Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy 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 Sci Rep . 2026 Mar 26;16:10505. doi: 10.1038/s41598-026-45245-2 Search in PMC Search in PubMed View in NLM Catalog Add to search Investigating expectations and needs regarding the use of large language models at Bavarian university clinics Juraj Vladika Juraj Vladika 1 Department of Computer Science, TUM School of Computation, Information, and Technology, Technical University of Munich, Garching, Germany Find articles by Juraj Vladika 1, ✉ , Alexander Fichtl Alexander Fichtl 1 Department of Computer Science, TUM School of Computation, Information, and Technology, Technical University of Munich, Garching, Germany Find articles by Alexander Fichtl 1 , Florian Matthes Florian Matthes 1 Department of Computer Science, TUM School of Computation, Information, and Technology, Technical University of Munich, Garching, Germany Find articles by Florian Matthes 1 Author information Article notes Copyright and License information 1 Department of Computer Science, TUM School of Computation, Information, and Technology, Technical University of Munich, Garching, Germany ✉ Corresponding author. Received 2025 Apr 16; Accepted 2026 Mar 17; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . PMC Copyright notice PMCID: PMC13031305 PMID: 41882171 Abstract Recent advancements in Artificial Intelligence (AI) have been driven by Large Language Models (LLMs), powerful tools capable of generating coherent text and solving diverse analytical tasks. While LLMs hold great potential to enhance healthcare by assisting physicians and improving patient treatment, their clinical adoption is limited, and there is a lack of statistically grounded information on the opinions of medical professionals, personnel, and students regarding LLM usage. To address this gap, we conducted an online survey from April to October 2024, gathering insights from 120 participants across five Bavarian university clinics (in Germany), including physicians, medical students, and administrative staff. Findings show that many participants already use LLMs for research support, summarization, translation, and report drafting. Most believe LLMs will positively influence their field, acknowledge their potential to automate mundane tasks, and believe they will help to achieve a more personalized, evidence-based, and cost-effective patient treatment. However, concerns were shown regarding their opaque nature, data privacy, and the potential loss of patient trust. Participants overwhelmingly feel their institutions are not well-prepared for LLM adoption, with suggestions for improvement including increased education and specialized training, investments in digitalization and infrastructure, ensuring legal compliance, and encouraging technological openness. We hope these insights will inform the design of future medical AI solutions. Keywords: Large language models, Artificial intelligence, Natural language processing, Medical AI, Digital transformation Subject terms: Computer science, Information technology, Medical research, Health care, Health policy Introduction The field of Artificial Intelligence (AI) focuses on building machines that can solve tasks requiring advanced analytical skills and human-level intelligence 1 . At the intersection of AI and linguistics stands Natural Language Processing (NLP), a discipline dealing with developing tools that can analyze and understand human-written text or generate new text from scratch 2 . Impressive progress in NLP in the last few years has been achieved due to the advent of Large Language Models (LLMs), powerful tools that can generate text and solve a wide array of tasks using their next-word prediction capabilities based on the transformer architecture and pre-training on massive textual corpora 3 . Along with popular general-purpose LLMs like ChatGPT 4 , models specifically trained for biomedical purposes have also emerged 5 . LLMs have shown high capabilities in assisting with diverse and complex medical tasks and a promising potential in improving patient treatment and hospital systems 6 . This includes generation of discharge summaries 7 , empathetic conversations with patients 8 , encoding clinical knowledge for answering medical questions 9 , 10 , or assisting medical students with preparations for entry exams 11 . Conversely, other studies revealed that LLMs can still struggle with accurate patient diagnosis, adhering to clinical guidelines, and being applied in real hospital settings 12 . Nevertheless, as they have already started transforming numerous parts of society, the quick pace of their improvement makes a broad inclusion of LLMs into hospital systems highly likely in upcoming years 13 . Therefore, it is of crucial importance to understand the implications of LLMs in medicine. Studies investigating the potential of LLMs in other high-stake fields like law 14 , science 15 , and education 16 , have started emerging. In the medical field, however, previous studies focused mostly on AI models working with numerical or visual data instead of natural language 17 , 18 . Initial studies discussing LLMs in medical work have started appearing 19 , 20 , but there is still a lack of comprehensive surveys on the opinions of clinicians and medical students on the use of LLMs in their work. While the United States and China lead the current AI revolution, Germany also ranks among the top contributors to AI research output 21 . However, a sharp contrast exists between this high research productivity and the real-world practical implementation of AI solutions. Germany is marked by significantly slower adoption of these technologies in the public sector, largely due to stricter legal regulations and a culture that prioritizes established practices over potentially risky innovations 22 . The legal regulations governing medical AI include the EU’s General Data Protection Regulation (GDPR) 23 , focusing on individuals’ rights to data protection and privacy including patient data, and Medical Device Regulation 24 , which addresses safety, quality, and risk management requirements on all medical technology of the market. The level of digitalization in German hospital systems is still rather modest 25 , including the low adoption of AI in healthcare 26 , and before introducing tools based on LLMs into the German medical sector, it is essential to understand how clinicians currently perceive these technologies. By understanding what medical professionals and students prioritize or are concerned about, developers and researchers can tailor biomedical NLP solutions to better fit the actual needs of the medical community. To help bridge this research gap, we conducted a quantitative study using a survey instrument which also included qualitative elements. The survey consists of 22 questions and targets three groups: (a) physicians, (b) students of medicine, and (c) hospital administrative staff. The survey inquires about the opinions and attitudes of these three groups on the use of medical LLMs in their daily work. The questions include asking about potential use cases they would like to use the LLMs for, opinions on how it will affect the workforce, evaluating the acceptable level of errors, inquiry about what the perceived benefits are but also the risks and dangers of such models. A total of 120 participants completed the survey (70 students, 36 physicians, 14 administrative) from 5 German (Bavarian) university clinics between April and October 2024. We hope the results of the survey will help bridge the gap between the development of LLM-based solutions and their actual adoption in applied clinical settings by medical professionals. Results In this section, we present the results of the research survey. We discuss the survey questions within their respective thematic groups. We present the most informative insights, while complete survey results are included in the Supplementary materials . Participant background In total, 120 complete questionnaire responses were received across the three target demographic groups. Their distribution was as follows: Ongoing studies in the field of medicine: 70 participants Working as a physician: 36 participants Administrative staff: 14 participants The physicians and resident students were specialized in various fields, including psychiatry (13), neurology (6), general practitioners (4), internal medicine (3), otorhinolaryngology (ENT) (3), orthopedic surgery (3), radiology (2), nephrology (2), urology (2), translational medicine (2) and a few others. The full list can be found in the Supplementary materials . The sample is slightly biased towards psychiatry and neurology because these departments had the most active distribution of our survey. Out of 80 participants that responded about their years of experience in medicine, 20 (25%) have worked below 5 years, 18 (22.5%) have worked 5–10 years, 13 (16.25%) have worked 10–20 years, 7 (8.75%) have worked 20–30 years, and 2 (2.5%) have worked more than 30 years. Current knowledge and use of NLP While 34 participants (28.33%) never used LLMs for work-related tasks, almost half of the participants (n=56; 46.67%) reported a daily or weekly use (Fig. 1 ). Around one quarter of students (n=18; 25.7%) and physicians (n=10; 27.8%) indicated that they never use LLMs, while around one half of students (n=33; 47.1%) and exactly one half of physicians (n=18; 50%) said they use LLMs on a weekly or daily basis. Administrative staff showed the lowest frequency of use, with 6 (33.3%) reporting never using LLMs and only 5 (27.8%) using it weekly or daily. Still, the usage between groups is relatively similar (p=0.335, V=0.195). Fig. 1. Open in a new tab Left plot ( a ):“How often do you use LLMs in your daily work?”Right plot ( b ):“Compared to your colleagues, how would you rate your knowledge of LLMs (on a scale of 1 to 5)?”. The participants’ confidence in their knowledge of LLMs varied significantly between groups (p = 0.037). While the largest chunk of students (n=27, 38.57%) considered their knowledge average (score of 3), physicians were very confident in their knowledge. The majority of physicians rated their knowledge of LLMs as good (4) or very good (5) (n=19, 52.78%, compared to only n=21, 30.00% of students). The responses of the administrative staff were less conclusive. When asked how they would rate their knowledge of LLMs compared to their colleagues, 42 participants (35.00%) reported that they feel they have a worse understanding of this technology than their colleagues. The participants mentioned a wide range of use cases for which they already utilize LLMs. The most popular use case was using LLMs as a general support tool for doing research (data analysis, literature search, research ideas, etc.), mentioned 17 times. Other popular use cases (with number of mentions) were using LLMs for answering questions (14), translation (11), explaining unfamiliar concepts (10), and general help with studying (5); various types of text generation such as drafting of e-mails (7), patient discharge notes ( Arztbriefe ) (5), and parts of research work like master’s thesis and scientific articles (8). Other mentioned use cases include summarization (6), programming (4), spellchecking (4), and speech transcription (3). Other than inquiring about their existing use cases, we also prepared a pre-defined set of popular practical use cases of LLMs and asked participants about the relevance for their daily work, even if they do not use them yet. Some of the proposed use cases overlap with those already found in the free-form field: translation, report creation, summarization, and question answering. Still, some of the use cases included in this question were rarely mentioned in the free-form answers: information extraction, speech transcription, and text simplification. The average results of relevance (scale 1–5) of potential LLM use cases in their work are in Fig. 2 . Each potential use case was rated on average above 2.35. The use cases seen as the most relevant across all groups were translating medical reports (4.04, 3.21, 3.81) and leveraging LLMs for transcribing spoken language into text (3.84, 3.50, 3.94). These two use cases were then closely followed by the three generative tasks, namely report creation, text summarization, and text simplification. The administrative staff considered the summarizing of long documents slightly more important than translation, and in general rated medical-specific tasks considerably lower than physicians and students did. Using LLMs for key information extraction and for providing medical reasoning were also relatively popular among physicians and students (average above 3), but less relevant for the administrative stuff. Finally, the least attractive use case among all three participant groups was answering patient questions during their daily work (2.46, 2.36, 2.56). Fig. 2. Open in a new tab “How relevant are the following use cases of language models to your daily work? (not only for your current use but also for potential future use)?”(on a scale from 1 to 5). Perceived influence More than one-third of all respondents (n=26, 38.24%) already perceive a noticeable impact of LLMs on their field of practice today. The second largest group of respondents (n=23; 33.82%) estimates that an impact will be evident within 5-10 years. Only two participants (2.94%) expect that it will take more than 10 years for an impact to be felt in their field. Notably, no medical students were among them, even though it is by far the largest participant group. The three groups responded similarly (p=0.175, V = 0.22). When asked about the nature of the expected or present impact (Fig. 3 ), the vast majority of participants anticipated that the integration of LLMs would improve their current or future professional domain (n=68; 56.67% agreed or strongly agreed). Notably, while students and physicians gave similar responses, the administrative staff was less enthusiastic, with most responses (n=6; 42.86%) corresponding to a neutral expectation. Fig. 3. Open in a new tab “How much do you agree with the following statement: The introduction of LLMs will improve my field (1: Completely disagree; 5: Completely agree).”. Figure 4 focuses on the perception regarding changes in workforce needs. While (n=41; 34.17%) expect a strong or very strong change in workforce needs due to LMs within this decade, (n=47; 39.17%) disagree and only predict small or very small changes. These numbers shift slightly when looking beyond the next decade, where only (n=33; 27.5%) still disagree or strongly disagree. Overall, students expect the workforce needs to change less so than physicians and administrative staff within this decade (p=0.025, =0.448) and beyond the next decade (p=0.005, =0.733) Fig. 4. Open in a new tab “To what extent will language models have an impact on workforce needs in your area of expertise within the next decade ( a ) versus beyond the next decade ( b )? (on a scale from 1=low impact to 5=high impact)”. While the previous two questions focused on the strength of impact on workforce needs, the next question asked about the type of change (increase, decrease, remain the same). Hereby, the participants in our survey indicated that they anticipate a decrease in workforce needs (n=52; 43.33%) much more strongly than an increase (n=7; 5.83%). Surprisingly, with this phrasing, the largest share of participants (n=61; 50.83%) anticipates that the workforce needs will neither increase nor decrease. Practical implications We questioned the participants about the required performance standards for LLMs (Fig. 5 ) in two different scenarios with two different medical groups. In the left plot, the participants were asked about the performance standards necessary when untrained healthcare personnel (like medical technicians) use the LLMs for general screening purposes like disease detection. Here, a considerable proportion of all respondents (n=52; 43.33%) find the LLM screening performance equal to the level of or even below the level of an average specialist to be sufficient. Fig. 5. Open in a new tab Left plot ( a ):“If a language model were to be used for screening purposes by untrained healthcare personnel in your field of expertise, what level of error in the model’s assessments do you consider acceptable?”Right plot ( b ):“What level of error in the assessments of a language model would be acceptable if it were used by trained professionals/physicians as diagnostic decision support?”. On the other hand, in the right plot, participants were asked about the required performance of LLMs when used by trained professionals (like physicians) for diagnostic decision support. In this case, only (n=24; 20.0%) deemed an average or below-average performance of the LLM systems to be sufficient. Instead, the vast majority (n=96; 80.0%) responded that LLM systems should exceed the proficiency of at least an average specialist. Approval of Example LLM Use Scenario In another survey question, the following workflow procedure was suggested: “ During a pandemic, a specialist responds to patient questions online. To save time, they generate responses with a language model and then only review them before sending ”. The large majority of the participants (n=95; 79.17%) were open to adopting this method, with physicians being the group most in favor of the approach (n=33; 91.67%). No physicians answered“no”to this question. This question was tailored to the focus of the biomedical NLP research community on question-answering tasks. Opportunities and risks Two large-scale questions asked the participants to rank potential advantages and concerns of using LLMs in their field. Figures 6 and 7 ) show radar plots that visualize the responses. Fig. 6. Open in a new tab “Which of the following do you perceive as the greatest potential advantage of the use of language model systems in your field?”Participants indicated their top three preferences from a list of set choices. Higher scores in the radar plot indicate a higher ranking. Fig. 7. Open in a new tab “Which of the following do you perceive as concerns to the utilization of language models in your field?”Participants indicated their top three concerns from a list of set choices. Participants indicated their top three preferences from a list of set choices. Higher scores in the radar plot indicate a higher ranking. Perceived advantages and benefits As visualized in Fig. 6 , all three occupational groups agreed that the most important benefit of introducing LLMs in the medical field lies in the potential to reduce time spent on monotonous and repetitive tasks (n=78, 65% responses ranked as 1st). The next two benefits that physicians rated as the most influential are“more personalized and evidence-based disease management”and“more cost-efficient healthcare”. On the other hand, students found the second most important benefit to be“greater confidence in making diagnoses”, while the third place was a close call between“more cost-efficient healthcare”and“greater consistency in diagnosis and management decisions”. Finally, administrative staff rated very closely“improved access to disease diagnoses for patients”and“higher accuracy in predicting disease progressions”. Perceived risks and concerns When it comes to the perceived risks and concerns (Fig. 7 ), physicians found the most important ones to be: (1) lack of trust in “black-box”(nontransparent) diagnoses, (2) outsourcing healthcare to large tech companies, and (3) data security and privacy concerns. Students also found the (1) lack of trust in“black-box”diagnoses to be the most important, followed by (2) data security and privacy, and (3) challenges to the trust-based relationship between patients and doctors. Finally, administrative staff found the biggest concern to be the impact on staffing needs, which was rated in last place by physicians and students. Tied from second to fourth place for the administrative staff were diagnosis and treatment made by non-specialist systems, outsourcing healthcare to large tech companies, and data security and privacy concerns. Preparedness of medical centers and university clinics for introduction of LLMs Most participants of our survey (n=77; 64.17%) believe that their institution (university/clinic/medical center) is not adequately equipped to deal with the introduction of LLMs. A much lower number of respondents (n=14; 11.67%) think their institutions are adequately prepared. In a free-form followup question, we asked the participants on suggestions on what has to change for better preparedness of their institutions for the introduction of LLM technologies. Even though the question was optional, we received a large number of responses (n=78; 65%), showing the topic is important to participants and they would like to be involved in the process. We provide the most common themes and factors for adoption in Table 1 . Illustrative quotes for most common themes are given in Supplementary Table 3 . Table 1. Overview of the most common key factors that survey respondents mentioned for improving the preparedness of their institutions for AI adoption. ID Key factor Mentions F1 Education about AI 46 F2 Ensure privacy & security 17 F3 Infrastructure & resources 13 F4 Provide examples, onboarding 10 F5 Digitalization of clinical texts 8 F6 Legal compliance 8 F7 Integration with hospital systems 6 F8 High accuracy of predictions 5 F9 Cooperation with IT departments 4 F10 Reduce bureaucracy 3 F11 Work with local language 3 Open in a new tab As seen in the thematic content-analysis matrix in Table 1 , the largest number of participants (38% of all and 59% of those who answered, n=46) identified education about AI and LLMs as the most crucial factor for successful adoption. This included responses such as the need for more seminars and courses about the abilities and potential of LLMs (one participant mentioned that“integrating the use of language models into the medical curriculum makes it easier to apply them later on”). The second most mentioned factor was Ensuring Privacy and Security of AI systems. The participants pointed out how medical paperwork often deals with highly confidential patient records and sensitive information like medical history or disease diagnoses. Another point was ensuring the security of LLM systems against any potential malicious use or adversarial attacks. The third most popular factor was acquiring the necessary technical infrastructure and resources to host and support the deployed AI systems. The fourth most popular factor was the need for on-boarding and clear instructions of how and when to use the LLM-based tools in medical settings. Many respondents emphasized that they cannot be expected to start using the system out of the box without proper instructions and examples. Furthermore, another important factor was the digitalization of clinical text. Eight participants informed us that a significant portion of the text they use is still in handwritten format and has never been scanned or digitized (with one emphasizing“early organization of a data protection structure, digitization of clinical findings, digitization of dialysis data”). Legal compliance, with eight mentions, emphasizes the need to respect and clarify legal regulations before introducing LLM systems into clinical practice. Integration within existing hospital systems (six mentions) refers to the need of seamlessly integrating newly adopted AI systems into those systems already in use. The requirement for high accuracy of predictions was mentioned five times, including considerations regarding liability for any medical errors. Finally, four participants mentioned the need for good cooperation with existing IT departments of their respective institutions. Three responses focused on reducing the overall level of bureaucracy in the system as a prerequisite for wider AI adoption. Lastly, three respondents pointed out how, while LLMs work very well with the English language, they have to be properly tested and work well with the local language (German), considering that most clinical data will be provided in German. Discussion Given the increasing amount of funding, research, and discussion dedicated to LLMs in multifold areas, it is essential to understand the expectations and needs of those using them. LLMs have the potential to transform healthcare and our study is among the first ones in the German context to investigate the opinions and attitudes of medical professionals toward the adoption of LLMs in medicine, especially with inclusion of medical students and administrative staff. The introduction of LLMs is mostly seen as positively influencing their field. This follows the findings of related surveys on the use of AI in medicine 17 , 27 , 28 , but also newer ones on the use of LLMs by physicians and students 29 – 32 . Around one-half of both physicians and students in our study use LLMs daily or weekly. This finding highlights the growing recognition of the utility of LLMs in automating mundane tasks, thereby saving time and enabling higher-quality results. While they were the smallest participant group, administrative staff also had the lowest frequency of LLM use. The outreach and promotion of LLMs to help the medical community should also be more inclusive of administrative workers in the future. Finally, a considerable number of students said they have a worse understanding of the technology than their peers, suggesting that many feel left behind in the face of rapid digital developments. Existing LLM use cases among respondents are diverse and multifaceted. Many utilize LLMs for general-purpose tasks, mirroring how the general population uses them 33 . Examples for such tasks are drafting emails, summarizing lengthy texts, translating articles, or conducting quick web searches. At the same time, the participants identified speech transcription and translating medical records into patients’ native languages as the most important use cases. Less frequently mentioned and perceived as less important were more complex medical tasks, like suggesting diagnoses or support in medical reasoning. This highlights the discrepancy in the focus of the NLP/AI community, which often prioritizes researching complex tasks over addressing the actual needs of medical workers. On the other hand, it is possible that medical experts are not informed of or acquainted with the possibilities that LLMs can provide and expect tools with user-friendly interfaces and instructions to start using them for more complex tasks. Finally, it is also evident that administrative staff rated as the most important those tasks that are closer to their daily work environment, such as speech transcription, document summarization, text and simplification, while rating the medical tasks like medical reasoning and concept extraction considerably lower than medical professionals. The least popular LLM use case was using LLMs to answer patients’ questions. At first glance, this finding contradicts the popularity of the proposed workflow scenario of using LLMs to draft an answer to a patient question, where the vast majority agreed they would use LLMs for that purpose. This suggests that physicians are open to utilizing LLMs to assist in drafting well-structured responses to questions but highly prefer to verify and refine the content of the answers themselves. The participants, especially physicians, mostly see the speed and level of impact of LLMs to be high. Interestingly enough, students were more moderate and careful in their assessment, with only a small group experiencing the current impact of LLMs. Since their medical careers are just about to commence and are expected to span multiple decades, they likely feel that there is enough time to adapt to the upcoming changes. Overall, students expect the workforce needs to change much less than physicians and administrative staff in both scenarios. Another interesting observation was that most students and physicians did not expect a very high performance standard of LLMs for routine screening tasks (like disease detection from anamnesis) but overwhelmingly required a high standard when helping them with diagnostic decisions and treatment plans. This reflects the lower potential risk associated with routine tasks even if model predictions are not highly accurate. Intuitively, untrained professionals could be less likely to spot intricate medical errors in LLM outputs. Still, this was not seen as detrimental in the screening phase because any error can be corrected in later patient treatment phases. Therefore, LLM-based digital assistants could fill the role of trainees for routine tasks when helping physicians but be like a peer when used for the more critical task of decision support in patient treatment. Nevertheless, recent studies have shown that LLMs perform better in routine classification tasks while struggling in realistic patient diagnostics and adhering to clinical guidelines 12 , 34 , highlighting a mismatch between actual capabilities of current LLMs and expected error rate thresholds from survey participants. All participant groups overwhelmingly see the reduced time spent on repetitive tasks as the biggest potential of LLMs in their work. This follows our insight on how they currently use LLMs and how they rated potential LLM use cases involving time-consuming tasks. Moreover, both physicians and students appreciated the opportunity of LLMs to lead to more cost-effective healthcare. By automating certain mundane tasks of everyday medical work, more time and energy of physicians can be directed to actual patient treatment, thus lowering the average cost of healthcare services 35 . Similarly, rural healthcare providers could utilize LLMs for support instead of spending expenses on getting help from metropolitan institutions 36 . Furthermore, physicians found important the idea of a more personalized and evidence-based disease management with LLMs, following the increasing trend in medicine on providing personalized treatment for each patient 37 . One way in which LLMs can help physicians here is, given a specific patient case, sifting through relevant medical sources like lengthy clinical guidelines or systematic reviews and summarizing the key points concisely to assist them in diagnosis and treatment. The biggest concern that physicians and students have with LLMs is the low trust in“black-box”diagnoses, i.e., the lack of transparency about how LLMs work internally. This highlights the need to augment LLMs with methods that clarify their decision-making process, such as decision trees 38 , symbolic logic reasoning 39 , chain-of-thought mechanism 40 , and other techniques of explainability 41 . Recent advanced LLMs like OpenAI o3 or DeepSeek R1 42 explain their reasoning by“thinking out loud”while generating an answer, making the generation process more transparent. Additionally, retrieval-augmented generation (RAG) 43 can make the generated text grounded in trustworthy sources by linking facts in LLM answers to exact passages from medical guidelines 44 or by fact-checking the answers 45 . Such measures are also important in the light of the so-called“control problem” 46 , which evidences humans becoming complacent with and over-reliant on reliable autonomous systems, leading to less rigorous manual checking of outputs. Another significant concern for physicians was outsourcing healthcare to large technology companies, which have developed numerous popular large language models (LLMs), highlighting the need for investment in locally developed and tailored solutions for individual clinics. Furthermore, a concern highly ranked by all three groups was privacy considerations – as the medical domain often works with sensitive patient data, it is paramount to ensure their personal information will be protected 47 . On top of that, students highlighted the loss of the fiduciary relationship between patients and doctors, which could arise if machines are the ones making the final decisions. Finally, administrative workers had the impact on workforce needs as their highest concern. One reason for this spike could be that they are particularly afraid of their work being automated, but this has to be confirmed in further research before conclusions can be drawn. Nevertheless, the role of LLMs as tools that assist humans instead of fully automating work should be emphasized when adopting the technology and in educational training. One major finding of our survey was the negative feeling about the level of preparedness of participants’ institutions for the introduction of LLMs and AI into work. More than 80% of physicians feel their clinics are not ready for it, with only around 10% of both students and physicians thinking they are ready. When asked how to improve it, the majority of all respondents found it crucial to provide proper education and onboarding for employees on how LLMs work and what they can achieve and be used for. We encourage future work and efforts to increase public outreach and organize seminars and tutorials where medical personnel would be introduced to the potential and practical use of LLMs. Other crucial topics that training should cover are best practices and caveats of prompting, as well as the aforementioned control problem. Another common factor for successful adoption was ensuring a high level of privacy and security, as well as compliance of models with laws and regulations. This can be achieved by deploying local LLMs, which would be run on private servers belonging to clinics and universities, thus ensuring that no sensitive data leaves the premises and providing autonomy over the AI tools. Similar recurring themes included investment in proper technical infrastructure, digitalization of paper-written clinical texts, proper coordination with local IT departments, and integration within existing hospital systems that workers are already accustomed to. However, especially the last fact could prove to be a significant challenge in Germany, as there is a lack of standardization in IT systems across clinics 48 . Finally, a smaller number of responses highlighted the need for high output factuality and also good LLM capabilities in the native language of those using the models. The finding of the institutional lack of preparedness to introduce LLMs stands in contrast to the existing frequent usage of LLMs by physicians. Based on statements provided in answers, this discrepancy is mostly explained by the private use of LLMs for work when the clinic does not provide any guidelines. At the same time, this represents a great risk to privacy and data security (since there is no regulation or training) and shows the high value of LLMs to physicians who are ready to accept this risk. A similar situation is present in the adoption of legal AI in law firms, where lawyers already use LLMs personally to help them with email drafting and legal cases, but firms often do not regulate or mandate how to use it 49 . Considering that many questions in our survey were based on the study by Scheetz and Rothschild et al. 50 , conducted in 2019 and focusing on AI for medical imaging, we can draw some comparisons. While only 20% of physicians in their survey actively used AI tools, 50% did so in ours – this shows the general rise in popularity of AI driven by user-friendly chat interfaces and widespread promotion of LLMs. While participants in both surveys agreed the impact of AI on their field is positive, more respondents agreed in their survey (71%) compared to ours (57%). This could point to the fact that the higher popularity of LLMs comes with a higher awareness of potential risks and liabilities associated with them. One notable difference is the question on preparedness of the hospitals for AI adoption (34% said“not prepared”in theirs and 83% in ours), showing the different degree of digitalization and challenges with AI adoption between the two regions. Another major difference was their finding of“improved patient access to disease screening”as the most promising benefit selected by physicians, which was not among top 3 in our survey. This is likely because visual models are mostly used in medicine to detect disease from imaging, while LLMs have a wider range of applications. Their survey also showed a two times higher percentage of respondents concerned about the change in workforce needs – the lower number in our survey might indicate that better familiarity with LLMs also comes with recognizing their limitations and how they mainly assist but cannot fully automate clinical work. Looking at the path forward, there are several important actionable considerations. There is a large need for organizing educations, seminars, and practical courses on the use of AI targeted towards the medical community and adapted for different groups and use cases. Three medical groups we investigated showed different needs and concerns, therefore the developed solutions should be tailored to different groups or provide a certain filter (mask) to select only those features relevant to that group. Furthermore, the developed systems should include a degree of explainability of their decisions, which can be achieved by grounding responses in trustworthy sources, automated fact-checking, reasoning chains, and other symbolic approaches that enhance interpretability. The current hospital systems should be digitalized, including documentation and guidelines, for easier integration of more sophisticated AI tools into their systems. Any developed AI solutions have to respect the privacy of patient data, which can be achieved through anonymization techniques and methods like federated learning. Finally, physicians, students, and staff should always be included and consulted in the solution development process. We acknowledge several limitations of our study. Filling out the survey was completely voluntary and anonymous, so the sample is not representative. The relatively low participant count for the administrative staff group adds to this issue, so the statistical results should be interpreted carefully. Subgroup differences or reported percentages may reflect random variation rather than true patterns in the broader population. Moreover, the sample is likely biased toward technology-savvy participants who already use LLMs and were more inclined to complete the survey. Additionally, the high proportion of medical students and the choice of university clinics (as opposed to non-academic and rural hospitals) likely skews the sample toward a younger and more digitally engaged demographic. This may limit generalizability and the lack of readiness for AI adoption that we identified is likely even more severe in the broader, non-academic workforce. Despite this tech-savvy bias, 30% of respondents still indicated they never use LLMs. Another limitation is the study’s limited geographic scope, as all participants are based in Germany, specifically in Bavaria. German health-system governance is decentralized, involving both the German federal and state governments, which can limit generalizability to the rest of Germany. Therefore, similar studies in other regions could help in evaluating the generalizability of our findings. A similar concern is the timeliness of results, since the survey was conducted in 2024 and AI is a technically fast-moving field so people’s perceptions of the technology can shift rapidly. An additional threat to validity is the potential for researcher bias, as some of the coding procedures and discussion points in our study involve subjective interpretations. We aimed to mitigate this by discussing the findings with a team of three researchers, limiting individual biases. Finally, the findings may be affected by social desirability bias, meaning the data and derived findings are based on the information the survey respondents wanted to share with us and may not necessarily capture the complete picture. Conclusion In this study, we conducted a questionnaire survey with the aim of investigating the expectations and opinions of the medical community in Germany on the use of increasingly popular large language models in their daily work. In total, 70 medical students, 36 physicians, and 14 administrative workers shared their thoughts about potential use cases, perceived degree of impact, quality expectations, advantages and risks, and factors for future adoption of LLMs in medicine. Many participants already use LLMs as a tool for research support, as well as summarization, translation, or report drafting, and see patient-centered use cases like simplification and translation of medical reports as particularly relevant. Most participants see LLMs as having a positive influence on their field and assess that the impact is already present and moderately high. They greatly appreciate using LLMs to save time by automating repetitive tasks involving written text, which can lead to more cost-effective and personalized patient treatment. Concerns were shown regarding the opaque nature of LLMs’ inner workings, confidential data privacy, the influence of large tech companies, and the potential loss of trust between patients and physicians. Participants overwhelmingly do not find their current institution well-equipped for the introduction of LLMs. Suggestions to bridge this gap include offering broader education and specialized training on the use of LLMs in medicine, making higher investments in digitalization, ensuring good infrastructure, standardizing IT systems, ensuring legal compliance and data protection, and fostering a culture of technological openness. We hope that our work will encourage better-informed design and development of future LLM-based tools for medical workers, taking into account their needs, thoughts, and requirements, leading to a better quality of patient treatment and higher job satisfaction of medical personnel. Methods Survey design Our study was based on an anonymous online survey in the form of a questionnaire. The survey consisted of 22 questions, out of which 17 were multiple-choice questions, and 5 were questions with free-form input. We separated questions into four logical blocks. The first block asks basic occupational questions and inquires about whether the participant already uses LLMs – if yes, for which use cases – and how relevant were some selected use cases in their daily work. The second question block revolves around the opinion on how soon the LLMs will have a noticeable impact in their field and how it will affect the workforce. The third question block asks about how ready they think their institution is to adopt AI technologies and what kind of error level and quality the language models should have to be helpful. The fourth question block focuses on the perceived benefits and perceived risks of the adoption of LLMs in their field. The final question is for any additional remarks on the survey. The survey questions were originally written in German, however, for this paper, we translated all questions and answers to English. All the original German and translated English questions can also be found in Supplementary materials . The majority of questions were based on the existing survey from Scheetz and Rothschild et al. 50 , conducted in 2019 on the use of vision AI among physicians. This allowed, among other findings, to compare how the attitudes have changed in the last five years and how different opinions on LLMs are compared to computer vision tools. After adapting the existing questions to be specific to LLM usage, we performed a pilot run and checks with two in-house university physicians and modified or rewrote any questions that were unclear. Data collection The questionnaire was primarily hosted on Bayern Collab 51 , a safe and secure server based on Confluence software and managed by Digitalverbund Bayern ( https://digitalverbund.bayern ). Confluence is a web-based corporate wiki software, and Digitalverbund Bayern is the official association of Bavarian universities and higher education institutions. The platform Bayern Collab was recommended by the Ethical Committee of the Technical University of Munich. The users had to use their university credentials to log in to the platform, but the data collection process itself was completely anonymous. The Bayern Collab platform records only the timestamp of a survey response, which is completely unidentifiable. The survey was open and accepting answers for six months, from April 2024 to October 2024. The collected electronic data was stored on a password-protected and backed-up computer drive on the safe server and remained confidential – only the two principal investigators (paper authors) had access to it and used it to run statistical descriptive analyses on the aggregated answers, to understand and interpret the results. Survey distribution The survey was targeted at three professional groups: medical doctors, medical administrative personnel, and students of medicine. The survey was distributed at five Bavarian university clinics: TUM University Clinic – Klinikum Rechts der Isar (MRI), LMU University Clinic (LMU Klinikum), University Clinic Augsburg, University Clinic Regensburg, and University Clinic Würzburg. The survey was distributed through official channels and intranet postings at the respective clinics and targeted at the three surveyed groups. Owing to the anonymous nature of the survey, we cannot distinguish how many responses came from which of the clinics. Our assumption is that the majority comes from the TU Munich, as the starting point of the study, and the other clinics are somewhat equally represented. Ethical approval and consent The study was performed in accordance with the Declaration of Helsinki. The prospective online survey was officially approved by the Ethical Committee of the Technical University of Munich 52 (decision number: 2023-596-S) and by the Data Protection Office of the Technical University of Munich 53 (decision number: VT-1057). These approvals were valid for the conduction of the survey on other Bavarian university clinics as well. Before starting the survey, each participant gave electronic informed consent and was also provided with a 4-page document describing the consent and how the data will be stored and analyzed (generated by the official tool eTIC –- electronic Tool for the compilation of Informed Consent documents 54 ). Additionally, participants were provided with a two-page study protocol describing the purpose and conduction of the study, as well as a one-page document outlining a brief introduction to the most recent advancements in biomedical NLP and a description of how LLMs work. These are provided as attachments to our submission. Data analysis methods Only fully completed survey responses were taken into consideration for data analysis. The data was analyzed using Python libraries numpy , pandas , and scipy , while the plots were generated with the Python library matplotlib . Categorical responses were compared according to the participants group using Pearson’s test, for the test of independence of the observed frequencies in contingency tables, while its effect size was estimated with Cramér’s V . Ordinal responses (Likert scale 1–5) were compared with regard to participant groups using the Kruskal-Wallis test to test whether samples originate from the same distribution, while its effect size was estimated with . To determine the most common themes in questions related to the existing use cases of LLMs and suggestions for successful LLM adoption, we employed the hybrid thematic analysis approach. Two authors followed a six-step framework proposed by Braun and Clarke 55 : (1) explore the free-form answers; (2) generate initial codes for observed phenomena; (3) identify potential themes and sub-themes in codes; (4) review and rephrase; (5) identify the overarching themes and merge similar ones; (6) write up a narrative. The codes were generated with an inductive, data-driven approach independently by the two authors after reading and reviewing the responses. Any discrepancy in coding was resolved through discussion and consensus in regular weekly meetings, where codes were discussed, clarified, and refined. To supplement our qualitative interpretation, the final themes were quantified through content analysis to determine their frequency in the dataset. To evaluate the responses to questions related to the benefits and risks of LLMs, we used a scoring system where the 1st ranked choice was given 3 points, 2nd choice 2 points, and 3rd choice 1 point. The total score was averaged by the number of participants in a group. Supplementary Information Supplementary Information 1. (392.9KB, pdf) Supplementary Information 2. (43.4KB, xlsx) Acknowledgements We would like to thank all those at the participating university clinics who helped distribute and promote the survey in their channels, in particular Hayim Malkhasy, Christoph Schmaderer, and Stefan Leucht from the TUM University Hospital. Author contributions J.V. and A.F. conceived the survey, promoted the survey, curated the data, performed data analysis, and wrote the original draft. F.M. provided supervision, handled communication with university clinics, did participant acquisition, and reviewed and edited the manuscript. Funding Open Access funding enabled and organized by Projekt DEAL. Data availability All data generated or analyzed during this study, including the raw export of all survey responses, are included in this published article and its supplementary information files. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary Information The online version contains supplementary material available at 10.1038/s41598-026-45245-2. References 1. Russell, S., & Norvig P. Artificial Intelligence, Global Edition A Modern Approach (Pearson Deutschland, 2021). 2. Eisenstein, J. Introduction to Natural Language Processing (The MIT Press, 2019). 3. Chang, Y. et al. A survey on evaluation of large language models. ACM Trans. Intell. Syst. Technol. 15 , 1–45 (2024). [ Google Scholar ] 4. OpenAI et al. GPT-4 Technical Report. arXiv e-prints arXiv:2303.08774 , (2023). 5. Wang, B. et al. Pre-trained language models in biomedical domain: A systematic survey. ACM Comput. Surv. 56 , 1–52 (2023). [ Google Scholar ] 6. Thirunavukarasu, A. J. et al. Large language models in medicine. Nat. Med. 29 , 1930–1940 (2023). [ DOI ] [ PubMed ] [ Google Scholar ] 7. Patel, S. B. & Lam, K. Chatgpt: The future of discharge summaries?. Lancet Digit. Health 5 , e107–e108 (2023). [ DOI ] [ PubMed ] [ Google Scholar ] 8. Ayers, J. W. et al. Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Intern. Med. 183 , 589–596 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Singhal, K. et al. Large language models encode clinical knowledge. Nature 620 , 172–180 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Vladika, J., Schneider, P. & Matthes, F. MedREQAL: Examining medical knowledge recall of large language models via question answering. In Findings of the Association for Computational Linguistics: ACL 2024 , (eds Ku, L.-W. et al.) 14459–14469, 10.18653/v1/2024.findings-acl.860 (Association for Computational Linguistics, 2024). 11. Gilson, A. et al. How does chatgpt perform on the United States Medical Licensing Examination (USMLE)? the implications of large language models for medical education and knowledge assessment. JMIR Med. Educ. 9 , e45312 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Hager, P. et al. Evaluation and mitigation of the limitations of large language models in clinical decision-making. Nat. Med. 30 , 2613–2622 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Clusmann, J. et al. The future landscape of large language models in medicine. Commun. Med. 3 , 141 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 14. Lai, J., Gan, W., Wu, J., Qi, Z. & Philip, S. Y. Large language models in law: A survey. AI Open (2024). 15. Birhane, A., Kasirzadeh, A., Leslie, D. & Wachter, S. Science in the age of large language models. Nat. Rev. Phys. 5 , 277–280 (2023). [ Google Scholar ] 16. Kasneci, E. et al. Chatgpt for good? on opportunities and challenges of large language models for education. Learn. Individ. Differ. 103 , 102274 (2023). [ Google Scholar ] 17. Pinto dos Santos, D. et al. Medical students’ attitude towards artificial intelligence: A multicentre survey. Eur. Radiol. 29 , 1640–1646 (2019). [ DOI ] [ PubMed ] 18. Scott, I. A., Carter, S. M. & Coiera, E. Exploring stakeholder attitudes towards ai in clinical practice. BMJ Health Inform. 28 (2021). [ DOI ] [ PMC free article ] [ PubMed ] 19. Abd-Alrazaq, A. et al. Large language models in medical education: Opportunities, challenges, and future directions. JMIR Med. Educ. 9 , e48291 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Tian, S. et al. Opportunities and challenges for chatgpt and large language models in biomedicine and health. Brief. Bioinform. 25 , bbad493 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Liu, N., Shapira, P. & Yue, X. Tracking developments in artificial intelligence research: Constructing and applying a new search strategy. Scientometrics 126 , 3153–3192 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Kuhlmann, S. & Heuberger, M. Digital transformation going local: Implementation, impacts and constraints from a German perspective. Public Money Manag. 43 , 147–155 (2023). [ Google Scholar ] 23. Zenker, S. et al. Data protection-compliant broad consent for secondary use of health care data and human biosamples for (bio) medical research: Towards a new german national standard. J. Biomed. Inform. 131 , 104096 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 24. Ladd, M. E. The medical device regulation and its impact on device development and research in Germany. Z. Med. Phys. 33 , 459 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. von Wedel, P. et al. Effects of hospital digitization on clinical outcomes and patient satisfaction: Nationwide multiple regression analysis across German hospitals. J. Med. Internet Res. 24 , e40124 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Weinert, L., Müller, J., Svensson, L. & Heinze, O. Perspective of information technology decision makers on factors influencing adoption and implementation of artificial intelligence technologies in 40 German hospitals: Descriptive analysis. JMIR Med. Inform. 10 , e34678 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Maassen, O. et al. Future medical artificial intelligence application requirements and expectations of physicians in German university hospitals: Web-based survey. J. Med. Internet Res. 23 , e26646. 10.2196/26646 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Jussupow, E., Spohrer, K. & Heinzl, A. Identity threats as a reason for resistance to artificial intelligence: Survey study with medical students and professionals. JMIR Formative Res. 6 , e28750 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 29. Tangadulrat, P. et al. Using chatgpt for clinical practice and medical education: Cross-sectional survey of medical students’ and physicians’ perceptions. JMIR Medical Education 9 , e50658 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 30. Ganjavi, C. et al. Chatgpt and large language models (LLMs) awareness and use. A prospective cross-sectional survey of us medical students. PLoS Digit. Health 3 , e0000596 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Spotnitz, M. et al. A survey of clinicians’ views of the utility of large language models. Appl. Clin. Inform. 15 , 306–312 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Pereira, S., Costa, J. & Hanscheid, T. Medical students’s attitudes, perceptions, and usage of large language models in education: A questionnaire-based study. Jornal da Sociedade das Ciências Médicas de Lisboa 169 (2025). 33. Chatterji, A. et al. How people use chatgpt. Working Paper 34255, National Bureau of Economic Research 10.3386/w34255 (2025). 34. Kim, J. et al. Limitations of large language models in clinical problem-solving arising from inflexible reasoning. Sci. Rep. 15 , 39426 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Klang, E. et al. A strategy for cost-effective large language model use at health system-scale. NPJ Digit. Med. 7 , 320 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 36. Ahmed, S. K. et al. The power of chatgpt in revolutionizing rural healthcare delivery. Health Sci. Rep. 6 , e1684 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. Vicente, A. M., Ballensiefen, W. & Jönsson, J.-I. How personalised medicine will transform healthcare by 2030: the icpermed vision. J. Transl. Med. 18 , 1–4 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 38. Yao, S. et al. Tree of thoughts: Deliberate problem solving with large language models. Adv. Neural. Inf. Process. Syst. 36 , 11809–11822 (2023). [ Google Scholar ] 39. Pan, L., Albalak, A., Wang, X. & Wang, W. Logic-LM: Empowering large language models with symbolic solvers for faithful logical reasoning. In Findings of the Association for Computational Linguistics: EMNLP 2023 (eds Bouamor, H. et al.) 3806–3824, 10.18653/v1/2023.findings-emnlp.248 (Association for Computational Linguistics, 2023). 40. Wei, J. et al. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems , Vol. 35 (eds Koyejo, S. et al.) 24824–24837 (Curran Associates, Inc., 2022). 41. Zhao, H. et al. Explainability for large language models: A survey. ACM Trans. Intell. Syst. Technol. 15 , 1–38 (2024). [ Google Scholar ] 42. Guo, D. et al. Deepseek-r1 incentivizes reasoning in llms through reinforcement learning. Nature 645 , 633–638 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Lewis, P. et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. Adv. Neural. Inf. Process. Syst. 33 , 9459–9474 (2020). [ Google Scholar ] 44. Ferber, D. et al. Gpt-4 for information retrieval and comparison of medical oncology guidelines. Nejm Ai 1 , AIcs2300235 (2024). 45. Vladika, J. & Matthes, F. Scientific fact-checking: A survey of resources and approaches. In Findings of the Association for Computational Linguistics: ACL 2023 , (eds Rogers, A. et al.) 6215–6230 10.18653/v1/2023.findings-acl.387 (Association for Computational Linguistics, 2023). 46. Zerilli, J., Knott, A., Maclaurin, J. & Gavaghan, C. Algorithmic decision-making and the control problem. Mind. Mach. 29 , 555–578 (2019). [ Google Scholar ] 47. Das, B. C., Amini, M. H. & Wu, Y. Security and privacy challenges of large language models: A survey. ACM Comput. Surv. 57 , 1–39 (2025). [ Google Scholar ] 48. Sunyaev, A., Dünnebeil, S., Mauro, C., Leimeister, J. M. & Krcmar, H. It-standards im gesundheitswesen: Überblick und entwicklungsperspektiven mit der einführung service-orientierter architekturen. In Informatik 2010 Service Science – Neue Perspektiven für die Informatik (2010). 49. Vladika, J., Meisenbacher, S., Preis, M., Klymenko, A. & Matthes, F. Towards a structured overview of use cases for natural language processing in the legal domain: A german perspective. In 30th Americas Conference on Information Systems, AMCIS 2024 (Association for Information Systems, 2024). 50. Scheetz, J. et al. A survey of clinicians on the use of artificial intelligence in ophthalmology, dermatology, radiology and radiation oncology. Nat. News . (2021). [ DOI ] [ PMC free article ] [ PubMed ] 51. Bayern collab. https://collab.dvb.bayern/ (Accessed 1 Apr 2025). 52. Tum ethics committee. https://www.tum.de/en/about-tum/organization/ethics-committee . (Accessed 1 Apr 2025). 53. Tum data protection office. https://www.datenschutz.tum.de/datenschutz/verarbeitungstaetigkeit/ . (Accessed 1 Apr 2025). 54. etic – electronic tool for the compilation of informed consent documents. https://etic.med.tum.de/ (Accessed 1 Apr 2025). 55. Braun, V. & Clarke, V. Using thematic analysis in psychology. Qual. Res. Psychol. 3 , 77–101 (2006). [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Information 1. (392.9KB, pdf) Supplementary Information 2. (43.4KB, xlsx) Data Availability Statement All data generated or analyzed during this study, including the raw export of all survey responses, are included in this published article and its supplementary information files. Articles from Scientific Reports are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (1.7 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel 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