin Berlin, Institute of Medical Informatics, Berlin, Germany (SJS), University Hospital Zurich, Institute of Anesthesiology and Perioperative Medicine, Zurich, Switzerland (JDI, MS), Charité – Universitätsmedizin Berlin, Department of Anesthesiology and Intensive Care Medicine (CCM/CVK), Chariteplatz 1, Berlin, Germany (ML, SJS), Charité – Universitätsmedizin Berlin, Department of Radiology, Berlin, Germany (ML), Medical University of Vienna, Department of Anaesthesia, Intensive Care Medicine and Pain Medicine, Division of General Anaesthesia and Intensive Care Medicine, Vienna, Austria (SJS) ✉ Correspondence to Jeffrey David Iqbal, PhD University Hospital Zurich, Raemistrasse 100, Zurich 8091, Switzerland. e-mail: jeffreydavid.iqbal@uzh.ch Received 2024 Dec 11; Accepted 2026 Mar 9; Collection date 2026 Jun. Copyright © 2026 The Author(s). Published by Wolters Kluwer Health, Inc. on behalf of the European Society of Anaesthesiology and Intensive Care. This is an open access article distributed under the terms of the Creative Commons Attribution-Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to download and share the work provided it is properly cited. The work cannot be changed in any way or used commercially without permission from the journal. http://creativecommons.org/licenses/by-nc-nd/4.0 PMC Copyright notice PMCID: PMC13232904  PMID: 42244824 Abstract BACKGROUND Integrating artificial intelligence (AI) and automation can potentially advance perioperative medicine, especially in the maintenance phase of general anaesthesia. Recent technological advancements, rising costs, a shortage of anaesthesiologists and the potential for improved quality make a strong case for increased utilisation of these innovations. Despite instances where machines have outperformed humans, automated anaesthesia systems have not been widely adopted. OBJECTIVES This research paper aims to explore opportunities, challenges, restrictions and requirements for the further automation of anaesthesia, focusing on the intraoperative maintenance phase of anaesthesia. Addressing this gap is crucial for realising the full potential of AI and automation in perioperative care, ultimately leading to improved patient outcomes and healthcare efficiency. DESIGN This explorative mixed methods study is based on expert interviews, including both qualitative and quantitative survey components. SETTING Two university medical centres in Switzerland (University Hospital Zurich) and Germany (Charité University Medicine Berlin) PARTICIPANTS Forty-two anaesthesiologists (24 residents, 18 consultants) with a mean tenure of 8.6 years (3.5 for residents and 15.4 for consultants; range 1–29 years) were interviewed. MAIN OUTCOME MEASURES The primary endpoints were thematic descriptions due to the exploratory nature of this study. RESULTS Anaesthesiologists interviewed welcome AI, list automation aspects as desired functionalities and see its potential for error reduction. Attitudes towards automation were influenced by the level of automation involved. While augmentative support received unanimous approval, replacement scenarios with high machine autonomy were seen more critically despite diverse opinions. Several opportunities and challenges were identified for stakeholder groups such as anaesthesiologists, nurses, patients, surgeons, hospitals and payers. Further, requirements and restrictions were identified to which automation systems should be subject. CONCLUSIONS Residents and board-certified anaesthesiologists are convinced that technology, including AI and automation, has the potential to address challenges in anaesthesia practice. They list concrete requirements and restrictions for their use and expect a range of opportunities and challenges. With increased adoption, the roles and responsibilities of anaesthesiologists and other perioperative healthcare providers will change. Given both adoption and technological barriers, the complete replacement of anaesthesiologists by machines seems unlikely in the mid-term. KEY POINTS Question: What are the opportunities, challenges and constraints of integrating artificial intelligence (AI) and automation into anaesthesiology practice during the maintenance phase of general anaesthesia? Findings: Anaesthesiologists acknowledge the potential of AI and automation in anaesthesiology, list concrete requirements and restrictions and favour augmentative support while expressing caution toward high-autonomy systems. Meaning: AI and automation could address challenges in anaesthesia, but full automation replacing anaesthesiologists is unlikely in the mid-term due to significant adoption and technological barriers identified. Introduction Anaesthesiology has a long history of adopting innovative technologies such as electrocardiography (ECG), electroencephalography (EEG) and sonography, which have significantly improved patient safety and comfort by mitigating risks inherent in the practice. Despite this tradition of embracing advancements, the current wave of artificial intelligence (AI), automation and robotics has yet to gain widespread traction in anaesthesiology. 1 This delay is surprising, given the increasing demand for healthcare services due to aging populations and underserved communities. By 2034, the United States could face a shortfall of up to 16 000 anaesthesiologists, 2 , 3 a trend mirrored in other countries like Switzerland. 4 This shortage, coupled with rising healthcare costs 5 and potential for quality improvement, underscores the need for more efficient solutions in perioperative care. While several AI-driven technologies, such as those for predicting hypotensive episodes 6 and optimising regional anaesthesia, 7 are under development, they primarily serve as assistive tools rather than fully automated systems. Efforts to introduce closed-loop anaesthesia management, which regulates anaesthesia based on real-time feedback from multiple variables, have been ongoing for years. 8 However, despite showing promise in clinical trials, 9 these systems have struggled to achieve clinical adoption. Technological advancements continue, with newer closed-loop systems now integrating the management of various components of anaesthesia, such as sedation, analgesia and haemodynamics. 10 Despite these advancements, challenges remain in realising the full potential of automation in anaesthesiology, highlighting the need for further research and development to bridge the gap between innovation and implementation. Despite many engineering and model-focused publications, opinions, comments and editorials in this area, no systematic empirical research has been published on what opportunities, challenges, restrictions and requirements exist for further automation of anaesthesia. This gap must be addressed to start paving the way and setting boundaries for the potential adoption of automated anaesthesia. Therefore, this research paper aims to explore opportunities, challenges, restrictions and requirements for the further automation of anaesthesia, focusing on the intraoperative maintenance phase of anaesthesia. As exploratory research, the objective is to generate an overview with hypotheses, not to provide definitive proof. Methods A mixed-methods approach consisting of structured expert interviews and integrated quantitative survey components was chosen. The scope of interest included closed-loop systems for multiple target parameters to be controlled simultaneously (multiparameter adaptation, for example, sedation, analgesia, muscle relaxation, ventilation). Local and regional anaesthesia methods were excluded from the scope. The scope also did not include TCI and closed-loop systems for single target parameters to be controlled (e.g., only muscle relaxation). The institutional review boards of the Medical Faculty of the University of Zurich, Switzerland (waiver dated 21 December 2022) and Technical University of Munich, Germany (waiver dated 7 March 2023) confirmed that the research was not subject to approval under applicable human research laws. Written informed consent materials were provided in German, and signed consent forms were obtained from each participant. Sample: anaesthesiologists during and after residency Two types of participants were recruited: resident physicians in anaesthesiology (residents) and board-certified anaesthesiologists (consultants) at two university hospitals (Charité – Universitätsmedizin Berlin, Germany and University Hospital Zurich, Switzerland). Recruitment was done via the personal approach of physicians. Once interest was signalled, potential participants received detailed written information, the questionnaire and an informed consent statement document. Participants were not provided with a payment or any other incentive for their participation. A target number of at least 10 to 15 individual interviews per hospital was defined. Demographic data were limited to board certification status and tenure in years due to research ethics considerations (deanonymisation risk to participants). Practicing anaesthesiologists were chosen over key opinion leaders/automation experts in order to gain fresh, less biased perspectives on the topic. Questionnaire and interviews A semi-structured English language interview guide (Appendix 1, Supplemental Digital Content) was designed by the authors and used during the individual interviews conducted 1 : 1 in the hospitals by the first author, a Postdoc-level qualitative researcher and medical student. Participants were allowed to answer in English or German. While four conceptual areas were prearranged, interviews were conducted openly to not limit and distort the answers by rigid interviewer action and to allow the expression of information not preconceived. The four conceptual areas were artificial intelligence, automation, autonomy & supervision and human–machine interaction & situational awareness. The quantitative survey component was set up to determine attitudes on automation and artificial intelligence, using Likert-style items, and served to contextualise the answers obtained in the qualitative part. Data extraction and qualitative content analysis Audio recordings of the interviews were taped and transcribed into text format, after which the audio recordings were deleted. Single-coder coding and qualitative content analysis (identification of concepts and categories) were conducted according to grounded theory 11 using the software MaxQDA 24 with themes derived from the data according to the semi-structured interview guide. All researchers reviewed coding and categorisation. Grounded theory is a qualitative social science methodology that does not rely on prior hypotheses but can form hypotheses to be examined in a subsequent step. Quantitative data were extracted from the transcripts and compiled using Microsoft Excel 365. Results Forty-two anaesthesiologists (24 residents, 18 consultants) with a mean tenure of 8.6 years (3.5 for residents and 15.4 for consultants; range 1–29 years) were interviewed between August to October 2023. The mean interview time was 38 min (range 19 to 56 min), and the interviews were balanced between centres Charité – Universitätsmedizin Berlin (19 interviews) and University Hospital Zurich (23 interviews). Although more residents ( n = 24) than consultants ( n = 18) were interviewed, we analysed responses stratified by training level, and thematic saturation was reached in both groups. Attitudes of anaesthesiologists towards artificial intelligence Anaesthesiologists agree that AI has the potential to play a role in anaesthesiology Given the lack of a commonly accepted definition of AI, 12 participants were asked about their understanding of the term AI. Despite various perspectives, the analysis revealed that 81.5% of codes predominantly addressed the process dimension, emphasising the machine-driven data conversion into actionable information and decisions. Notably, unanimous consent was observed among all participating anaesthesiologists regarding the potential of AI in shaping the future of anaesthesiology. Specifically, an overwhelming 97.6% of anaesthesiologists agreed on AI's potential to contribute to the maintenance phase of general anaesthesia during surgical procedures. There are already artificial intelligence technologies in use in anaesthesiology clinical practice today EEG interpretation scores such as Patient State Index (PSI) and Bispectral Index (BIS), sepsis screening and warning systems based on Electronic Health Records (EHR), and arterial pulse wave form-based hypotension prediction were named as examples of AI technologies already in today's clinical use. Other monitoring and early warning systems, such as intelligent alarms with high specificity and sensitivity, forecasting of postanaesthesia awakening time, preoperative risk evaluation and closed-loop sedation, ventilation, hemodynamic control and neuromuscular relaxation systems, were given as examples of systems currently in research and development (R&D). Potential positive impact of artificial intelligence use on error reduction When asked about the impact of AI on the reduction of medical errors, including medication errors – the most common type of error in anaesthesiology 13 – 73.8% of respondents expect AI would reduce error rates. All others (26.2%) were undecided; none expected an increase in errors. Advocates for error reduction cited various reasons, including the prevention of inadvertent administration of incorrect drugs or doses, augmentation of anaesthesiologists’ vigilance, attention and focus, avoidance of drug administration despite contraindications or allergies recorded in EHRs and precise timing of drug administration. Anaesthesiologists generally welcome artificial intelligence Anaesthesiologists are undecided on whether AI may replace them. When confronted with the statement that fewer anaesthesiologists will be needed given constant demand for anaesthesia procedures and the introduction and use of AI on Likert response categories of 1 (disagree strongly) to 5 (agree strongly), the median answer was 4 [2 to 5]. Consultants scored this item lower, 3 [1 to 5] than residents, 4 [3 to 5]. For the statement “I am welcoming the increased use of AI within the practice of anaesthesia”, there was general agreement: all anaesthesiologists, 5 [4 to 5]; consultants, 5 [4 to 5]; residents, 4 [3 to 5]. The desire for artificial intelligence support and other technologies is present When anaesthesiologists were asked to outline their desired functionalities for AI technologies to streamline perioperative patient management, a diverse array of preferences emerged, documented in Table 1 . These preferences are structured according to the perioperative workflow. Table 1. The “AI wish list” – items represent functionalities desired by anaesthesiologists Preoperative Intraoperative Postoperative Preoperative risk evaluation support Prediction of complications & critical events Prediction of complications (postoperative pain, PONV, etc.) End-to-end check of requirements (checklists, discipline consultations, etc.) Maintenance of C (circulation) only during critical phases to enable focus on other aspects Post hoc feedback on anaesthesia to anaesthesiologist (evaluation & education tool) Automated/supported anaesthesia introduction Ventilation optimisation (sufficiency vs. protection) Estimate/steering of awakening Intubation robot Pharma robot calculating and injecting drugs (from battery of prespecified range of drugs) Reliable sedation/hypnosis, relaxation, noninvasive blood pressure, volume monitoring Peri-operative Automation of anaesthesia documentation Cross-check control function (second pair of eyes) Wireless monitoring Recommendations for action integrating all data sources (incl. laboratory results) Guidance system with Head Display Warning system monitoring vigilance status of anaesthesiologist Voice-guided surgical table movement for surgeons Dashboard for remote management of multiple procedures Open in a new tab Grades of automation in anaesthesiology We introduced an adapted medical version of a machine autonomy grading scheme, 14 commonly used in aviation and automobile sectors, to anaesthesiologists for feedback. While retaining the fundamental five-level structure, we included additional subsections under level three following participant feedback (Fig. 1 , Table 2 ). Fig. 1. Open in a new tab Conceptual framework illustrating increasing levels of machine autonomy in anaesthesia, from data presentation (Level 1) to full automation without active clinician involvement (Level 5). Intermediate levels involve progressively greater system autonomy with varying degrees of clinician oversight., modified from Bitterman et al. 14 . Table 2. Responses were collected using a five to point Likert type item response scale (1 = least desirable, 5 = most desirable) Level 1 Level 2 Level 3 Level 4 Level 5 All 5 [5 to 5] 5 [4 to 5] 4 [3 to 5] 2 [2 to 3] 1 [1 to 2] Consultants 5 [5 to 5] 5 [4 to 5] 5 [3 to 5] 3 [2 to 3] 2 [1 to 2] Residents 5 [5 to 5] 5 [5 to 5] 4 [3 to 4] 2 [1 to 3] 1 [1 to 2] Open in a new tab Likert responses were treated as ordinal data; median [IQR] therefore correspond to observed response categories. Data are median [IQR]. Level-1 systems are designed to present data in a structured format to clinicians, enhancing their effectiveness and efficiency. For instance, a tool extracting allergies, medication lists and previous hypotensive episodes from EHR for preoperative consultations. Level-2 systems extend beyond data presentation to suggest drug regimens and control limits tailored to individual patients, leveraging updated guidelines and large patient population data inaccessible to clinicians otherwise. Level-3 systems introduce the concept of closed-loop anaesthesia, with (3a) systems requiring clinician confirmation of system decisions and (3b) systems executing decisions autonomously within a defined time window, allowing clinician intervention if necessary while remaining “bedside”. At Level-4, clinicians transition from the one “bedside” to monitoring multiple rooms, similar to supervising physicians overseeing residents or nurse anaesthetists. Level-5 represents full automation, where clinicians are no longer actively involved in anaesthesia execution. Such systems would require medium to low use case specificity, capable of handling unplanned scenarios, including emergencies, beyond the planned scope of anaesthesia procedures. Except for one resident, all anaesthesiologists (97.6%) confirmed they would accept the use of Level-3 systems in their practice, while only 78.6% of anaesthesiologists (83.3% of consultants, 75.0% of residents) would do so for Level-4 systems. When asked the number of low-risk patients one anaesthesiologist could supervise in parallel in a Level-4 scenario, the median was 4 patients [range, 2 to 100]; only 30/42 anaesthesiologists chose to respond with a number stating their uncertainty. A differentiated picture evolved when anaesthesiologists were asked about the desirability of the increasing use of Level-1 to Level-5 systems within anaesthesia practice using a five-point Likert-type item response scale of 1 (least desirable) to 5 (most desirable). While Assistive AI systems are much desired, systems with higher grades of autonomy (Levels-4 and -5) may be seen more skeptically. Opportunities/benefits and challenges/disadvantages of closed-loop automation Anaesthesiologists mentioned many opportunities and challenges for stakeholders: anaesthesiologists, nurses, surgeons, patients, health insurers/payers and hospitals (Table 3 ). Regarding nurse anaesthetists (equivalent to the U.S. CRNAs, a profession prevalent in Switzerland but not Germany), respondents emphasised that benefits and risks would be comparable to those applicable to anaesthesiologists. Consequently, the term ‘nurses’ refers specifically to nonanaesthetist nurses fulfilling supportive roles. Table 3. Themes mentioned by anaesthesiologists reflect perceived stakeholder-specific opportunities and challenges for key stakeholder groups anaesthesiologists, nurses, surgeons, patients, insurances & payers as well as hospitals Stakeholder group Opportunities & benefits Challenges & disadvantages Anaesthesiologists Reduced workload Reduced job security Focus on high-value, core aspects of care Overreliance on system Error reduction Loss of basic anaesthesia skills Focus on most relevant aspects of situation Difficulty of training residents, future anaesthesiologists Enrichment, more interesting work Unclear liability More pleasant work environment Automation bias Superior performance of AI over human May get boring, unsatisfying with restricted “freedom to play” Reduced work hours Vigilance requirement demanding Nurses Reduced workload Reduced job security No impact expected Downskilling potential Focus on high-value, core aspects of care Missing dangerous situations Up skilling potential restricted “freedom to play”, machine paternalism Optimised workflows, easier planning on required materials Concentration of work Surgeons No impact expected Lack of safety net and assurance by anaesthesiologist Improved, more predictable anaesthesia Risk of malfunction or other low performance Less conflict potential between surgeon and anaesthesiologist, quiet environment – “Anytime anaesthesia” availability by turning a switch – Same effectiveness at lower cost – Patients Increased patient safety Reduced patient safety Increased patient comfort Lack of trust in machine Shorter induction and recovery times Training data bias and inability to handle complex patients Patient empowerment vs. physicians Anaesthesiologists’ skills might be impaired due to a higher levels of automation Automation perceived favourably Unclear liability & legal situation Insurances & payers Cost reduction (avoidance of medical errors, not-best care) Unclear liability & legal situation Cost reduction (personnel) Technology expensive to introduce Cost reduction (material) Lack of incentives for adoption by providers Cost reduction (infrastructure) Threat of price-gouging, artificial shortages Higher cost transparency – Hospitals Less recruitment pressure High initial investment needed Cost reduction Additional physical space required Marketability of AI “Modern/cutting edge provider” Threat of price-gouging, artificial shortages Capacity increase – Open in a new tab Increased and reduced patient safety using closed-loop automation were mentioned 43 and 14 times in the respective interview sections, respectively. Anaesthesiologists argued that the following mechanisms could contribute to increased patient safety (in descending order of frequency of mention): (1) Enhanced adherence to guidelines and evidence-based practices. (2) Improved medication management, encompassing delirium prevention, prevention of intraoperative awareness, mitigation of neurotoxicity risks and expedited administration of emergency medications. (3) Quality control mechanisms to monitor the performance of anaesthesiologists. (4) Capability to address rare events through preprogrammed experiences. (5) Continuity of patient care amid changes in team composition during prolonged surgery. Requirements for the use of automated anaesthesia systems Anaesthesiologists mentioned several prerequisites to adopting Level-3 and Level-4 automated anaesthesia systems. They are structured in Table 4 . Table 4. Requirements were derived from qualitative content analysis of expert interviews with anaesthesiologists Requirement group Requirement Examples of codes Performance Tested and approved Systems must be rigorously tested for and used only within specified use cases and limits; must be approved as medical devices User friendliness Convenience of data acquisition (minimise manual inputs), Design according to user experience/user interface (UX/UI) best practices Transparency Limits of operation must be known to user; system recommendations must be backed up with reference to evidence Human backup, fail safe Human anaesthesiologist must be available and able to take over if needed Data quality Training data & models need to reflect breadth of use cases, Interconnectability with multiple input sources needs to be ensured Input integration Situational monitoring Live inputs of what interventionist/surgeon is doing, e.g. cauterisation, must go into system Relevant inputs currently not digitally available must be considered by system Blood loss, operating table position & body position changes, colour/temperature of extremities Controls Override User must be able to override and take control; spatial proximity required to intervene; user must be able to override system check programmes and controls at all times Opt-in/opt-out Anaesthesiologist must have choice to use or not to use Control limits User must be able to set alarms, automated stops according to individual preferences Camera supervision User must be able to always see OR and patient if not in room Hospital processes Sufficient supervisory staff Sufficient capable staff must be on hand to intervene, also in case of multiple emergencies Roles & responsibilities (R&R) Clearly assigned R&R with standard operating procedures (SOPs) must be in place; workflows will need to be clearly defined Users Training Users must be rigorously trained “like a commercial driving license with regular checks” AI Literacy Basic statistics and data science understanding must be strengthened in medical training Creation Co-creation Industry must co-create this innovation with anaesthesiologists, not without Legal Liability Liability assignment must be transparent with regulatory class Use framework Machine–human complex becomes relevant unit that is to be governed; upper limits on care ratio to be established Open in a new tab Categories represent themes identified during grounded theory–based coding, with examples summarising representative statements from participants. User experience (UX) and user interface (UI) refer to the design of digital products so that they are easy, efficient, and pleasant for users to interact with. Restrictions for the use of automated anaesthesia systems Anaesthesiologists held varied perspectives regarding the restrictions or conditions of use for automated anaesthesia systems. While a notable consensus emerged on incorporating the ASA classification of patients as a criterion for determining use, divergence was evident on specific patient categories. A few physicians argued that ASA 4 and 5 patients should be excluded from use. In contrast, others argued that this group would profit most from using automated anaesthesia systems and that no patient should be excluded in principle. Both groups agreed that a rollout of these systems should start with less complex patients classified as ASA 1 or 2. Intrahospital polytrauma and resuscitation were the most mentioned use cases that should be excluded. Informed consent for the use of Level-3 systems was considered necessary by 68.3% of respondents, while 92.5% considered it essential for Level-4 system use. Proponents argued that obtaining informed consent would be required initially until the technology was established as standard practice and/or that informed consent always needed to be achieved, while opponents mentioned it was just another tool in the toolbox of anaesthesiologists. Some noted the patient should be given a choice between machine or human controlled anaesthesia. In contrast, others argued informed consent should entail a mention of machine use with its risks and benefits but not choice. Three distinct viewpoints emerged regarding the seniority levels of anaesthesiologists most likely to benefit from automated anaesthesia system use: one group believed junior anaesthesiologists would benefit the most, as the systems could compensate for their lack of experience; another group contended that junior anaesthesiologists should first master traditional manual techniques before transitioning to automated systems; while a third group asserted that all anaesthesiologists, regardless of seniority, could derive immediate benefits from the adoption of automated systems. It was proposed that for junior and tech-savvy clinicians such systems may be easier to get accustomed to, but that anaesthesiologists who do not want to work with such systems should not be forced to, as this could risk creating a self-fulfilling prophecy. The most enormous opportunity was seen for hospitals and clinics covering highly standardised surgical cases, such as gastric bypass procedures or those with primarily long, steady-state surgeries, such as hand surgeries in low-risk patients. These would represent an opportunity for Level-4 systems where one anaesthesiologist could supervise several operating rooms at a time. A group of anaesthesiologists argued that university hospitals should be the first to take up the technology to collect experience and to serve as first movers, as they have the necessary expertise for evaluation but also because of their role in training residents. Anaesthesiologists also argued that in the later stages of adoption, there would be no difference between smaller/regional hospitals versus large-centre hospitals regarding who can benefit most. Benefits to smaller/regional hospitals provided by the technology may include an increase of available expertise that would otherwise be missing: e.g. the potential for tele-anaesthesia, and associated ease of recruiting and retention of staff. The impact of automation on situational awareness is unclear Situational awareness is the “perception of the elements in the environment within a volume of time and space, the comprehension of their meaning and the projection of their status in the near future”. It is deemed an essential component of effective decision-making. 15 Previous research outside of medicine points to the risk of a person's situational awareness decreasing with the increasing autonomy of a system. 16 Anaesthesiologists were undecided on the impact of closed-loop anaesthesia systems in that 40.5% of respondents (38.9% of consultants, 41.7% of residents) said such systems would reduce the situational awareness of anaesthesiologists, 31.0% (38.9% of consultants, 25.0% of residents) expected an increase in situational awareness and 28.6% (22.2% of consultants, 33.3% of residents) remained neutral. Those with a negative sentiment argued that anaesthesiologists would over-rely on the system, including its monitoring functions, or may be overwhelmed by information overload. Those with a positive sentiment argued that such systems would enable a faster reaction by the anaesthesiologist to issues, nudge anaesthesiologists into action, lower their cognitive load, allow them to focus on the essentials and aid them in predicting the future state of the patient. An array of mitigation and support measures on how to keep up or increase situational awareness of anaesthesiologists working with closed-loop systems was provided. Intelligent alarms should have high specificity and could be paired with a suggestion on how to react; human confirmation could be required before drug doses or other changes, and the system should require user interaction or ask for patient-specific information (e.g., “How has the heart rate changed in the last ten minutes?”). Intuitive graphical representations and regular training measures should be included. Discussion This mixed-methods study provides an overview of opportunities, challenges, requirements and constraints of implementing automated anaesthesia systems and contextualises them with anaesthesiologists’ attitudes towards AI and automation with a focus on the anaesthesia maintenance phase. The interviews provide evidence that anaesthesiologists of different training stages generally believe that AI and automation can address many challenges in anaesthesiology. One such area is closed-loop management, which considers individual variability to achieve optimal outcomes by monitoring the depth of anaesthesia and reducing side effects such as postoperative nausea and vomiting, and delirium. In addition, preventing human error – a critical factor in anaesthesia errors 17 – might contribute to quality improvements in anaesthesia. When inexperienced or fatigued care providers are supported by technology, common errors 18 could be avoided. 19 Although a definitive statement on the exact impact of automation systems on anaesthesiologists’ situational awareness has yet to be made, it is plausible that the need for such awareness among anaesthesia caregivers will change with the introduction of rigorously tested, approved and label-compliant automated anaesthesia machines – in line with expert opinion. 20 Consequently, there is a likelihood that errors in situational awareness can be reduced in all three types – perception, comprehension and projection – which account for 81.5% of errors in anaesthesia and critical care. The most daunting challenge lies in navigating the technical-clinical trajectory toward achieving utmost reliability across the diverse cases encompassing myriad difficulties of OR scenarios. The lack of AI technical support personnel and quality assurance mechanisms, across all anaesthesia providers but more likely in smaller hospitals and practices, may pose significant implementation risks if system failures occur. This underlines the importance of developing vendor-independent certification standards and maintenance frameworks to ensure reliability and accountability across diverse healthcare settings. The roles and responsibilities of anaesthesiologists will change The profession of anaesthesiology is likely to change with the adoption of AI and automation technology. 22 Rather than “change-resistant” anaesthesiologists being excluded from use, there will likely be a transition phase with sequential progression along the automation grade spectrum with skills growing and attitudes changing alongside technology. While some have argued that human supervision in intraoperative automation will always be indispensable, 23 this automation grade progression may be primarily a function of time if the requirements and restrictions identified here are fulfilled. Along this timeline, automated anaesthesia systems in Levels 2 and 3 could serve as educational tools for resident and consultant anaesthesiologists learning from and having access to up-to-date guidelines, evidence and real-world data of (ideally) millions of cases and adding to an array of approaches in anaesthesiology education. 24 Such systems could be run in different modes such as nudging, recommendations, action or postoperative feedback. The least invasive mode, – postoperative feedback on anaesthesia performed by a human anaesthesiologist – is what many residents regularly complain is lacking from their supervising consultants. This may also provide an opportunity to speed up residency and increase the supply of anaesthesiologists. The shift from patients to patient-consumers 25 will also apply to perioperative medicine and result in increased shared decision-making. Anaesthesiologists may spend less time in an OR and more time establishing perioperative treatment plans together with patient-consumers. Work may, therefore, become more complex and enriched, emphasising this human-to-human interaction. In the context of the shift of work away from the operating theatre and the trend towards shorter working hours, 4 AI systems may not only help shorten working hours but also improve the quality and efficiency of perioperative care by supporting anaesthesiologists in performing complex tasks within a reduced timeframe. The complete replacement of anaesthesiologists by machines appears unlikely in the mid-term The successful adoption of new technologies and associated workflows in healthcare requires robust clinical evidence. Additionally, addressing ethical challenges such as obtaining patient consent, ensuring proof of safety, and maintaining data privacy is crucial. Legal challenges must also be managed, including regulatory approval and liability issues. 26 Ongoing legal-regulatory initiatives, such as the EU Acts on Artificial Intelligence, Data and Data Governance, complicate this landscape further. Revising reimbursement policies could be necessary if cost reductions and quality improvements are achieved. 27 These factors can be managed without significant support from anaesthesiologists. However, anaesthesia societies must develop clinical guidelines and recommendations for using AI in perioperative medicine, which will likely enhance patient and clinician acceptance. Previous efforts, such as McSleepy, iControl-RP and Sedasys have failed to either gain industrial backers 28 or market acceptance. 29 , 30 Ultimately, any commercialisation faces high risks of failure given the lack of mentioned adoption antecedents. Strengths and limitations The strengths of this study include the large number of interviews, which achieved theoretical saturation, and the inclusion of two hospitals in different countries, thereby avoiding centre bias and broadening the applicability of the findings. The study also analysed residents and board-certified anaesthesiologists separately and employed a structured methodological approach. However, the methods used have inherent limitations. The experts interviewed were a convenience sample and did not represent the broader population of anaesthesiologists in Switzerland and Germany. Additionally, all participants worked at academic hospitals, whereas anaesthesia practice varies widely, such as in smaller hospitals and private practices. The sample included more residents than consultants, which may influence the relative weight of certain perspectives. While this imbalance was modest and both groups were analysed separately, the findings may slightly reflect the views of early-career anaesthesiologists more strongly. As the subject of automated anaesthesia is mainly hypothetical, the information provided pertains to potential future developments and is inherently uncertain. Nonetheless, this study is a starting point for further research on anaesthesia automation and offers valuable insights. Conclusion and outlook In conclusion, the study indicates that residents and board-certified anaesthesiologists are generally convinced that technology, including AI and automation, can address existing challenges in anaesthesia practice. They list concrete requirements and restrictions for the use of higher grade automation systems. The interviews highlight that adopting such techniques is accompanied by a change in the roles and responsibilities of anaesthesiologists and other perioperative staff. Given both adoption and technological barriers, the complete replacement of anaesthesiologists by machines appears unlikely in the mid-term. Anaesthesiology faces multiple clinical and operational challenges that AI could help mitigate. Clinically, increasing case complexity and multimorbidity heighten the risk of human error, particularly in drug administration and haemodynamic control, amid high cognitive load and dynamic intraoperative conditions. Operationally, the specialty is affected by workforce shortages, night shift fatigue and rising documentation demands, which can compromise vigilance and focus. AI offers potential solutions through clinical decision support, closed-loop systems and intelligent monitoring. Such tools may enhance adherence to best practices, reduce medication and monitoring errors, and offload routine tasks to maintain physiologic targets with greater precision. AI may also potentially improve situational awareness by processing complex data into predictive insights and intuitive visualisations, while intelligent alarms help reduce alert fatigue. From a training perspective, AI could deliver personalised performance feedback and simulate rare scenarios, addressing gaps in supervision. Automation of documentation through natural language processing may further reduce nonclinical burden. Finally, AI may help extend anaesthesia expertise to smaller or understaffed institutions, supporting tele-anaesthesia models and standardising care across settings. These advances, if developed and governed appropriately, may enable more efficient, safer and equitable anaesthetic care. 21 Supplementary Material Supplemental Digital Content ejaic-5-e0107-s001.docx (50.4KB, docx) Acknowledgements relating to this article Assistance with the study: the authors thank the anaesthesiologists who participated in this study for their time and valuable input. Financial support and sponsorship: the authors are thankful for the support by the Graduate Campus (GRC) and the Global Affairs Office of the University of Zurich. Conflict of interest: SJS received grants and non-financial support from Reactive Robotics GmbH (Munich, Germany), ASP GmbH (Attendorn, Germany), STIMIT AG (Biel, Switzerland), ESICM (Geneva, Switzerland), as well as grants, personal fees, and non-financial support from Fresenius Kabi Deutschland GmbH (Bad Homburg, Germany), grants from the Innovationsfond of the Federal Joint Committee (G-BA) and EU / European Commission, personal fees from Springer Verlag GmbH (Vienna, Austria) for educational purposes, and Advanz Pharma GmbH (Bielefeld, Germany), non-financial support from national and international societies (and their congress organisers) in the fields of anaesthesiology and intensive care medicine, outside the submitted work. Dr. Schaller holds small amounts of stock in Alphabet Inc., Ascendis Pharma Inc., Bayer AG, and Siemens AG; these holdings have not influenced any decisions regarding his research or this study. Presentation: none declared. This manuscript was handled by Dan Longrois Footnotes Supplemental digital content is available for this article. References 1. Lonsdale H, Jalali A, Gálvez JA, et al. Artificial intelligence in anesthesiology: hype, hope, and hurdles. Anesth Analg 2020; 130:1111–1113. [ DOI ] [ PubMed ] [ Google Scholar ] 2. Menezes J, Zahalka C. Anesthesiologist shortage in the United States: a call for action. J Med Surg Public Health 2024; 2:100048. [ Google Scholar ] 3. Association of American Medical Colleges. The complexities of physician supply and demand: projections from 2018 to 2033. https://www.aamc.org/media/45976/download [Accessed October 6, 2024]. [ Google Scholar ] 4. Haller G, Heim C, Meier K, et al. Physician anaesthesia providers in Switzerland today and tomorrow: results of the National Anaesthesia Workforce Study (NAWOS). Swiss Med Wkly 2021; 151 (33–34): [ DOI ] [ PubMed ] [ Google Scholar ] 5. Childers CP, Maggard-Gibbons M. Understanding costs of care in the operating room. JAMA Surg 2018; 153:e176233. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Hatib F, Jian Z, Buddi S, et al. Machine-learning algorithm to predict hypotension based on high-fidelity arterial pressure waveform analysis. Anesthesiology 2018; 129:663–674. [ DOI ] [ PubMed ] [ Google Scholar ] 7. McKendrick M, Yang S, McLeod GA. The use of artificial intelligence and robotics in regional anaesthesia. Anaesthesia 2021; 76 (Suppl 1):171–181. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Bickford RG. Automatic electroencephalographic control of general anesthesia. Electroencephalogr Clin Neurophysiol 1950; 2:93–96. [ DOI ] [ PubMed ] [ Google Scholar ] 9. Spataru A, Eiben P, Pluddemann A. Performance of closed-loop systems for intravenous drug administration: a systematic review and meta-analysis of randomised controlled trials. J Clin Monit Comput 2024; 38:5–18. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Joosten A, Rinehart J, Bardaji A, et al. Anesthetic management using multiple closed-loop systems and delayed neurocognitive recovery: a randomized controlled trial. Anesthesiology 2020; 132:253–266. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Strauss A, Corbin J. Grounded theory methodology: an overview. Handbook of qualitative research 1994; Sage Publications, Inc., 273–285. [ Google Scholar ] 12. Iqbal JD, Vinay R. Are we ready for Artificial Intelligence in Medicine? Swiss Med Wkly 2022; 152:w30179. [ DOI ] [ PubMed ] [ Google Scholar ] 13. Dhawan I, Tewari A, Sehgal S, et al. Medication errors in anesthesia: unacceptable or unavoidable? Braz J Anesthesiol 2017; 67:184–192. [ DOI ] [ PubMed ] [ Google Scholar ] 14. Bitterman DS, Aerts HJWL, Mak RH. Approaching autonomy in medical artificial intelligence. The Lancet Digital Health 2020; 2:e447–e449. [ DOI ] [ PubMed ] [ Google Scholar ] 15. Endsley MR. Toward a theory of situation awareness in dynamic systems. Hum Factors 1995; 37:32–64. [ Google Scholar ] 16. Jiang J, Karran AJ, Coursaris CK, et al. A Situation awareness perspective on human-ai interaction: tensions and opportunities. Int J Hum–Comput Interact 2023; 39:1789–1806. [ Google Scholar ] 17. Kelly FE, Frerk C, Bailey CR, et al. Human factors in anaesthesia: a narrative review. Anaesthesia 2023; 78:479–490. [ DOI ] [ PubMed ] [ Google Scholar ] 18. Cooper J, Newbower R, Long C, McPeek B. Preventable anesthesia mishaps: a study of human factors. Anesthesiology 1978; 49:399–406. [ DOI ] [ PubMed ] [ Google Scholar ] 19. Webster CS. Healthcare technology, the human–machine interface, and patient safety during intravenous anesthesia. In: Absalom AR, Mason KP, eds. Total intravenous anesthesia and target controlled infusions: a comprehensive global anthology . Springer International Publishing; 2017:667–83. [ Google Scholar ] 20. Coeckelenbergh S, Joosten A, Cannesson M, Rinehart J. Closing the loop: automation in anesthesiology is coming. J Clin Monit Comput 2024; 38:1–4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Schulz CM, Krautheim V, Hackemann A, et al. Situation awareness errors in anesthesia and critical care in 200 cases of a critical incident reporting system. BMC Anesthesiol 2016; 16:4. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Lopes S, Rocha G, Guimarães-Pereira L. Artificial intelligence and its clinical application in anesthesiology: a systematic review. J Clin Monit Comput 2024; 38:247–259. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Coeckelenbergh S, Boelefahr S, Alexander B, et al. Closed-loop anesthesia: foundations and applications in contemporary perioperative medicine. J Clin Monit Comput 2024; 38:487–504. [ DOI ] [ PubMed ] [ Google Scholar ] 24. Bastola P, Atreya A, Bhandari PS, et al. The evolution of anesthesiology education: embracing new technologies and teaching approaches. Health Sci Rep 2024; 7:e1765. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Iqbal JD, Biller-Andorno N. The regulatory gap in digital health and alternative pathways to bridge it. Health Policy Technol 2022; 11:100663. [ Google Scholar ] 26. Gerke S, Minssen T, Cohen G. Ethical and legal challenges of artificial intelligence-driven healthcare. Artificial intelligence in healthcare 2020; Elsevier, 295–336. [ Google Scholar ] 27. Yu K-H, Beam AL, Kohane IS. Artificial intelligence in healthcare. Nat Biomed Eng 2018; 2:719–731. [ DOI ] [ PubMed ] [ Google Scholar ] 28. Frankel TC. We are convinced the machine can do better than human anesthesiologists. Published May 15, 2015. https://www.washingtonpost.com/news/the-switch/wp/2015/05/15/one-anesthesiology-robot-dips-its-toes-into-whats-possible-this-one-jumps-all-in/ [Accessed October 6, 2024]. [ Google Scholar ] 29. Frankel TC. It's game over for the robot intended to replace anesthesiologists. The Washington Post. Published March 28, 2016. https://www.washingtonpost.com/news/the-switch/wp/2016/03/28/its-game-over-for-the-robot-intended-to-replace-anesthesiologists/ [Accessed October 6, 2024]. [ Google Scholar ] 30. Gapinski K. The Sinking of Sedasys. Published April 6, 2016. https://www.aorn.org/outpatient-surgery/article/2016-April-the-sinking-of-sedasys [Accessed 6 October 2024]. [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplemental Digital Content ejaic-5-e0107-s001.docx (50.4KB, docx) Articles from European Journal of Anaesthesiology and Intensive Care are provided here courtesy of Wolters Kluwer Health ACTIONS View on publisher site PDF (1.2 MB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Understanding implementation of the 'high volume, low complexity' surgical hubs programme in England: a qualitative study.#388582
• The Effect of Perioperative Medications on the Outcomes of Patients Undergoing Cardiac Surgery#894674
• Bioethical Considerations of Deploying Artificial Intelligence in Clinical Orthopedic Settings: A Narrative Review.#242306
• Artificial intelligence in medicine: a clinician-oriented guide and operational framework to evaluate new AI devices#971212
• Point-of-Care Ultrasound (PoCUS) Teleeducation Curriculum for Hospital-at-home Care#838711