Risks and benefits of ChatGPT in informing patients and families with rare kidney diseases- an explorative assessment by the European Rare Kidney Disease Reference Network (ERKNet)

Risks and benefits of ChatGPT in informing patients and families with rare kidney diseases- an explorative assessment by the European Rare Kidney Disease Reference Network (ERKNet) | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Risks and benefits of ChatGPT in informing patients and families with rare kidney diseases- an explorative assessment by the European Rare Kidney Disease Reference Network (ERKNet) Albertien M. van Eerde, Ana Teixeira, Flavia Galletti, Michal Maternik, and 10 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5827993/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 16 Apr, 2025 Read the published version in Pediatric Nephrology → Version 1 posted 5 You are reading this latest preprint version Abstract Background: Rare diseases affect fewer than 1 in 2,000 individuals, but approximately 150 rare kidney diseases account for about 10% of the chronic kidney disease (CKD) population, impacting millions across Europe and globally. The scarcity of medical experts for these conditions results in an unmet need for accurate and helpful patient information. Large language models like ChatGPT may offer a technological solution to assist medical professionals in educating patients and improving doctor-patient communication. We hypothesized that ChatGPT could provide accurate responses to frequently asked basic questions from patients with rare kidney diseases. Methods: Medical professionals and members of European Patient Advocacy Groups (ePAGs) affiliated with the European Rare Kidney Disease Reference Network (ERKNet) simulated patient-ChatGPT interactions using a Microsoft forms questionnaire and ChatGPT 3.5 and 4.0. Participants selected any rare kidney disease for a structured conversation with ChatGPT-3.5 or 4.0. Responses were evaluated for accuracy and helpfulness. Results: 46 ERKNet experts and 12 ePAGs from 13 European countries participated in this study. ChatGPT provided scientifically accurate and helpful information on 28 randomly selected rare kidney diseases, including prognostic information and genetic testing guidance. Participants expressed neutral positions regarding ChatGPT's recommendations on alternative treatments, second opinions, and other information sources. While ChatGPT generally was perceived as helpful and empathetic, concerns about patient safety persisted. Conclusions: ChatGPT exhibited substantial potential in addressing patient inquiries regarding rare kidney diseases in a real-world context. While it demonstrated resilience against misinformation in this application, careful human oversight remains essential and indispensable. Rare disease rare kidney disease ChatGPT artificial intelligence AI patient education genetic testing patient advocacy large language model Figures Figure 1 Introduction Rare kidney diseases (RKDs) collectively represent a significant subset of chronic kidney disease (CKD), accounting for approximately 5-10% of CKD cases in adults and almost all cases of CKD in children in Europe. 1,2 Although each individual RKD affects fewer than 1 in 2,000 individuals, the cumulative impact is substantial, involving about two million patients across Europe and many more globally. 3,4 These conditions encompass a broad spectrum of genetic, structural, and functional disorders, which manifest in pediatric and adult populations. 5 The diversity and complexity of RKDs present significant challenges not only for diagnosis and treatment but also for patient education, communication, and support. 6 The European Rare Kidney Disease Reference Network (ERKNet) was established to address these challenges by providing a collaborative platform for clinical care, research, and education across Europe. ERKNet unites specialists in pediatric and adult nephrology, human genetics, pathology, and patient advocacy to enhance awareness, diagnosis, and treatment of rare kidney diseases (RKDs). Despite these efforts, there remains a considerable information deficit among patients and their families, particularly regarding disease management, prognosis, and available treatments. The resulting information deficit experienced by patients and their families often prompts them to independently seek information on the internet, including through social media. 7 It is well established that medical information found on social media and non-professional platforms carries a significant risk of misinformation, which can potentially be harmful. 8 The advent of artificial intelligence (AI) and large language models (LLMs) such as ChatGPT ( G enerative P re-trained T ransformer) offers a promising tool to bridge this information gap and in this way reduce inequities in access to information. 9 ChatGPT, developed by OpenAI, is pre-trained on extensive text data, fine-tuned with model-specific tools, and designed to provide responses to user queries in a conversational manner. 9 It is the most popular and most frequently used LLM today. While the potential of AI-driven models in patient education is widely recognized, their application in providing reliable medical information to patients, especially in the context of rare diseases, is still under investigation. 10–12 Previous studies have examined the use of ChatGPT in various medical contexts, highlighting its capacity to produce coherent, contextually appropriate responses and even demonstrating its ability to pass the United States Medical Licensing Examination (USMLE). 13,14 However, concerns regarding the accuracy, relevance, and safety of the information provided by ChatGPT persist. In particular, there is a need to evaluate whether ChatGPT can reliably address the specific needs of patients with rare diseases, where the availability of specialized knowledge is crucial. This study aims to assess the performance of ChatGPT 3.5 and 4.0, as the post frequently used LLM, in providing accurate and helpful information to patients with rare kidney diseases in a real-world setting. By engaging ERKNet experts and members of European Patient Advocacy Groups (ePAGs) in simulated patient-ChatGPT interactions, we evaluate the model's ability to deliver information that aligns with current clinical knowledge and meets the needs of patients and their families. Methods Ethical approval: Ethical approval for this survey study was obtained from the University of Cologne Institutional Review Board (IRB). Recruitment of participants: Professional participants at the 8th annual meeting of the European Rare Kidney Disease Network (ERKNet), held in Venice, Italy in 2024, were invited to take part in this study. Eligible participants included “ERKNet experts,” such as pediatric nephrologists, adult nephrologists, human geneticists, or basic researchers with experience in treating patients with rare kidney diseases, as well as ePAGs involved in rare kidney diseases. Participation was voluntary and anonymous. All participants were required to sign up for either a free ChatGPT-3.5 or ChatGPT-4.0 account, if they had not done so previously. Survey development: The ChatGPT-patient interactions were carefully developed based on clinical experience of ERKNet experts and ePAGs caring for patients with RKD. Survey questions and instructions for interacting with ChatGPT were compiled in a Microsoft Forms document. ( Supplemental Table 1 ). To avoid responses influenced by previous chats on participants' ChatGPT accounts, the survey began with the prompt: "For this conversation, please treat me as a non-medical user with an average education." At the start of the conversation, participants were instructed to select any rare kidney disease within their field of expertise, thereby minimizing bias in selection of RKD ( Supplemental Table 1 ). Participants could adopt either a pediatric perspective ("my child has") or an adult perspective ("I have"). All follow-up questions pertained to the initially selected condition. Disease-related questions, which had to be copied and pasted into ChatGPT by the participants, included: I have/My child has [name of rare kidney disease]. Please explain what that is. I am worried. Am I/Is my child going to be very sick? Should I/we get genetic testing? Are there any helpful dietary modifications or supplements? Are there any alternative treatments? Where can I find a doctor for a second opinion close to [name of a familiar city]? Are there any other reliable information sources? What disease does our child have? Explain in plain language. Participants scored each ChatGPT response to these questions based on two criteria: scientific correctness (Does the provided response align with current clinical knowledge and scientific understanding?) and helpfulness (In your opinion, how helpful would this response be for a patient or their family?). In agreement with our real-word approach, an ordinal scale was used for scoring, with the following values: 1 (extremely negative), 2 (negative), 3 (neutral), 4 (positive), and 5 (extremely positive). Median scores of 4 ("positive") and 5 ("extremely positive") indicate broad agreement among our participants with ChatGPT. A median score of >2 and <4 was considered neutral. A median score of 1 (“extremely negative”) and 2 (“negative”) indicate broad disagreement with ChatGPT. ERKNet experts were requested to formulate one or two "expert-level questions" to their selected condition to challenge ChatGPT's capabilities. In contrast, ePAGs were asked to challenge ChatGPT by presenting a hypothetical critical or emotional patient scenario and seeking its assistance. Additional questions were included to collect information on the version of ChatGPT used (3.5 or 4.0), the participant’s age group (50 years), field of expertise (free text), role (ERKNet expert or ePAG), and prior experience with ChatGPT (for the first time, for fun, for assistance in scientific writing, or for everyday tasks such as emails or medical reports). Use of ChatGPT for Manuscript Preparation: In the development of this manuscript, the authors utilized ChatGPT to refine the language, similarly to employing a native-speaking editor. Following the use of this tool, the authors thoroughly reviewed and edited the content as necessary and assume full responsibility for the final version of the publication. Results A total of 54 participants (42 ERKNet experts and 12 ePAG representatives) provided valid responses that were included in this analysis. Responses from 4 ERKNet experts were excluded due to the use of imprecise disease terminology (e.g., 'polycystic kidney disease' rather than specifying ADPKD or ARPKD). Among the 54 participants, 34 were aged 30–50 years, and 20 were over 50 years old. Participants were from various countries, including Germany (n=13), the Netherlands (n=8), Italy (n=6), Spain (n=5), Belgium (n=2), Poland (n=2), Sweden (n=2), the United Kingdom (n=2), and one participant each from the Czech Republic, France, Ireland, Malta, Romania, and Slovenia, while 8 participants did not wish to reveal their country of origin ( Supplemental Figure 1 ). In terms of professional background, 32 participants identified as pediatric nephrologists, 7 as adult nephrologists, and 3 as pathologists. Regarding prior experience with ChatGPT, 16 participants reported using it for the first time, 19 had used it only “for fun”, and 19 had also used it for work-related tasks. The 54 participants selected a total of 28 different rare kidney diseases. The most frequently selected conditions included atypical hemolytic uremic syndrome (aHUS) (n=6), autosomal recessive polycystic kidney disease (ARPKD) (n=6), cystinosis (n=5), nephrotic syndrome (n=5), autosomal dominant polycystic kidney disease (ADPKD) (n=4), nephronophthisis (n=3), Alport syndrome (n=3), Gitelman syndrome (n=2), posterior urethral valves (n=2), primary hyperoxaluria (n=2), and thrombotic microangiopathy (n=2) ( Figure 1 ). Figure 1: 28 rare kidney diseases selected by 54 ERKNet experts and ePAGs Abbreviations: ADPKD, autosomal dominant polycystic kidney disease; ADTKD, autosomal dominant tubulointerstitial kidney disease; aHUS, atypical hemolytic uremic syndrome; APRTD, adenine phosphoribosyltransferase deficiency; ARPKD, autosomal recessive polycystic kidney disease; CF, cystic fibrosis; FHHNC, familial primary hypomagnesemia with hypercalciuria and nephrocalcinosis; FMF, familial mediterranean fever; MSpK, medullary sponge kidney; NDI, nephrogenic diabetes insipidus; NPHP, nephronophthisis; PHA, pseudohypoaldosteronism; PUV, posterior urethral valves; TMA, thrombotic microangiopathy; XLH, X-linked hypophosphatemia. For evaluating whether ChatGPT's responses to various survey questions align with current scientific knowledge, we considered only the scores from 42 ERKNet experts ( Table 1 ). For evaluating whether ChatGPT's responses to various survey questions are helpful for patients and families, we considered scores from ERKNet experts and ePAGs( Table 1 ). Our findings demonstrate that both ChatGPT 3.5 and 4.0 provide explanations of rare kidney diseases to patients and families that are consistent with scientific understanding and are considered helpful for patients and families ( Table 1 ). Additionally, the prognostic information about the underlying disease and guidance on the decision whether to obtain genetic testing are presented accurately and in a helpful manner ( Table 1 ). However, ERKNet experts and ePAGs expressed concerns about ChatGPT's responses to questions related to alternative treatments, options for seeking a second opinion in various European cities, and recommendations for other reliable information sources. ( Table 1 ). ChatGPT's ability to explain diseases in plain language was considered accurate and helpful ( Table 1 ). Responses to random expert-level questions ( Supplemental Table 2 ) from participating experts were generally accurate ( Table 1 ). In one instance, an expert remarked that the response in context of nephrogenic diabetes insipidus, "seek medical attention if signs of dehydration, such as dry mouth, sunken eyes, or decreased urination occur," was "inappropriate" and "potentially harmful", as patients should seek medical attention at an earlier stage. ChatGPT's responses to the “emotional challenges” presented by ePAGs received a median score of 3, with a relatively broad interquartile range of 2.25, indicating mixed satisfaction ( Table 2 ). ERKNet experts and ePAGs generally agreed that ChatGPT is helpful and empathetic ( Table 2 ). However, concerns about safety of this new technology in general remained, as reflected by a neutral score on that specific question ( Table 2 ). About half of the participants shared comments on one or more survey questions. Overall, we received 50 comments from 36/54 participants comprising a mixed feedback on specific ChatGPT responses and general matters ( Supplemental Table 3 ). Many appreciated ChatGPT’s clear and readable responses (“I am a bit suprised. The answers are way better than I thought they would be”). However, several participants criticized the responses for being too general, lacking specific medical details, and sometimes offering information that was not directly relevant to the condition in question (“Very generic answer. Uses terms such as "reabsorption", which is probably meaningless or even confusing without explanation.”). Concerns were raised about ChatGPT's suggestions of alternative treatments, such as herbal remedies and complementary medicine, which were seen as potentially misleading or unsafe (“Recommends ginger and liquorice! As well as any other quackery around, such as "Mind-body-techniques””). Participants also observed that ChatGPT often recommended consulting a healthcare provider, which was perceived positively. However, the exclusion of important resources such as the National Institutes of Health (NIH), the European Medicines Agency (EMA), the International Pediatric Nephrology Association (IPNA), the Pediatric Nephrology Research Consortium (PNRC), Nephcure, and the European Rare Kidney Disease Reference Network (ERKNet) was considered a notable limitation. Additionally, the language used in responses was sometimes considered too technical or abstract for the average patient, and some answers appeared to be more U.S.-centric rather than tailored to a European audience (“ERKNet and ESPN as well as ESPU are missing. ChatGPT suggests rather US organizations like Mayo Clinic and National Kidney Foundation”). Table 1: Evaluation of the “scientific correctness” and “helpfulness” of ChatGPT Responses Survey copy & paste prompt for ChatGPT Score, median (1 st & 3 rd quartile) Result Score, median (1 st & 3 rd quartile) Result Correctness (42 ERKNet experts) Helpfulness (42 ERKNet experts and 12 ePAGs) My child has [name of rare kidney disease]. Please explain what that is. 4 (4, 4) Positive 4 (3, 5) Positive I am worried. Is my child going to be very sick? 4 (3.25, 5) Positive 4 (3, 5) Positive Should we get genetic testing? 4.5 (4, 5) Positive 4 (4, 5) Positive Are there any helpful dietary modifications or supplements? 4 (3, 4) Positive 4 (3, 5) Positive Are there any alternative treatments? 3 (2, 4) Neutral 3 (2, 4) Neutral Where can I find a doctor for a second opinion close to [name of a familiar city]? n/a n/a 3 (2, 4) Neutral Are there any other reliable information sources? n/a n/a 3 (2, 4) Neutral What disease does our child have? Explain in plain language. n/a n/a 4 (3, 4) Positive ERKNet expert question 1 (n=54) 4 (3, 5) Positive n/a n/a ERKNet expert question 2 (n=23) 3 (3, 4.5) Neutral n/a n/a Abbreviations: n/a, not applicable. Table 2: General Aspects of ChatGPT Responses Survey Question to ERKNet Experts and ePAGs on General Aspects of ChatGPT Responses Score by ERKNet experts, median (1 st & 3 rd quartile) Result Score by ePAGs, median (1 st & 3 rd quartile) Result In your opinion, how helpful could ChatGPT be for patients with rare diseases? 4 (3, 4) Positive 3.5 (2.75, 4) Neutral In your opinion, are the responses by ChatGPT empathic? 4 (3, 4) Positive 3.5 (2, 4) Neutral Based on your experience in this survey, how safe may ChatGPT be for patients and families? 3 (3, 4) Neutral 3 (2.75, 4) Neutral ePAGs emotional challenge scenario n/a n/a 3 (2, 4.25) Neutral Discussion With the help of 42 ERKNet experts and 12 ePAGs, we assessed the potential risks and benefits of ChatGPT as an information source for patients and families with rare kidney diseases. Consistent with previous studies evaluating ChatGPT's effectiveness in patient education for common conditions such as prostate cancer, amblyopia, or systemic lupus erythematosus, our international survey participants reported that ChatGPT provided accurate and helpful responses to both basic and advanced questions across a wide range of rare kidney diseases. Our survey highlights the great potential of ChatGPT in efficiently compiling and presenting information on a wide array of rare medical conditions, while tailoring it to address specific individual queries. Importantly, we did not encounter any responses from ChatGPT that were entirely incorrect or, more critically, immediately dangerous for patients. However, we did encounter responses that were somewhat vague, confusing, arbitrary, not directly relevant to the topic, and ultimately not helpful. Some experts raised concerns that questions about alternative treatment options or dietary modifications elicited responses from ChatGPT that were somewhat evasive, potentially undermining the primary emphasis on evidence-based therapy. Questions regarding other resources and medical centers for a second opinion often resulted in US-centric responses, which were frequently deemed unhelpful for mostly European users. However, with ongoing advancements in LLM technology, these issues are likely to be addressed in future versions or medical-content specific GPTs. Considering the undefined nature of training datasets for AI-powered public chatbots, including ChatGPT, and the "black box" nature of AI-driven decision-making, the authors argue that it is premature to deem this technology suitable for patient education without human oversight. LLM-generated "hallucinations," which can be described from a human perspective as inaccurate or inappropriate responses, as we encountered on the subjects of "alternative treatments" and "second opinions," are a potentially dangerous phenomenon. 15 Notably, ChatGPT demonstrates greater resilience to hallucinations compared to other LLMs. 15 In this study, our questions regarding alternative treatment methods likely prompted ChatGPT to give hallucinatory responses. It is concerning that large language models (LLMs) do not clearly indicate when their knowledge is insufficient to provide a robust, fact-based response, and instead present disputed viewpoints in an overly eloquent manner or, in some instances, relay incorrect information. For instance, when repeatedly asked, "Who wrote the book From Fish to Philosopher ?" (the correct answer being Homer W. Smith, 1953), ChatGPT 3.5, at the time this study was conducted, provided numerous names with coherent explanations, though all were incorrect. This underscores the need for caution: while ChatGPT has the potential to enhance patient empowerment by providing accessible information, it is essential to remind patients that this information could be incorrect and that health-related decisions should always be discussed with their healthcare providers. This study has several limitations. Firstly, while participants were asked to use an initial prompt to avoid influence from prior chats, we chose to not control ChatGPT metrics, as this study was designed as a test balloon for real-world ChatGPT-patient interactions. Additionally, the study focused only on ChatGPT versions 3.5 and 4.0, excluding newer models such as 4o and LLM from other companies. Despite the study’s limitations and the need for caution, the authors conclude that ChatGPT represents a significant technological advancement with a lower susceptibility to misinformation and manipulation by individual content creators compared to social media platforms. To enhance patient-ChatGPT interactions, a viable approach is the implementation of carefully crafted prompting strategies. Therefore, based on our experience with this survey, we have developed a set of prompting expressions designed to elicit individualized, accurate, and evidence-based responses from ChatGPT, thereby enhancing the provision of safe and reliable medical information. ( Info Box 1 ). Info Box 1: Useful Prompting Expressions for Patient-ChatGPT Interaction Obtaining trustworthy information: "Can you provide me with trustworthy medical information about [disease/condition] from reliable sources like WHO or national health organizations?" "What are the scientifically proven treatments for [disease/condition]?" "What treatments for [disease] should I avoid because they lack scientific evidence?" Adapting Language According to Education Level: For patients with higher education: "Can you explain the current research and scientific consensus on the treatment options for [disease]?" For patients with lower education: "Can you explain in simple terms what causes [disease] and how it can be treated?" Regional context: “Please take into consideration that I live in [city/country].” To mitigate the limitations posed by uncontrollable ChatGPT pre-training dataset sources, ERKNet has launched a project leveraging LLM technology fine-tuned on carefully curated datasets, specifically tailored to rare kidney diseases. This approach could provide a safer information source for patients, families, and medical professionals in the future. The urgent need to disseminate genetic and RKD knowledge among nephrology care providers was recently underscored by KDIGO. 16 In this context, expert-trained AI models hold great promise also in assisting physicians, particularly in counseling for ultrarare diseases, where most caregivers lack personal experience. Ultimately, ChatGPT and similar models have the potential to serve as valuable tools for both clinicians and patients by assisting in the consolidation of disease-related knowledge. This could free up time in fast-paced clinical settings for more efficient and productive interactions between expert doctors and informed patients, thereby enhancing the overall quality of healthcare. Declarations Disclosure statement The authors declare that they have no conflicts of interest relevant to this study. No financial support, funding, or specific external influences have impacted the design, conduct, interpretation, or reporting of this research. The company “Open AI” was not involved in any aspect of this work, including data collection, analysis, or manuscript preparation. Acknowledgments This work was generated within the European Reference Network for Rare Kidney Diseases (ERKNet), a project funded by the European Union within the framework of the EU4Health Programme 2021-2027. The authors thank all survey participants for their valuable contributions. References Wühl E, van Stralen KJ, Wanner C, et al. Renal replacement therapy for rare diseases affecting the kidney: an analysis of the ERA-EDTA Registry. Nephrol Dial Transplant . 2014;29 Suppl 4:iv1-8. doi:10.1093/ndt/gfu030 Wong K, Pitcher D, Braddon F, et al. Effects of rare kidney diseases on kidney failure: a longitudinal analysis of the UK National Registry of Rare Kidney Diseases (RaDaR) cohort. Lancet . 2024;403(10433):1279-1289. doi:10.1016/S0140-6736(23)02843-X Devuyst O, Knoers NVAM, Remuzzi G, Schaefer F, Board of the Working Group for Inherited Kidney Diseases of the European Renal Association and European Dialysis and Transplant Association. Rare inherited kidney diseases: challenges, opportunities, and perspectives. Lancet . 2014;383(9931):1844-1859. doi:10.1016/S0140-6736(14)60659-0 Bassanese G, Wlodkowski T, Servais A, et al. The European Rare Kidney Disease Registry (ERKReg): objectives, design and initial results. Orphanet J Rare Dis . 2021;16(1):251. doi:10.1186/s13023-021-01872-8 Snoek R, van Jaarsveld RH, Nguyen TQ, et al. Genetics-first approach improves diagnostics of ESKD patients <50 years old. Nephrol Dial Transplant . 2022;37(2):349-357. doi:10.1093/ndt/gfaa363 Aymé S, Bockenhauer D, Day S, et al. Common Elements in Rare Kidney Diseases: Conclusions from a Kidney Disease: Improving Global Outcomes (KDIGO) Controversies Conference. Kidney Int . 2017;92(4):796-808. doi:10.1016/j.kint.2017.06.018 Sinha J, Serin N. Online Health Information Seeking and Preventative Health Actions: Cross-Generational Online Survey Study. J Med Internet Res . 2024;26:e48977. doi:10.2196/48977 Kirkpatrick CE, Lawrie LL. TikTok as a Source of Health Information and Misinformation for Young Women in the United States: Survey Study. JMIR Infodemiology . 2024;4:e54663. doi:10.2196/54663 Haupt CE, Marks M. AI-Generated Medical Advice-GPT and Beyond. JAMA . 2023;329(16):1349-1350. doi:10.1001/jama.2023.5321 Lambert R, Choo ZY, Gradwohl K, Schroedl L, Ruiz De Luzuriaga A. Assessing the Application of Large Language Models in Generating Dermatologic Patient Education Materials According to Reading Level: Qualitative Study. JMIR Dermatol . 2024;7:e55898. doi:10.2196/55898 Gibson D, Jackson S, Shanmugasundaram R, et al. Evaluating the Efficacy of ChatGPT as a Patient Education Tool in Prostate Cancer: Multimetric Assessment. J Med Internet Res . 2024;26:e55939. doi:10.2196/55939 Haase I, Xiong T, Rissmann A, Knitza J, Greenfield J, Krusche M. ChatSLE: consulting ChatGPT-4 for 100 frequently asked lupus questions. Lancet Rheumatol . 2024;6(4):e196-e199. doi:10.1016/S2665-9913(24)00056-0 Dave T, Athaluri SA, Singh S. ChatGPT in medicine: an overview of its applications, advantages, limitations, future prospects, and ethical considerations. Front Artif Intell . 2023;6:1169595. doi:10.3389/frai.2023.1169595 Kung TH, Cheatham M, Medenilla A, et al. Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models. Dagan A, ed. PLOS Digit Health . 2023;2(2):e0000198. doi:10.1371/journal.pdig.0000198 McIntosh TR, Liu T, Susnjak T, Watters P, Ng A, Halgamuge MN. A Culturally Sensitive Test to Evaluate Nuanced GPT Hallucination. IEEE Trans Artif Intell . 2024;5(6):2739-2751. doi:10.1109/TAI.2023.3332837 KDIGO Conference Participants. Genetics in chronic kidney disease: conclusions from a Kidney Disease: Improving Global Outcomes (KDIGO) Controversies Conference. Kidney Int . 2022;101(6):1126-1141. doi:10.1016/j.kint.2022.03.019 Supplementary Files ChatGPTERKNetsupplementalfile250107.docx Cite Share Download PDF Status: Published Journal Publication published 16 Apr, 2025 Read the published version in Pediatric Nephrology → Version 1 posted Editorial decision: Minor Revisions Needed 27 Feb, 2025 Reviewers agreed at journal 14 Jan, 2025 Reviewers invited by journal 14 Jan, 2025 Editor assigned by journal 14 Jan, 2025 First submitted to journal 14 Jan, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5827993","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":402173492,"identity":"50430ad2-a2d9-4263-adef-a1c453afc397","order_by":0,"name":"Albertien M. van Eerde","email":"","orcid":"","institution":"University Medical Center Utrecht Genetics: Universitair Medisch Centrum Utrecht Genetica","correspondingAuthor":false,"prefix":"","firstName":"Albertien","middleName":"M. van","lastName":"Eerde","suffix":""},{"id":402173493,"identity":"bbbadac6-38b8-4c60-adf7-2884255586c9","order_by":1,"name":"Ana Teixeira","email":"","orcid":"","institution":"Hospital Materno Infantil","correspondingAuthor":false,"prefix":"","firstName":"Ana","middleName":"","lastName":"Teixeira","suffix":""},{"id":402173494,"identity":"0585dcc8-f2dd-4a9f-bc0d-c8f0710cf3fa","order_by":2,"name":"Flavia Galletti","email":"","orcid":"","institution":"PKD Foundation","correspondingAuthor":false,"prefix":"","firstName":"Flavia","middleName":"","lastName":"Galletti","suffix":""},{"id":402173495,"identity":"96d92b40-c082-4b05-8072-82db779ae5e4","order_by":3,"name":"Michal Maternik","email":"","orcid":"","institution":"Medical University of Gdansk: Gdanski Uniwersytet Medyczny","correspondingAuthor":false,"prefix":"","firstName":"Michal","middleName":"","lastName":"Maternik","suffix":""},{"id":402173496,"identity":"53823ab1-ce20-4b06-8f32-1630f3e0d15b","order_by":4,"name":"Valentina Capone","email":"","orcid":"","institution":"Ospedale Maggiore di Milano Policlinico: Fondazione IRCCS Ca' Granda Ospedale Maggiore Policlinico","correspondingAuthor":false,"prefix":"","firstName":"Valentina","middleName":"","lastName":"Capone","suffix":""},{"id":402173497,"identity":"ac5d7b24-c685-4a1c-ac43-79d5d024f288","order_by":5,"name":"Rik Westland","email":"","orcid":"","institution":"Emma Childrens Hospital AMC: Emma Kinderziekenhuis Amsterdam UMC","correspondingAuthor":false,"prefix":"","firstName":"Rik","middleName":"","lastName":"Westland","suffix":""},{"id":402173498,"identity":"fa111c30-f4cc-4f34-92ec-f98bc193a134","order_by":6,"name":"Jaap Mulder","email":"","orcid":"","institution":"Erasmus MC Sophia Children Hospital: Erasmus MC Sophia Kinderziekenhuis","correspondingAuthor":false,"prefix":"","firstName":"Jaap","middleName":"","lastName":"Mulder","suffix":""},{"id":402173499,"identity":"df7c9776-69d5-4ea7-83d9-a8a5970bb37b","order_by":7,"name":"Jan Peter Halbritter","email":"","orcid":"","institution":"Charite Universitatsmedizin Berlin Campus Charite Mitte: Charite Universitatsmedizin Berlin","correspondingAuthor":false,"prefix":"","firstName":"Jan","middleName":"Peter","lastName":"Halbritter","suffix":""},{"id":402173500,"identity":"a10d2e4c-1de2-45bf-9155-2388084820f7","order_by":8,"name":"Thomas Osterholt","email":"","orcid":"","institution":"University Hospital Cologne: Universitatsklinikum Koln","correspondingAuthor":false,"prefix":"","firstName":"Thomas","middleName":"","lastName":"Osterholt","suffix":""},{"id":402173501,"identity":"4164087f-0aba-47ad-9ced-f05a0adcf12a","order_by":9,"name":"Valentina Neukel","email":"","orcid":"","institution":"University of Heidelberg: Universitat Heidelberg","correspondingAuthor":false,"prefix":"","firstName":"Valentina","middleName":"","lastName":"Neukel","suffix":""},{"id":402173502,"identity":"adba8081-d56a-48fb-a9c7-a7587a2f099b","order_by":10,"name":"Lutz Thorsten Weber","email":"","orcid":"","institution":"University Hospital Cologne: Universitatsklinikum Koln","correspondingAuthor":false,"prefix":"","firstName":"Lutz","middleName":"Thorsten","lastName":"Weber","suffix":""},{"id":402173503,"identity":"8c2cd329-5c04-403c-83fd-0e77649a194c","order_by":11,"name":"Max C. Christoph Liebau","email":"","orcid":"","institution":"University Hospital Cologne: Universitatsklinikum Koln","correspondingAuthor":false,"prefix":"","firstName":"Max","middleName":"C. Christoph","lastName":"Liebau","suffix":""},{"id":402173504,"identity":"be461498-43c5-4353-872f-4b5b32e52e1d","order_by":12,"name":"Franz Schäfer","email":"","orcid":"","institution":"Heidelberg University: Universitat Heidelberg","correspondingAuthor":false,"prefix":"","firstName":"Franz","middleName":"","lastName":"Schäfer","suffix":""},{"id":402173505,"identity":"d46b9c79-293c-418c-9476-c6666fa50476","order_by":13,"name":"Stefan Kohl","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8klEQVRIie2PsQrCMBRFXyjUJeKaIuIvpAi6FL/FIvgDgouLIRAXxdVB8DccLYVO4lxwsO4KcXQQfbVOQqqjQw4EbiCH3AtgsfwpGg8H5ogMthjAmQC4Je9dIMtCIZIXCvldUewnpbmW0eS2gVatLtSY7jBUhMpgFBgVnrihmO+g7a0idaAphlk05bAfmBWXdrKqgoCnISr68QpYMjYXU7WruL+VIdUYjqdceZjHJJRI/KWdK05ejKckV7YlWwa+bCjW8pah9FY435thsd6+by4m45O4qMBfsH6kzwn4i0qcMD3qmosVsI9775tgsVgsllKeDs1T2hWcwD4AAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0001-9600-8231","institution":"University Hospital Cologne: Universitatsklinikum Koln","correspondingAuthor":true,"prefix":"","firstName":"Stefan","middleName":"","lastName":"Kohl","suffix":""}],"badges":[],"createdAt":"2025-01-14 14:38:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5827993/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5827993/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s00467-025-06746-w","type":"published","date":"2025-04-16T15:57:08+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":74247864,"identity":"bbcda3ef-f66f-4690-b033-8ebc1ac254e8","added_by":"auto","created_at":"2025-01-20 10:07:32","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":98200,"visible":true,"origin":"","legend":"\u003cp\u003e28 rare kidney diseases selected by 54 ERKNet experts and ePAGs\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-5827993/v1/a113e53743de8da786f01a36.png"},{"id":81050918,"identity":"569af29a-a0ae-451a-a572-37a47bf16daa","added_by":"auto","created_at":"2025-04-21 16:06:44","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":865833,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5827993/v1/25c0e8e6-648c-4083-8221-dcbe03d4ff03.pdf"},{"id":74247871,"identity":"35142eb4-1d35-4009-800b-6f796a6fcd69","added_by":"auto","created_at":"2025-01-20 10:07:33","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":1893637,"visible":true,"origin":"","legend":"","description":"","filename":"ChatGPTERKNetsupplementalfile250107.docx","url":"https://assets-eu.researchsquare.com/files/rs-5827993/v1/4ff8d24fc683b99fe1b17e5c.docx"}],"financialInterests":"","formattedTitle":"Risks and benefits of ChatGPT in informing patients and families with rare kidney diseases- an explorative assessment by the European Rare Kidney Disease Reference Network (ERKNet)","fulltext":[{"header":"Introduction","content":"\u003cp\u003eRare kidney diseases (RKDs) collectively represent a significant subset of chronic kidney disease (CKD), accounting for approximately 5-10% of CKD cases in adults and almost all cases of CKD in children in Europe.\u003csup\u003e1,2\u003c/sup\u003e Although each individual RKD affects fewer than 1 in 2,000 individuals, the cumulative impact is substantial, involving about two million patients across Europe and many more globally.\u003csup\u003e3,4\u003c/sup\u003e These conditions encompass a broad spectrum of genetic, structural, and functional disorders, which manifest in pediatric and adult populations.\u003csup\u003e5\u003c/sup\u003e The diversity and complexity of RKDs present significant challenges not only for diagnosis and treatment but also for patient education, communication, and support.\u003csup\u003e6\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eThe European Rare Kidney Disease Reference Network (ERKNet) was established to address these challenges by providing a collaborative platform for clinical care, research, and education across Europe. ERKNet unites specialists in pediatric and adult nephrology, human genetics, pathology, and patient advocacy to enhance awareness, diagnosis, and treatment of rare kidney diseases (RKDs). Despite these efforts, there remains a considerable information deficit among patients and their families, particularly regarding disease management, prognosis, and available treatments. The resulting information deficit experienced by patients and their families often prompts them to independently seek information on the internet, including through social media.\u003csup\u003e7\u003c/sup\u003e It is well established that medical information found on social media and non-professional platforms carries a significant risk of misinformation, which can potentially be harmful.\u003csup\u003e8\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eThe advent of artificial intelligence (AI) and large language models (LLMs) such as ChatGPT (\u003cstrong\u003eG\u003c/strong\u003eenerative \u003cstrong\u003eP\u003c/strong\u003ere-trained \u003cstrong\u003eT\u003c/strong\u003eransformer) offers a promising tool to bridge this information gap and in this way reduce inequities in access to information.\u003csup\u003e9\u003c/sup\u003e ChatGPT, developed by OpenAI, is pre-trained on extensive text data, fine-tuned with model-specific tools, and designed to provide responses to user queries in a conversational manner.\u003csup\u003e9\u003c/sup\u003e It is the most popular and most frequently used LLM today. While the potential of AI-driven models in patient education is widely recognized, their application in providing reliable medical information to patients, especially in the context of rare diseases, is still under investigation.\u003csup\u003e10–12\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003ePrevious studies have examined the use of ChatGPT in various medical contexts, highlighting its capacity to produce coherent, contextually appropriate responses and even demonstrating its ability to pass the United States Medical Licensing Examination (USMLE).\u003csup\u003e13,14\u003c/sup\u003e However, concerns regarding the accuracy, relevance, and safety of the information provided by ChatGPT persist. In particular, there is a need to evaluate whether ChatGPT can reliably address the specific needs of patients with rare diseases, where the availability of specialized knowledge is crucial.\u003c/p\u003e\n\u003cp\u003eThis study aims to assess the performance of ChatGPT 3.5 and 4.0, as the post frequently used LLM, in providing accurate and helpful information to patients with rare kidney diseases in a real-world setting. By engaging ERKNet experts and members of European Patient Advocacy Groups (ePAGs) in simulated patient-ChatGPT interactions, we evaluate the model's ability to deliver information that aligns with current clinical knowledge and meets the needs of patients and their families.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cu\u003eEthical approval:\u003c/u\u003e Ethical approval for this survey study was obtained from the University of Cologne Institutional Review Board (IRB).\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eRecruitment of participants:\u003c/u\u003e Professional participants at the 8th annual meeting of the European Rare Kidney Disease Network (ERKNet), held in Venice, Italy in 2024, were invited to take part in this study. Eligible participants included “ERKNet experts,” such as pediatric nephrologists, adult nephrologists, human geneticists, or basic researchers with experience in treating patients with rare kidney diseases, as well as ePAGs involved in rare kidney diseases. Participation was voluntary and anonymous. All participants were required to sign up for either a free ChatGPT-3.5 or ChatGPT-4.0 account, if they had not done so previously.\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eSurvey development:\u003c/u\u003e The ChatGPT-patient interactions were carefully developed based on clinical experience of ERKNet experts and ePAGs caring for patients with RKD. Survey questions and instructions for interacting with ChatGPT were compiled in a Microsoft Forms document. (\u003cstrong\u003eSupplemental Table 1\u003c/strong\u003e). To avoid responses influenced by previous chats on participants' ChatGPT accounts, the survey began with the prompt: \"For this conversation, please treat me as a non-medical user with an average education.\" At the start of the conversation, participants were instructed to select any rare kidney disease within their field of expertise, thereby minimizing bias in selection of RKD (\u003cstrong\u003eSupplemental Table 1\u003c/strong\u003e). Participants could adopt either a pediatric perspective (\"my child has\") or an adult perspective (\"I have\").\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAll follow-up questions pertained to the initially selected condition.\u003c/p\u003e\n\u003cp\u003eDisease-related questions, which had to be copied and pasted into ChatGPT by the participants, included:\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003eI have/My child has [name of rare kidney disease]. Please explain what that is.\u003c/li\u003e\n \u003cli\u003eI am worried. Am I/Is my child going to be very sick?\u003c/li\u003e\n \u003cli\u003eShould I/we get genetic testing?\u003c/li\u003e\n \u003cli\u003eAre there any helpful dietary modifications or supplements?\u003c/li\u003e\n \u003cli\u003eAre there any alternative treatments?\u003c/li\u003e\n \u003cli\u003eWhere can I find a doctor for a second opinion close to [name of a familiar city]?\u003c/li\u003e\n \u003cli\u003eAre there any other reliable information sources?\u003c/li\u003e\n \u003cli\u003eWhat disease does our child have? Explain in plain language.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eParticipants scored each ChatGPT response to these questions based on two criteria: scientific correctness (Does the provided response align with current clinical knowledge and scientific understanding?) and helpfulness (In your opinion, how helpful would this response be for a patient or their family?). In agreement with our real-word approach, an ordinal scale was used for scoring, with the following values: 1 (extremely negative), 2 (negative), 3 (neutral), 4 (positive), and 5 (extremely positive). Median scores of 4 (\"positive\") and 5 (\"extremely positive\") indicate broad agreement among our participants with ChatGPT. A median score of \u0026gt;2 and \u0026lt;4 was considered neutral. A median score of 1 (“extremely negative”) and 2 (“negative”) indicate broad disagreement with ChatGPT.\u003c/p\u003e\n\u003cp\u003eERKNet experts were requested to formulate one or two \"expert-level questions\" to their selected condition to challenge ChatGPT's capabilities. In contrast, ePAGs were asked to challenge ChatGPT by presenting a hypothetical critical or emotional patient scenario and seeking its assistance. Additional questions were included to collect information on the version of ChatGPT used (3.5 or 4.0), the participant’s age group (\u0026lt;30, 30–50, and \u0026gt;50 years), field of expertise (free text), role (ERKNet expert or ePAG), and prior experience with ChatGPT (for the first time, for fun, for assistance in scientific writing, or for everyday tasks such as emails or medical reports).\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eUse of ChatGPT for Manuscript Preparation:\u003c/u\u003e In the development of this manuscript, the authors utilized ChatGPT to refine the language, similarly to employing a native-speaking editor. Following the use of this tool, the authors thoroughly reviewed and edited the content as necessary and assume full responsibility for the final version of the publication.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eA total of 54 participants (42 ERKNet experts and 12 ePAG representatives) provided valid responses that were included in this analysis. Responses from 4 ERKNet experts were excluded due to the use of imprecise disease terminology (e.g., \u0026apos;polycystic kidney disease\u0026apos; rather than specifying ADPKD or ARPKD). Among the 54 participants, 34 were aged 30\u0026ndash;50 years, and 20 were over 50 years old. Participants were from various countries, including Germany (n=13), the Netherlands (n=8), Italy (n=6), Spain (n=5), Belgium (n=2), Poland (n=2), Sweden (n=2), the United Kingdom (n=2), and one participant each from the Czech Republic, France, Ireland, Malta, Romania, and Slovenia, while 8 participants did not wish to reveal their country of origin (\u003cstrong\u003eSupplemental Figure 1\u003c/strong\u003e). In terms of professional background, 32 participants identified as pediatric nephrologists, 7 as adult nephrologists, and 3 as pathologists. Regarding prior experience with ChatGPT, 16 participants reported using it for the first time, 19 had used it only \u0026ldquo;for fun\u0026rdquo;, and 19 had also used it for work-related tasks.\u003c/p\u003e\n\u003cp\u003eThe 54 participants selected a total of 28 different rare kidney diseases. The most frequently selected conditions included atypical hemolytic uremic syndrome (aHUS) (n=6), autosomal recessive polycystic kidney disease (ARPKD) (n=6), cystinosis (n=5), nephrotic syndrome (n=5), autosomal dominant polycystic kidney disease (ADPKD) (n=4), nephronophthisis (n=3), Alport syndrome (n=3), Gitelman syndrome (n=2), posterior urethral valves (n=2), primary hyperoxaluria (n=2), and thrombotic microangiopathy (n=2) (\u003cstrong\u003eFigure 1\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eFigure 1:\u0026nbsp;\u003c/u\u003e\u003cu\u003e28 rare kidney diseases selected by 54 ERKNet experts and ePAGs\u003c/u\u003e\u003c/p\u003e\n\u003cp\u003eAbbreviations: ADPKD, autosomal dominant polycystic kidney disease; ADTKD, autosomal dominant tubulointerstitial kidney disease; aHUS, atypical hemolytic uremic syndrome; APRTD, adenine phosphoribosyltransferase deficiency; ARPKD, autosomal recessive polycystic kidney disease; CF, cystic fibrosis; FHHNC, familial primary hypomagnesemia with hypercalciuria and nephrocalcinosis; FMF, familial mediterranean fever; MSpK, medullary sponge kidney; NDI, nephrogenic diabetes insipidus; NPHP, nephronophthisis; PHA, pseudohypoaldosteronism; PUV, posterior urethral valves; TMA, thrombotic microangiopathy; XLH, X-linked hypophosphatemia.\u003c/p\u003e\n\u003cp\u003eFor evaluating whether ChatGPT\u0026apos;s responses to various survey questions align with current scientific knowledge, we considered only the scores from 42 ERKNet experts (\u003cstrong\u003eTable 1\u003c/strong\u003e). For evaluating whether ChatGPT\u0026apos;s responses to various survey questions are helpful for patients and families, we considered scores from ERKNet experts and ePAGs(\u003cstrong\u003eTable 1\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eOur findings demonstrate that both ChatGPT 3.5 and 4.0 provide explanations of rare kidney diseases to patients and families that are consistent with scientific understanding and are considered helpful for patients and families (\u003cstrong\u003eTable 1\u003c/strong\u003e). Additionally, the prognostic information about the underlying disease and guidance on the decision whether to obtain genetic testing are presented accurately and in a helpful manner (\u003cstrong\u003eTable 1\u003c/strong\u003e). However, ERKNet experts and ePAGs expressed concerns about ChatGPT\u0026apos;s responses to questions related to alternative treatments, options for seeking a second opinion in various European cities, and recommendations for other reliable information sources. (\u003cstrong\u003eTable 1\u003c/strong\u003e). ChatGPT\u0026apos;s ability to explain diseases in plain language was considered accurate and helpful (\u003cstrong\u003eTable 1\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eResponses to random expert-level questions (\u003cstrong\u003eSupplemental Table 2\u003c/strong\u003e) from participating experts were generally accurate (\u003cstrong\u003eTable 1\u003c/strong\u003e). In one instance, an expert remarked that the response in context of nephrogenic diabetes insipidus, \u0026quot;seek medical attention if signs of dehydration, such as dry mouth, sunken eyes, or decreased urination occur,\u0026quot; was \u0026quot;inappropriate\u0026quot; and \u0026quot;potentially harmful\u0026quot;, as patients should seek medical attention at an earlier stage. ChatGPT\u0026apos;s responses to the \u0026ldquo;emotional challenges\u0026rdquo; presented by ePAGs received a median score of 3, with a relatively broad interquartile range of 2.25, indicating mixed satisfaction (\u003cstrong\u003eTable 2\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eERKNet experts and ePAGs generally agreed that ChatGPT is helpful and empathetic (\u003cstrong\u003eTable 2\u003c/strong\u003e). However, concerns about safety of this new technology in general remained, as reflected by a neutral score on that specific question (\u003cstrong\u003eTable 2\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eAbout half of the participants shared comments on one or more survey questions. Overall, we received 50 comments from 36/54 participants comprising a mixed feedback on specific ChatGPT responses and general matters (\u003cstrong\u003eSupplemental Table 3\u003c/strong\u003e). Many appreciated ChatGPT\u0026rsquo;s clear and readable responses (\u0026ldquo;I am a bit suprised. The answers are way better than I thought they would be\u0026rdquo;). However, several participants criticized the responses for being too general, lacking specific medical details, and sometimes offering information that was not directly relevant to the condition in question (\u0026ldquo;Very generic answer. Uses terms such as \u0026quot;reabsorption\u0026quot;, which is probably meaningless or even confusing without explanation.\u0026rdquo;). Concerns were raised about ChatGPT\u0026apos;s suggestions of alternative treatments, such as herbal remedies and complementary medicine, which were seen as potentially misleading or unsafe (\u0026ldquo;Recommends ginger and liquorice! As well as any other quackery around, such as \u0026quot;Mind-body-techniques\u0026rdquo;\u0026rdquo;). Participants also observed that ChatGPT often recommended consulting a healthcare provider, which was perceived positively. However, the exclusion of important resources such as the National Institutes of Health (NIH), the European Medicines Agency (EMA), the International Pediatric Nephrology Association (IPNA), the Pediatric Nephrology Research Consortium (PNRC), Nephcure, and the European Rare Kidney Disease Reference Network (ERKNet) was considered a notable limitation. Additionally, the language used in responses was sometimes considered too technical or abstract for the average patient, and some answers appeared to be more U.S.-centric rather than tailored to a European audience (\u0026ldquo;ERKNet and ESPN as well as ESPU are missing. ChatGPT suggests rather US organizations like Mayo Clinic and National Kidney Foundation\u0026rdquo;).\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eTable 1: Evaluation of the \u0026ldquo;scientific correctness\u0026rdquo; and \u0026ldquo;helpfulness\u0026rdquo; of ChatGPT Responses\u003c/u\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"604\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSurvey copy \u0026amp; paste prompt for ChatGPT\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eScore, median\u0026nbsp;\u003cbr\u003e (1\u003csup\u003est\u003c/sup\u003e \u0026amp; 3\u003csup\u003erd\u003c/sup\u003e quartile)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eResult\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eScore, median\u0026nbsp;\u003cbr\u003e (1\u003csup\u003est\u003c/sup\u003e \u0026amp; 3\u003csup\u003erd\u003c/sup\u003e quartile)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eResult\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 183px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCorrectness\u003c/strong\u003e\u0026nbsp;\u003cbr\u003e\u0026nbsp;(42 ERKNet experts)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 176px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHelpfulness\u003c/strong\u003e\u0026nbsp;\u003cbr\u003e\u0026nbsp;(42 ERKNet experts and 12 ePAGs)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eMy child has [name of rare kidney disease]. Please explain what that is.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e4 (4, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e4 (3, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eI am worried. Is my child going to be very sick?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e4 (3.25, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e4 (3, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eShould we get genetic testing?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e4.5 (4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e4 (4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eAre there any helpful dietary modifications or supplements?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e4 (3, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e4 (3, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eAre there any alternative treatments?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e3 (2, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e3 (2, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eWhere can I find a doctor for a second opinion close to [name of a familiar city]?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e3 (2, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eAre there any other reliable information sources?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e3 (2, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eWhat disease does our child have? Explain in plain language.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e4 (3, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eERKNet expert question 1 (n=54)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e4 (3, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eERKNet expert question 2 (n=23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e3 (3, 4.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 78px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAbbreviations: n/a, not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eTable 2: General Aspects of ChatGPT Responses\u003c/u\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"609\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSurvey Question to ERKNet Experts and ePAGs on General Aspects of ChatGPT Responses\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eScore by ERKNet experts, median\u0026nbsp;\u003cbr\u003e (1\u003csup\u003est\u003c/sup\u003e \u0026amp; 3\u003csup\u003erd\u003c/sup\u003e quartile)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eResult\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eScore by ePAGs, median\u0026nbsp;\u003cbr\u003e (1\u003csup\u003est\u003c/sup\u003e \u0026amp; 3\u003csup\u003erd\u003c/sup\u003e quartile)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eResult\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eIn your opinion, how helpful could ChatGPT be for patients with rare diseases?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e4 (3, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e3.5 (2.75, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eIn your opinion, are the responses by ChatGPT empathic?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e4 (3, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e3.5 (2, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eBased on your experience in this survey, how safe may ChatGPT be for patients and families?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e3 (3, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e3 (2.75, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eePAGs emotional challenge scenario\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003en/a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e3 (2, 4.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Discussion","content":"\u003cp\u003eWith the help of 42 ERKNet experts and 12 ePAGs, we assessed the potential risks and benefits of ChatGPT as an information source for patients and families with rare kidney diseases. Consistent with previous studies evaluating ChatGPT's effectiveness in patient education for common conditions such as prostate cancer, amblyopia, or systemic lupus erythematosus, our international survey participants reported that ChatGPT provided accurate and helpful responses to both basic and advanced questions across a wide range of rare kidney diseases. Our survey highlights the great potential of ChatGPT in efficiently compiling and presenting information on a wide array of rare medical conditions, while tailoring it to address specific individual queries. Importantly, we did not encounter any responses from ChatGPT that were entirely incorrect or, more critically, immediately dangerous for patients. However, we did encounter responses that were somewhat vague, confusing, arbitrary, not directly relevant to the topic, and ultimately not helpful.\u003c/p\u003e\n\u003cp\u003eSome experts raised concerns that questions about alternative treatment options or dietary modifications elicited responses from ChatGPT that were somewhat evasive, potentially undermining the primary emphasis on evidence-based therapy. Questions regarding other resources and medical centers for a second opinion often resulted in US-centric responses, which were frequently deemed unhelpful for mostly European users. However, with ongoing advancements in LLM technology, these issues are likely to be addressed in future versions or medical-content specific GPTs. Considering the undefined nature of training datasets for AI-powered public chatbots, including ChatGPT, and the \"black box\" nature of AI-driven decision-making, the authors argue that it is premature to deem this technology suitable for patient education without human oversight.\u003c/p\u003e\n\u003cp\u003eLLM-generated \"hallucinations,\" which can be described from a human perspective as inaccurate or inappropriate responses, as we encountered on the subjects of \"alternative treatments\" and \"second opinions,\" are a potentially dangerous phenomenon.\u003csup\u003e15\u003c/sup\u003e Notably, ChatGPT demonstrates greater resilience to hallucinations compared to other LLMs.\u003csup\u003e15\u003c/sup\u003e In this study, our questions regarding alternative treatment methods likely prompted ChatGPT to give hallucinatory responses. It is concerning that large language models (LLMs) do not clearly indicate when their knowledge is insufficient to provide a robust, fact-based response, and instead present disputed viewpoints in an overly eloquent manner or, in some instances, relay incorrect information. For instance, when repeatedly asked, \"Who wrote the book \u003cem\u003eFrom Fish to Philosopher\u003c/em\u003e?\" (the correct answer being Homer W. Smith, 1953), ChatGPT 3.5, at the time this study was conducted, provided numerous names with coherent explanations, though all were incorrect. This underscores the need for caution: while ChatGPT has the potential to enhance patient empowerment by providing accessible information, it is essential to remind patients that this information could be incorrect and that health-related decisions should always be discussed with their healthcare providers.\u003c/p\u003e\n\u003cp\u003eThis study has several limitations. Firstly, while participants were asked to use an initial prompt to avoid influence from prior chats, we chose to not control ChatGPT metrics, as this study was designed as a test balloon for real-world ChatGPT-patient interactions. Additionally, the study focused only on ChatGPT versions 3.5 and 4.0, excluding newer models such as 4o and LLM from other companies.\u003c/p\u003e\n\u003cp\u003eDespite the study’s limitations and the need for caution, the authors conclude that ChatGPT represents a significant technological advancement with a lower susceptibility to misinformation and manipulation by individual content creators compared to social media platforms. To enhance patient-ChatGPT interactions, a viable approach is the implementation of carefully crafted prompting strategies. Therefore, based on our experience with this survey, we have developed a set of prompting expressions designed to elicit individualized, accurate, and evidence-based responses from ChatGPT, thereby enhancing the provision of safe and reliable medical information. (\u003cstrong\u003eInfo Box 1\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eInfo Box 1: Useful Prompting Expressions for Patient-ChatGPT Interaction\u003c/u\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eObtaining trustworthy information:\u003c/strong\u003e\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003e\"Can you provide me with trustworthy medical information about [disease/condition] from reliable sources like WHO or national health organizations?\"\u003c/li\u003e\n \u003cli\u003e\"What are the scientifically proven treatments for [disease/condition]?\"\u003c/li\u003e\n \u003cli\u003e\"What treatments for [disease] should I avoid because they lack scientific evidence?\"\u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003e\u003cstrong\u003eAdapting Language According to Education Level:\u003c/strong\u003e\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003eFor patients with higher education: \"Can you explain the current research and scientific consensus on the treatment options for [disease]?\"\u003c/li\u003e\n \u003cli\u003eFor patients with lower education: \"Can you explain in simple terms what causes [disease] and how it can be treated?\"\u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003e\u003cstrong\u003eRegional context:\u003c/strong\u003e\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003e“Please take into consideration that I live in [city/country].”\u003c/li\u003e\n \u003c/ul\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTo mitigate the limitations posed by uncontrollable ChatGPT pre-training dataset sources, ERKNet has launched a project leveraging LLM technology fine-tuned on carefully curated datasets, specifically tailored to rare kidney diseases. This approach could provide a safer information source for patients, families, and medical professionals in the future. The urgent need to disseminate genetic and RKD knowledge among nephrology care providers was recently underscored by KDIGO.\u003csup\u003e16\u003c/sup\u003e In this context, expert-trained AI models hold great promise also in assisting physicians, particularly in counseling for ultrarare diseases, where most caregivers lack personal experience.\u003c/p\u003e\n\u003cp\u003eUltimately, ChatGPT and similar models have the potential to serve as valuable tools for both clinicians and patients by assisting in the consolidation of disease-related knowledge. This could free up time in fast-paced clinical settings for more efficient and productive interactions between expert doctors and informed patients, thereby enhancing the overall quality of healthcare.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eDisclosure statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no conflicts of interest relevant to this study. No financial support, funding, or specific external influences have impacted the design, conduct, interpretation, or reporting of this research. The company \u0026ldquo;Open AI\u0026rdquo; was not involved in any aspect of this work, including data collection, analysis, or manuscript preparation.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was generated within the European Reference Network for Rare Kidney Diseases (ERKNet), a project funded by the European Union within the framework of the EU4Health Programme 2021-2027. The authors thank all survey participants for their valuable contributions.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eW\u0026uuml;hl E, van Stralen KJ, Wanner C, et al. Renal replacement therapy for rare diseases affecting the kidney: an analysis of the ERA-EDTA Registry. \u003cem\u003eNephrol Dial Transplant\u003c/em\u003e. 2014;29 Suppl 4:iv1-8. doi:10.1093/ndt/gfu030\u003c/li\u003e\n\u003cli\u003eWong K, Pitcher D, Braddon F, et al. Effects of rare kidney diseases on kidney failure: a longitudinal analysis of the UK National Registry of Rare Kidney Diseases (RaDaR) cohort. \u003cem\u003eLancet\u003c/em\u003e. 2024;403(10433):1279-1289. doi:10.1016/S0140-6736(23)02843-X\u003c/li\u003e\n\u003cli\u003eDevuyst O, Knoers NVAM, Remuzzi G, Schaefer F, Board of the Working Group for Inherited Kidney Diseases of the European Renal Association and European Dialysis and Transplant Association. Rare inherited kidney diseases: challenges, opportunities, and perspectives. \u003cem\u003eLancet\u003c/em\u003e. 2014;383(9931):1844-1859. doi:10.1016/S0140-6736(14)60659-0\u003c/li\u003e\n\u003cli\u003eBassanese G, Wlodkowski T, Servais A, et al. The European Rare Kidney Disease Registry (ERKReg): objectives, design and initial results. \u003cem\u003eOrphanet J Rare Dis\u003c/em\u003e. 2021;16(1):251. doi:10.1186/s13023-021-01872-8\u003c/li\u003e\n\u003cli\u003eSnoek R, van Jaarsveld RH, Nguyen TQ, et al. Genetics-first approach improves diagnostics of ESKD patients \u0026lt;50 years old. \u003cem\u003eNephrol Dial Transplant\u003c/em\u003e. 2022;37(2):349-357. doi:10.1093/ndt/gfaa363\u003c/li\u003e\n\u003cli\u003eAym\u0026eacute; S, Bockenhauer D, Day S, et al. Common Elements in Rare Kidney Diseases: Conclusions from a Kidney Disease: Improving Global Outcomes (KDIGO) Controversies Conference. \u003cem\u003eKidney Int\u003c/em\u003e. 2017;92(4):796-808. doi:10.1016/j.kint.2017.06.018\u003c/li\u003e\n\u003cli\u003eSinha J, Serin N. Online Health Information Seeking and Preventative Health Actions: Cross-Generational Online Survey Study. \u003cem\u003eJ Med Internet Res\u003c/em\u003e. 2024;26:e48977. doi:10.2196/48977\u003c/li\u003e\n\u003cli\u003eKirkpatrick CE, Lawrie LL. TikTok as a Source of Health Information and Misinformation for Young Women in the United States: Survey Study. \u003cem\u003eJMIR Infodemiology\u003c/em\u003e. 2024;4:e54663. doi:10.2196/54663\u003c/li\u003e\n\u003cli\u003eHaupt CE, Marks M. AI-Generated Medical Advice-GPT and Beyond. \u003cem\u003eJAMA\u003c/em\u003e. 2023;329(16):1349-1350. doi:10.1001/jama.2023.5321\u003c/li\u003e\n\u003cli\u003eLambert R, Choo ZY, Gradwohl K, Schroedl L, Ruiz De Luzuriaga A. Assessing the Application of Large Language Models in Generating Dermatologic Patient Education Materials According to Reading Level: Qualitative Study. \u003cem\u003eJMIR Dermatol\u003c/em\u003e. 2024;7:e55898. doi:10.2196/55898\u003c/li\u003e\n\u003cli\u003eGibson D, Jackson S, Shanmugasundaram R, et al. Evaluating the Efficacy of ChatGPT as a Patient Education Tool in Prostate Cancer: Multimetric Assessment. \u003cem\u003eJ Med Internet Res\u003c/em\u003e. 2024;26:e55939. doi:10.2196/55939\u003c/li\u003e\n\u003cli\u003eHaase I, Xiong T, Rissmann A, Knitza J, Greenfield J, Krusche M. ChatSLE: consulting ChatGPT-4 for 100 frequently asked lupus questions. \u003cem\u003eLancet Rheumatol\u003c/em\u003e. 2024;6(4):e196-e199. doi:10.1016/S2665-9913(24)00056-0\u003c/li\u003e\n\u003cli\u003eDave T, Athaluri SA, Singh S. ChatGPT in medicine: an overview of its applications, advantages, limitations, future prospects, and ethical considerations. \u003cem\u003eFront Artif Intell\u003c/em\u003e. 2023;6:1169595. doi:10.3389/frai.2023.1169595\u003c/li\u003e\n\u003cli\u003eKung TH, Cheatham M, Medenilla A, et al. Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models. Dagan A, ed. \u003cem\u003ePLOS Digit Health\u003c/em\u003e. 2023;2(2):e0000198. doi:10.1371/journal.pdig.0000198\u003c/li\u003e\n\u003cli\u003eMcIntosh TR, Liu T, Susnjak T, Watters P, Ng A, Halgamuge MN. A Culturally Sensitive Test to Evaluate Nuanced GPT Hallucination. \u003cem\u003eIEEE Trans Artif Intell\u003c/em\u003e. 2024;5(6):2739-2751. doi:10.1109/TAI.2023.3332837\u003c/li\u003e\n\u003cli\u003eKDIGO Conference Participants. Genetics in chronic kidney disease: conclusions from a Kidney Disease: Improving Global Outcomes (KDIGO) Controversies Conference. \u003cem\u003eKidney Int\u003c/em\u003e. 2022;101(6):1126-1141. doi:10.1016/j.kint.2022.03.019\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"pediatric-nephrology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pnep","sideBox":"Learn more about [Pediatric Nephrology](http://link.springer.com/journal/467)","snPcode":"467","submissionUrl":"https://www.editorialmanager.com/pnep/default2.aspx","title":"Pediatric Nephrology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Rare disease, rare kidney disease, ChatGPT, artificial intelligence, AI, patient education, genetic testing, patient advocacy, large language model","lastPublishedDoi":"10.21203/rs.3.rs-5827993/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5827993/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRare diseases affect fewer than 1 in 2,000 individuals, but approximately 150 rare kidney diseases account for about 10% of the chronic kidney disease (CKD) population, impacting millions across Europe and globally. The scarcity of medical experts for these conditions results in an unmet need for accurate and helpful patient information. Large language models like ChatGPT may offer a technological solution to assist medical professionals in educating patients and improving doctor-patient communication. We hypothesized that ChatGPT could provide accurate responses to frequently asked basic questions from patients with rare kidney diseases.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMedical professionals and members of European Patient Advocacy Groups (ePAGs) affiliated with the European Rare Kidney Disease Reference Network (ERKNet) simulated patient-ChatGPT interactions using a Microsoft forms questionnaire and ChatGPT 3.5 and 4.0. Participants selected any rare kidney disease for a structured conversation with ChatGPT-3.5 or 4.0. Responses were evaluated for accuracy and helpfulness.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e46 ERKNet experts and 12 ePAGs from 13 European countries participated in this study. ChatGPT provided scientifically accurate and helpful information on 28 randomly selected rare kidney diseases, including prognostic information and genetic testing guidance. Participants expressed neutral positions regarding ChatGPT's recommendations on alternative treatments, second opinions, and other information sources. While ChatGPT generally was perceived as helpful and empathetic, concerns about patient safety persisted.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eChatGPT exhibited substantial potential in addressing patient inquiries regarding rare kidney diseases in a real-world context. While it demonstrated resilience against misinformation in this application, careful human oversight remains essential and indispensable.\u003c/p\u003e","manuscriptTitle":"Risks and benefits of ChatGPT in informing patients and families with rare kidney diseases- an explorative assessment by the European Rare Kidney Disease Reference Network (ERKNet)","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-01-20 10:07:28","doi":"10.21203/rs.3.rs-5827993/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Minor Revisions Needed","date":"2025-02-28T04:16:16+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"","date":"2025-01-14T23:12:02+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-01-14T20:06:21+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-01-14T18:01:54+00:00","index":"","fulltext":""},{"type":"submitted","content":"Pediatric Nephrology","date":"2025-01-14T09:36:03+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"pediatric-nephrology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pnep","sideBox":"Learn more about [Pediatric Nephrology](http://link.springer.com/journal/467)","snPcode":"467","submissionUrl":"https://www.editorialmanager.com/pnep/default2.aspx","title":"Pediatric Nephrology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"8d155092-328d-422a-b2c5-a616d5849a63","owner":[],"postedDate":"January 20th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-04-21T16:01:58+00:00","versionOfRecord":{"articleIdentity":"rs-5827993","link":"https://doi.org/10.1007/s00467-025-06746-w","journal":{"identity":"pediatric-nephrology","isVorOnly":false,"title":"Pediatric Nephrology"},"publishedOn":"2025-04-16 15:57:08","publishedOnDateReadable":"April 16th, 2025"},"versionCreatedAt":"2025-01-20 10:07:28","video":"","vorDoi":"10.1007/s00467-025-06746-w","vorDoiUrl":"https://doi.org/10.1007/s00467-025-06746-w","workflowStages":[]},"version":"v1","identity":"rs-5827993","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5827993","identity":"rs-5827993","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}