Attitudes of Anesthesiologists toward Artificial Intelligence in Anesthesia: A Multicenter, Mixed Qualitative–Quantitative Study

Artificial intelligence (AI) is predicted to play an increasingly important role in perioperative medicine in the very near future. However, little is known about what anesthesiologists know and think about AI in this context. This is important because the successful introduction of new technologies depends on the understanding and cooperation of end users. We sought to investigate how much anesthesiologists know about AI and what they think about the introduction of AI-based technologies into the clinical setting. In order to better understand what anesthesiologists think of AI, we recruited 21 anesthesiologists from 2 university hospitals for face-to-face structured interviews. The interview transcripts were subdivided sentence-by-sentence into discrete statements, and statements were then grouped into key themes. Subsequently, a survey of closed questions based on these themes was sent to 70 anesthesiologists from 3 university hospitals for rating. In the interviews, the base level of knowledge of AI was good at 86 of 90 statements (96%), although awareness of the potential applications of AI in anesthesia was poor at only 7 of 42 statements (17%). Regarding the implementation of AI in anesthesia, statements were split roughly evenly between pros (46 of 105, 44%) and cons (59 of 105, 56%). Interviewees considered that AI could usefully be used in diverse tasks such as risk stratification, the prediction of vital sign changes, or as a treatment guide. The validity of these themes was probed in a follow-up survey of 70 anesthesiologists with a response rate of 70%, which confirmed an overall positive view of AI in this group. Anesthesiologists hold a range of opinions, both positive and negative, regarding the application of AI in their field of work. Survey-based studies do not always uncover the full breadth of nuance of opinion amongst clinicians. Engagement with specific concerns, both technical and ethical, will prove important as this technology moves from research to the clinic.


Introduction
The artificial intelligence (AI) revolution in medicine is well underway [1]. The list of potential applications of AI and its subfield machine learning (ML, here used as synonyms for technologies that aim to replicate human cognitive functions with computer algorithms) is ever-expanding, and many research programs are now bearing clinical fruit [2,3]. Following the first FDA approval of an AI algorithm in 2016, barely a year later, 64 AI-based technologies had been brought to market [1]. At the same time, although the literature is in its infancy, these technologies appear to be starting to perform as well as human physicians [4,5]. Thus, it seems likely that the roles AI-based technologies play in medicine will continue to grow.

Part Two: Online Survey
To construct the online survey, first, the most frequently recurring themes from the coded interviews in the first part of the study were used to create six representative statements. Then, each statement was reviewed for content and construct (content validity) by two members of the research group who were not involved in creating the statements but who have experience in survey creation and AI. Finally, two anesthesiologists from the University Hospital Zurich checked the six statements for comprehensibility (face validity).
The final six representative statements (see Appendix C) were answerable on a fivepoint Likert scale with the divisions "1, strongly disagree", "2, disagree", "3, neutral", "4, agree", and "5, strongly agree". In order to quantify the level of agreement or disagreement with these statements in as wide a pool of practicing anesthesiologists as possible, a link to these statements in questionnaire format (Google Forms, Google LLC, Mountain View, CA, USA) was sent via email to all participants of a concurrently running anesthesia simulation study (and which included the original 21 interview candidates). The questionnaire remained active for a period of three weeks from July to August 2022. A single reminder email was sent halfway through this period.

Statistical Analysis
Data from part one of the study are reported as the number and percentage of responses corresponding to each code. The consistency of coding according to the final coding scheme between study authors DH and TR was assessed by calculating percent agreement and interrater reliability with Cohen's kappa [29].
The results of the online survey are presented as numbers, medians, and interquartile ranges (IQR). The Wilcoxon signed-rank test was used as a test of statistical significance. We considered a deviation from neutral (i.e., a value of 3, "neutral") as of practical significance and a p-value of <0.05 as statistically significant.
We used Microsoft Word, Microsoft Excel (Microsoft Corporation, Redmond, WA, USA), and R version 4.2.0 (R Foundation for Statistical Computing, Vienna, Austria) to manage and analyze our data.

Results
In the first half of 2022, we recruited 21 anesthesiologists from 2 centers for the first part of the study, the interview. In the second part of the study, the online survey, 49 anesthesiologists from across the 3 study centers participated (further study and participant characteristics are listed in Table 1).

Part One: In-Depth Interviews
In total, across all questions, 21 codes were derived via inductive coding by study authors DH and TRR. Interrater reliability, as measured by Cohen's Kappa, was 0.908, and the percentage agreement was 91.4%. The 21 codes could be grouped into 3 main themes, namely, (1) a good pre-existing understanding of AI, (2) a balanced view of the pros and cons of AI as applied to anesthesia, and (3) a generally positive view of the use of AI to predict clinical events. An overview of responses corresponding to each question and code (as number and percentage, as well as example statements) organized by the themes above is provided in Tables 2-4. Figure 1 is a word cloud representing the most common words used by participants in their answers. Complete transcripts are available in Appendix D.
Statements derived from questions 1 and 2 demonstrated a good pre-existing understanding of AI. Participants' statements mainly referenced AI as an "information technology" (44 of 90, 49%) and the "capabilities" (34 of 90, 38%) of AI. Of note, however, question 2 also revealed a lack of awareness of the applications of AI in anesthesia, with the majority of statements coded as "none" (23 of 42, 55%), followed by "research" (7 of 42, 17%) and "non-AI/-ML example" (7 of 42, 17%). Statements in response to question 3 demonstrated a balanced view of the pros (46 of 105, 44%) and cons (59 of 105, 56%) of AI as applied to anesthesia. Question 4 revealed a generally positive (32 of 67, 48%) or neutral (19 of 67, 28%) view of the use of AI to predict clinical events, with only a minority being negative in sentiment (6 of 67, 9%). Responses to question 5 were very varied, with statements referencing vital sign predictions (36 of 92, 39%), event type (19 of 92, 21%), treatment guide (17 of 92, 18%), and risk stratification (14 of 92, 15%) as potential targets for AI.   "Acute events in the operating theatre" (participant 7) "Things that are quite subtle, and that could easily be overlooked" (participant 20) Statements derived from questions 1 and 2 demonstrated a good pre-existing understanding of AI. Participants' statements mainly referenced AI as an "information technology" (44 of 90, 49%) and the "capabilities" (34 of 90, 38%) of AI. Of note, however, question 2 also revealed a lack of awareness of the applications of AI in anesthesia, with the majority of statements coded as "none" (23 of 42, 55%), followed by "research" (7 of 42, 17%) and "non-AI/-ML example" (7 of 42, 17%). Statements in response to question 3 demonstrated a balanced view of the pros (46 of 105, 44%) and cons (59 of 105, 56%) of AI as applied to anesthesia. Question 4 revealed a generally positive (32 of 67, 48%) or neutral (19 of 67, 28%) view of the use of AI to predict clinical events, with only a minority being negative in sentiment (6 of 67, 9%). Responses to question 5 were very varied, with statements referencing vital sign predictions (36 of 92, 39%), event type (19 of 92, 21%), treatment guide (17 of 92, 18%), and risk stratification (14 of 92, 15%) as potential targets for AI.

Part Two: Online Survey
Overall participants were in agreement with the survey statements. A minimum of 37 of 49 participants (75%) agreed or strongly agreed with all statements, except for statement two, "I don't currently use any technology based on AI or ML at work", where there was a bimodal distribution of answers (14 of 49 participants (29%) agreed, and the same number disagreed with the statement). A more detailed breakdown of results is presented as donut diagrams in Figure 2. Overall participants were in agreement with the survey statements. A minimum of 37 of 49 participants (75%) agreed or strongly agreed with all statements, except for statement two, "I don't currently use any technology based on AI or ML at work", where there was a bimodal distribution of answers (14 of 49 participants (29%) agreed, and the same number disagreed with the statement). A more detailed breakdown of results is presented as donut diagrams in Figure 2.

Principal Findings
The primary aim of this study was to explore what anesthesiologists already know

Principal Findings
The primary aim of this study was to explore what anesthesiologists already know and think about AI. We were able to derive three main themes from physician anesthesiologists' responses to a series of in-depth interviews and a follow-up questionnaire, including (1) a good pre-existing understanding of AI, (2) a balanced view of the pros and cons of AI as applied to anesthesia, and (3) a generally positive view of the use of AI to predict clinical events. Our dataset demonstrates a good level of pre-existing knowledge of AI in our sample of practicing anesthesiologists. Notably, all participants were able to give a definition of AI, with, at the interviews, only 4 of 90 statements (4%) coded as "little/no prior knowledge". Furthermore, in the follow-up questionnaire, 38 of 49 survey participants (78%) "agreed" or "strongly agreed" that they had a general idea of what AI/ML is. In terms of content, many interviewees effectively paraphrased Arthur Samuel's original definition of machine learning as "programming computers to learn from experience" [30]. Participant 3, for example, defined AI as "Computers [which] adjust and perfect their predictions based on experience". Some participants were very well informed indeed. Participant 6, for example, offered a definition of a neural network: "Machine learning today [comprises] a neural network with different nodes, which have an input and an output, and the output can then be passed on to several nodes in the next level, and this output is weighted, i.e., amplified or degraded, before being passed on".
In contrast, few interviewees were aware of the applications of AI in the field of anesthesia despite a fast-growing body of research literature. Other interviewees gave examples from unrelated specialties, especially radiology, or gave examples of a technology not based on AI principles. This knowledge gap was seen in part two of the study too, where there was a bimodal response (i.e., participants "agreed" and "disagreed" in equal measure) to the question "I don't currently use any technology based on artificial intelligence or machine learning at work".
What explains the gap between the ability to describe AI in the abstract and the inability to name a single application of AI in anesthesia, even if only from research? One explanation lies in the fact that, for the practicing clinician, AI is still largely an abstract technology and not yet in widespread clinical use (the Acumen Hypotension Prediction Index from Edwards Lifesciences remains the only FDA-approved example of an AI-based device [8]). Moreover, the average practicing physician consults the primary literature only relatively rarely [31]. Thus, there exists, as yet, no real clinical need to be familiar with AI-based technologies in the relatively little amount of time available to read research literature. Even in radiology, where AI has had the most impact, in one survey, a third of resident physicians had not read a single paper featuring AI in the preceding year [32]. Given this, it is perhaps unsurprising that concrete examples of AI implementations are relatively rare in our dataset.
Participants were generally able to take a balanced view of the application of AI to anesthesia. In this context, positive statements tended to reference more supposed technical capabilities of AI, for example, that an AI might be less prone to error, less biased, have more "experience" to draw on, or learn faster than a human operator might. Participant 8, for example, stated, "Machines don't get tired, they don't have bad days, they usually function better, they have a better memory than any human being and they have an unlimited capacity for learning". In this regard, participants in this study echoed many of the putative benefits of AI, as described in the literature [2].
Negative statements also referenced technical issues, including biased training data or faulty or too rigid algorithms. This again mirrors much of the published literature on AI in medicine and also serves to underline the good pre-existing level of knowledge of AI in our cohort [33,34]. For example, "[It] is only as good as the material with which it has been trained" (participant 6); "In medicine . . . there is often a gender bias, especially in drug studies, and that is a challenge to overcome" (participant 18); or "The situation can be diverse or much more differentiated than an algorithm can reckon with" (participant 2).
However, in contrast to the positive statements, the majority of negative statements focused on human-computer interactions instead of technical factors. Here, participants focused on different challenges, for example, the potential for anesthesiologists to deskill, or become obsolete. For example, participant 11 stated, "The forward thinking you need as an anesthesiologist can be lost". Similarly, and in common with previous survey data, in which concerns regarding obsolescence have also been reported, participant 20 stated, "We might make ourselves totally superfluous at some point" [9].
Another concern was the conflict that might arise were an AI to recommend a course of action that the human operator does not agree with. As other commentators have pointed out, an algorithm does not learn to diagnose; it learns to predict the chain of human events leading to a diagnosis [22]. AIs are thus more rightfully seen as "thinking partners" than replacements. Many participants in this study seemed to intuitively understand this: that a deep understanding of AI is required in order to be able to safely incorporate it into the operating theatre. Per participant 17, "If you're not familiar with the process, how the data is created, then you can't know what kind of errors can arise". Likewise, participant 19 was concerned that "if you decide to go against that recommendation, then there's kind of an ethical and moral dilemma, right? What if the patient then dies? Then you must ask, 'What could I have done better?' Should one always do the therapy recommended by artificial intelligence? So that's kind of very difficult". These statements reflect many of the discussion points found in the literature regarding the ethics and practical implementation of AI, for example, concerns regarding "explainability" and responsibility [35].
Nevertheless, participants in this study were generally, although not exclusively, positive, in both parts, about the use of AI in a predictive capacity. This is notable, given that AI is increasingly being applied in this fashion, whether to predict a difficult airway [36], intraoperative vital sign changes [8], or postoperative analgesia requirements [37]. However, an interesting subset of respondents (10 of 67, 15%) was skeptical about the role AI will play in the future in anesthesia. Participant 1, for example, when asked about using AI as a predictive technology, replied, "it is so extremely different, from patient to patient, that I can't imagine that an AI can manage that". Again, this mirrors previous work with clinicians on AI in medicine, who found that general practitioners largely considered the potential of AI in their field to be limited [38], as well as related literature on technology adoption, which frequently identifies a core of "active resisters" to new technology [16].

Comparison to Prior Work
In the existing literature, which consists almost exclusively of survey-based data, most studies have found that a plurality of physicians in a variety of specialties see the integration of AI into medicine as a positive development and that, the more technologically adept physicians are, the more positive their attitudes [9][10][11][12]39,40]. One of the only studies reporting a negative association, in which general practitioners in the UK were largely skeptical of the ability of AI to contribute meaningfully to their work, is also one of the only studies to utilize qualitative data [38]. This study extends these findings to the field of anesthesia, finding an overall strikingly positive response to closed questions but a rich collection of nuanced observations, both positive and negative, in interviews. On the basis of these data, it could be argued that these prior surveys have not captured the full range of clinicians' thoughts and opinions regarding AI. Here, it is insightful to look at efforts to develop more comprehensive surveys, such as the new General Attitudes towards Artificial Intelligence Scale (GAAIS) from Schepman et al., which appears to better explore the full range of participants' opinions [41].

Limitations
The data for this study were gathered using an inductive approach to the thematic analysis of interview data. These interviews were open-ended, standardized, and con-ducted by three different interviewers. This hypothesis-free-but-hypothesis-generating tactic, applied to a large dataset, lends credibility to these findings [42]. Furthermore, combining interview and survey data, on the one hand, yielded insights that either method alone would not have and, on the other hand, led to a perhaps somewhat leading survey, based as it was specifically on themes arising during the interviews.
However, it should be noted that participating anesthesiologists in both parts of the study trended younger and more female than the workforce at large. This was not intentional-and, indeed, the gender mix in the institutions where this study was carried out is roughly equal-but rather resulted from the relative availability of younger anesthesiologists in the course of the working day. In addition, age is a well-established moderator of attitudes toward new technology, with younger people generally more positively predisposed to new technologies. Finally, participants were recruited exclusively from university-affiliated hospital settings in which clinicians are arguably more used to trialing new technologies. All of this could perhaps explain the overall enthusiasm for AI in our cohort. Moreover, comparisons to similar cohorts are not possible, as this is the first such study in the field of anesthesia. Given this, further studies, especially as AI technology begins to be more widely adopted clinically, are warranted.
In addition, it should also be noted that the online survey, brief as it was, was not able to cover all the nuances in the data from the qualitative part of the study. Thus, the generalizability of many interesting points raised by our study participants remains unclear.

Conclusions
In this study of what anesthesiologists already know and think about AI, we have established that anesthesiologists appear to be generally well informed about AI and take a balanced view of the integration of AI in anesthesia. They are aware of some of the pitfalls of AI while being cautiously optimistic about, especially, the technical benefits it could bring to patients, above all in terms of its predictive capacity. Our results further suggest that there remains a small group of skeptics who will require a high degree of evidence to be "won over" to AI-based technologies. If AI is to successfully make the jump from research into the clinic, developers will need to build on clinicians' pre-existing knowledge base and help practicing anesthesiologists navigate the strengths and weaknesses of this new technology to achieve successful implementation.

Institutional Review Board Statement:
The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the Canton of Zurich, Zurich, Switzerland, which reviewed the study protocol and issued a declaration of no objection on 11 March 2022 (Business Administration System for Ethics Committees Req-2022-00302). Additionally, all participants signed an informed consent form in which they agreed to the use of their anonymized demographic data and interview answers for medical research.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: All data are available in the appendices to this paper.

Appendix A. Interview Questions
English Translation

1.
What do you understand by the terms "artificial intelligence" or "machine learning"?
After question 1, study participants were provided with the following definitions of artificial intelligence and machine learning: (English) "Artificial intelligence is a branch of computer science that deals with the automation of intelligent behaviour. Machine learning is a sub-area of artificial intelligence." (German) «Künstliche Intelligenz ist ein Teilgebiet der Informatik, das sich mit der Automatisierung von intelligentem Verhalten befasst. Maschinelles Lernen ist ein Teilbereich der künstlichen Intelligenz.»

2.
Are you aware of any applications of artificial intelligence or machine learning in anesthesia?
After question 2, study participants were shown a collection of recent examples of applications of artificial intelligence to anesthesia (see Figure A1). biet der Informatik, das sich mit der Automatisierung von intelligentem Verhalten befasst. Maschinelles Lernen ist ein Teilbereich der künstlichen Intelligenz.» 2. Are you aware of any applications of artificial intelligence or machine learning in anesthesia?
After question 2, study participants were shown a collection of recent examples of applications of artificial intelligence to anesthesia (see Figure A1). 3.
What do you think are the advantages and disadvantages of artificial intelligence in anesthesia? 4.
In particular, what do you think about the use of artificial intelligence to make predictions? 5.
Which predictions do you think are most useful clinically?
German Translation

2.
Are you aware of applications of artificial intelligence in anaesthesia?
English: "Not in the clinic. [In research it is for various predictions]. There are also [models that you either select clinically through algorithms, that is clinically through regression, or through some kind of tree... map, I think it's called, where you let it select through that], but I know it so far, it hasn't convinced me yet, because it has always performed worse than if you have selected things clinically beforehand and you have tested them."

3.
What do you think are the advantages and disadvantages of artificial intelligence in anaesthesia?
English: "I can imagine that in some cases, [because you have so many parameters, it's an advantage], because [there are many parameters where a person might need a few learning cases before he can predict as well as the AI], that it's better. But  1. What do you understand by the terms "artificial intelligence" or "machine learning"?
English: "Machine learning today? You have [a neural network with different nodes, which have input and output and the output can then be passed on to several nodes at the next level. And depending on how this output is weighted, i.e., how the input is amplified or degraded and passed on], a neural network can then [make decisions] afterwards, depending on how it was set up."

2.
Are you aware of applications of artificial intelligence in anaesthesia?
English: "There's bound to be something done in anaesthesia [I don't know yet]."

3.
What do you think are the advantages and disadvantages of artificial intelligence in anaesthesia?
English: "Well, in general, artificial intelligence [is only as good as the material with which it has been trained]. And here [you probably need a lot of input to eventually get a reasonable output]. And then it's also the case that [if you hypothetically let an artificial intelligence make predictions about anaesthesia, we humans also play a role as disruptive factors]. Because [the anaesthetist can now spontaneously decide okay, I'll give fentanyl now and the AI only knows that when it is given as an input]. I can imagine that [when you give a drug at that moment, reality takes a completely different direction than the prediction of artificial intelligence]."

4.
In particular, what do you think about the use of artificial intelligence to make predictions?
English: "As we have already said, [we need an input and the output is only as good as the input]. [

2.
Are you aware of applications of artificial intelligence in anaesthesia?
English: "So I don't know at all whether this is actually artificial intelligence, that is, [decision-making, support systems and decision-making aids are widespread in medicine in anaesthesia]... what is really done with artificial intelligence... [I'm not personally aware of anything], but it could also be my personal ignorance. Well, I only know that [at least in radiology it has been done more often], that the artificial intelligence of algorithms... we want that, but... we are in the process, but I..."

3.
What do you think are the advantages and disadvantages of artificial intelligence in anaesthesia?
English: "Well, the advantage is that [a machine learns things more often], if you can put it that way, and through these experiences and [can therefore make more informed decisions] and perhaps also with less stress. But of course [it's also a computer-based learning, which may be error-prone or may have errors in the learning process] or [may be subject to some biases]. [Of course, one can always say that the human being may still be able to assess more all-encompassing capabilities with experience in a different way.]. And I think [we are also often totally biased as a result of our experiences]. For example, in the case that you have a patient who is for some reason tachycardic and hypotensive: one person sees a haemorrhage, and the next sepsis, and the next, a surgeon, doesn't want to admit there's anything wrong at all. So [everyone is somehow subject to their' own personal relationship and their previous experiences]. And not the now... And [that bias comes into play with us, but can be filtered out by technology or by the computer]."

4.
In particular, what do you think about the use of artificial intelligence to make predictions?
English: "Yes, a lot. So that [in itself is something totally innovative] and I think [still too little used in medicine], because we are all relative beginners in the profession and we also learn a great deal from experience and from situations. That was the case in such and such a situation, especially in anaesthesia... that's the only thing that happens. You have a direct consequence from an action and [from algorithms that show you something, they are much cleverer than humans and faster and probably more reliable if they have already gone through it millions of times in their computer]. [We need years for that, maybe to achieve what the machine achieves]. But [I think it's very difficult to train it well with data]. I could imagine. So with [patient data or surgery data, it's already difficult to integrate that into a machine learning programme and have it learn well]. So I can imagine... because [the interface between technology and informatics and medicine is already a big hurdle]. So similar in our project and data protection, the law. [In itself, it's a super, super good idea] and [I think it can take us a long way and maybe also in medicine], but [it still needs a a while until it's ready]."

5.
Which predictions do you think are most useful clinically?
English: "Well, starting with [acute predictions]. So [if I give this drug and that drug, what happens to heart rate, blood pressure etc.] but also [outcome relevant points. Is the patient at risk of PONV and postoperative myocardial infarction, stroke, etc.?] There are many, many risk scores and predictions, but they are not in our everyday life and maybe you do it better than we do, but in anaesthesia you still sit in your pre-med [room] and look at a patient and think hmm he is sick, he is healthy. But [if you could somehow calculate that there is such and such a high risk of suffering a heart attack or such a high risk and that you have to pay special attention to this on the one hand and also have ways of avoiding this if you know that there is perhaps a higher probability that this will occur], that I think that it is [the acute events in the operating theatre], whereby [it is of course already clear that if I inject someone with 200 propofol that he will then become hypotensive]. Maybe I don't need a prediction, but [maybe there are major complications, and I think that's very relevant]. Yes." Participant 8

1.
What do you understand by the terms "artificial intelligence" or "machine learning"?

2.
Are you aware of applications of artificial intelligence in anaesthesia?
English: "[Not yet, to be honest], but [it will exist somewhere, I think, running in the background] But [I don't know now]."

3.
What do you think are the advantages and disadvantages of artificial intelligence in anaesthesia?
English: "I think one disadvantage is [coming to rely too much on machines]. [There are always potential sources of error, whether from the user or something being transmitted incorrectly, the machine can also be wrong]. And I believe that [if you rely too much on machines, at some point, there is a danger that you no longer look at the patient yourself]. For example, the device may say the patient is asleep, when that may not be the case. But I see more advantages than disadvantages, because [AI and ML] work with algorithms.
[Machines don't get tired, they don't have bad days, they usually function better, they have a better memory than any human being and they have an unlimited capacity for learning] and can support us in many decisions. [Maybe not in taking decisions, but in many situations, computers can support us], because algorithms are simply logical paths that you work through one after the other, which is what you do in everyday life, but of course [as a human being you are much more prone to error than a machine]. [So, ultimately, I think these technologies are more useful than harmful]."

4.
In particular, what do you think about the use of artificial intelligence to make predictions?

3.
What do you think are the advantages and disadvantages of artificial intelligence in anaesthesia?
English: "So the advantages are definitely that [there is finally a system that can evaluate the vast amounts of data that we are currently recording digitally], via anaesthesia protocols, where it is documented which drug is given when, which patient it is, how the vital parameters are, and finally with a large data set and then [make predictions]. That is of course super good. I would almost say that anaesthesia is almost only advantages in other areas. After all, [we have interpersonal contact. It can't be calculated away, of course, and of course an artificial intelligence can't do that]. But apart from that, I think there has to be much, much more. Yes, in my head there are only advantages. I always say that for learning, too. [Certain empirical knowledge comes from experience, of course, that will always be the case. But AI can also help you to understand a certain experience, to see certain things developing. So that, even when you don't have the system, you might think the AI has always warned me in this case, or made me aware of certain side effects. And so on. That that can have a certain teaching effect]. That's why I don't think there are any disadvantages in this area."

4.
In particular, what do you think about the use of artificial intelligence to make predictions?
English: "I think [it can definitely give a hint]. [Whether the hint is actually true or the prediction gives a clear indication to act is questionable]. So if the doctor says, okay, after giving a certain medication the heart rate or the blood pressure will drop with medium-urgency and I should then give this medication as a preventive measure. [ 1. What do you understand by the terms "artificial intelligence" or "machine learning"?
English: "By artificial intelligence and machine learning, I understand that [a computer scans pathologies according to predefined parameters and highlights a range of diagnoses and differential diagnoses accordingly]."

2.
Are you aware of applications of artificial intelligence in anaesthesia?
English: "[No, honestly not]. Not consciously, anyway. No, not with us. But I haven't been in the operating theatre for a long time. It could just be that it's because of that."

3.
What do you think are the advantages and disadvantages of artificial intelligence in anaesthesia?
English: "[Advantages are safety, that sometimes you don't notice something because you're busy with other things on the patient, repositioning or something else, that something is just shown]. But one disadvantage is perhaps that [you then switch off your head and simply rely on the PC or the programme] and [perhaps at some point you can no longer really understand the pathophysiology]. Or [the forward thinking that you need as an anaesthetist can be lost]."

4.
In particular, what do you think about the use of artificial intelligence to make predictions?
English: "[I think that's good, because you could then prepare oneself accordingly for what needs to be done, and not be caught unawares]."

5.
Which predictions do you think are most useful clinically?
English: "Anaesthesia is such an [acute emergency subject and as I said, if you just somehow knew that a step beforehand, then you can intervene much earlier], really prepare for it and that is certainly better for patient care, which simply makes patients safer." Participant 12 1. What do you understand by the terms "artificial intelligence" or "machine learning"? . I know, for example, that there was an artificial intelligence project or a company that wanted to establish such a device for nociception and that's also something that you learn to evaluate yourself, whether a patient is in pain during the operation or at least under stress, and when that is more or less taken away from you, [you then rely on values and no longer interpret other vital parameters yourself], with which you perhaps could arrive at the same result much more quickly."

4.
In particular, what do you think about the use of artificial intelligence to make predictions?
English: "[I think it is an excellent additional tool, especially for beginners and young colleagues to gain a bit more confidence through it]. [You just have to be careful not to rely purely on it], [I think it could be dangerous when used alone, because it can't read in any other way]."

5.
Which predictions do you think are most useful clinically?
English: "[Everything that can potentially endanger the patient], in other words, everything that can cause harm to the patient or make harm more likely, is of course a good thing to have." Participant 13 1. What do you understand by the terms "artificial intelligence" or "machine learning"?
English: "Machine learning is learning [on the basis of big data] and, in contrast to medicine, where [you always have a causality somewhere, in machine learning, you simply have correlations] from the [volume of data], which you may not be able to store at all! And artificial intelligence is either simply machine learning with [very, very, very, very large amounts of data] or a little less technical but the application of data and [the evaluation of data] [with the help of algorithms] or machine learning."

2.
Are you aware of applications of artificial intelligence in anaesthesia?
English: "[I've heard of it in radiology]. In anaesthesia... [we now have Narcotrend, i.e., cerebral state index, a depth-of-anaesthesia measurement. I assume that this is relevant]. Otherwise I don't think I can think of... [I think there is also anaesthesia prediction]. I would somehow summarise that under the topic, but that's all I can think of now."

3.
What do you think are the advantages and disadvantages of artificial intelligence in anaesthesia?
English: "[It is likely to capture more nuance] simply through artificial intelligence and machine learning. One disadvantage is [perhaps that in some cases the causality is not known, and the anaesthetist may not know how to react to it because he only knows how a vital sign might change, but not why it is changing. For example, hypertension is predicted, but it is not clear why it will happen]. Maybe."

4.
In particular, what do you think about the use of artificial intelligence to make predictions?
English: "Um, [I think it's good as an additional tool], [so long as it remains an aid and not a substitute, a way of flagging a problem, but then it's still acted upon by someone with the academic background knowledge, so to speak, and it's not simply flatly accepted].
[So long as there's still someone behind it who interprets it or what it means and reacts properly to it]."

5.
Which predictions do you think are most useful clinically?
English: "The need for [post-operative analgesia], I think, would be quite interesting. [Anaesthetic depth measurement] is a useful tool, but if that could be improved, that would also be good. And yes, of course, the prediction of intraoperative incidents such as [hypotension] or [hypoxaemia], would also certainly be good, to be warned in advance, so that you can remain vigilant and perhaps know that during this anaesthetic I have to pay a little more attention, but [I think what I have not yet seen was the post-operative analgesic requirement, which would certainly be exciting]." Participant 14 1. What do you understand by the terms "artificial intelligence" or "machine learning"?
English: "So under artificial intelligence? I basically understand [a computer or some other kind of electronic device] that is [able to evaluate data] and [make a forecast from it]. In the end, that's exactly what it is, i.e., [the creation of a forecast for the future] [based on a data set] that [the computer] has, which [the human brain could also do, but specifically from the computer]. Yes."

2.
Are you aware of applications of artificial intelligence in anaesthesia?
English: "In anaesthesia now less than [in radiology, for example]. So now in the clinic it's not clear. Our [Perseus can make a respiratory gas prediction, that is perhaps also a lower, lower artificial intelligence], but otherwise [I can't think of anything]."

3.
What do you think are the advantages and disadvantages of artificial intelligence in anaesthesia?
English: "The advantage is that it is [less distorted by emotions or personal prejudices], especially with regard to the duration of the operation, for example, that it is [really only based on rational information]. The disadvantage is that [perhaps some human factors are not taken into account by the computer], that an operation at 3 a.m. always takes longer than during the day."

4.
In particular, what do you think about the use of artificial intelligence to make predictions?
English: "[I actually think it's good]. Also [because you sometimes find yourself in the situation when an operation has been going on for three hours, the patient is completely stable, you have set the alarms and pay relatively little attention to anaesthesia, and it's nice to have a prediction every now and then, like "hey, I have to do something again now"].
Because the patient is about to wake up or something."

5.
Which predictions do you think are most useful clinically?
English: "[Patient is waking up], which usually doesn't work for me. We have a Narcotrend. If it tells me that the patient is awake, then the patient will be awake in ten minutes. Yes, also [things like ST segment], things like that, I don't necessarily find at first glance in our one-line ECG. [Blood pressure] too. Those are the most important ones now." Participant 15