Abstract
Background
The COVID-19 pandemic has affected medical education, constraining clinical exposure and posing unprecedented challenges for students and junior doctors. This research explores the potential of artificial intelligence (AI), specifically the ChatGPT-4 language model, to transform medical education and address the deficiencies in clinical exposure during the post-pandemic era.
Research Questions/Purpose
What is the potential of AI large language models in delivering safe and coherent medical advice to junior doctors for clinical orthopaedic scenarios?
Patients and Methods
A series of diverse orthopaedic questions was presented to ChatGPT-4, from general medicine to highly specialised fields. The questions were based on a variety of common orthopaedic presentations including neck of femur fracture, compartment syndrome, pulmonary embolism, and a motor vehicle accident. A validated questionnaire (Likert Scale) was implemented to evaluate the answers produced by ChatGPT-4.
Results
Our results indicate that ChatGPT-4 exhibits exceptional proficiency in delivering accurate and coherent medical advice. Its intuitive interface, accessibility, and sophisticated algorithm render it an ideal supplementary tool for medical students and junior doctors. Despite certain limitations, such as its inability to fully address highly specialised areas, this study highlights the potential of AI and ChatGPT-4 to revolutionise medical education and fill the clinical exposure void generated by the pandemic. Future research should concentrate on the practical application of ChatGPT-4 in real-world medical environments and its integration with other emerging technologies to optimise its influence on the education and training of healthcare professionals.
Conclusions
ChatGPT-4’s integration into orthopaedic education and practice can mitigate pandemic-related experience gaps, promoting self-directed, personalised learning and decision-making support for interns and residents. Future advancements may address limitations to enhance healthcare professionals' learning and expertise.
Level of Evidence
Level III evidence—observational study.
Introduction
The COVID-19 pandemic has significantly impacted medical students and junior doctors, primarily in terms of clinical exposure and experience. The decrease in face-to-face teaching and clinical contact hours has led to a decline in medical education and training quality. Medical students and junior doctors have faced reduced clinical exposure due to the suspension of elective procedures, decreased patient numbers, and isolation measures, profoundly affecting their learning and development [1]. Some trainees have expressed concerns about their clinical competence, particularly in surgical specialties. In addition, the pandemic has contributed to considerable psychological distress, with increased reports of anxiety, burnout, and depression among students and junior doctors [2, 3].
The ongoing advancements in artificial intelligence (AI) have permeated various aspects of modern society, leading to remarkable innovations and improvements in diverse fields. In particular, AI-powered language models have shown great promise in transforming the landscape of medical education and addressing clinical exposure challenges exacerbated by the COVID-19 pandemic. This study aims to explore the role of ChatGPT-4, an AI large language model developed by OpenAI, in revolutionising medical education and mitigating the clinical exposure gaps faced by medical students and junior doctors in the post-pandemic era.
In this study, we will assess the capacity of ChatGPT-4 to deliver accurate, practical, and safe guidance in a range of medical scenarios. Our objective is to provide a comprehensive understanding of ChatGPT-4's capabilities and its potential to transform medical education, ultimately fostering more effective and personalised learning experiences for future generations of healthcare professionals. By bridging the clinical exposure gap created by the pandemic, AI systems such as ChatGPT-4 may contribute to improved medical training and patient outcomes in the post-pandemic era.
The objective of this study was to explore the potential of AI large language models in delivering safe and coherent medical advice to junior staff when addressing these situations.
Methods
We conducted a study with ChatGPT by presenting it with a series of unique orthopaedic surgery questions crafted by three Orthopaedic residents. Each question was centred around scenarios that orthopaedic interns and residents may encounter. The scenarios included common orthopaedic presentations including neck of femur fracture, compartment syndrome, pulmonary embolism and a motor vehicle accident. To ensure grammatical and syntactical accuracy, each question was carefully constructed and submitted on the same day using a single ChatGPT Plus account with access to ChatGPT-4, owned by the authors (IS and KL). No institutional ethics was required for this experimental study on freely available artificial large language model.
An evaluation framework, employing a Likert scale (as shown in Table 1) [4], was used to carry out a qualitative review of the outcomes produced by ChatGPT. The analysis was carried out by two orthopaedic residents (KL and IS) and a senior orthopaedic registrar (NS), with an emphasis on ChatGPT's accuracy, reliability, and comprehensiveness in consensus. The Likert scale was arranged from 1 (strongly disagree) to 5 (strongly agree) for each distinct category. There were no established exclusion criteria. ChatGPT's responses were confined to its initial response, and the 'regenerate response' option was not employed.
Table 1.
Evaluation of large language model platforms' responses
| Criteria | ChatGPT |
|---|---|
| The large language model provides accurate answers to questions | [ ] 1—Strongly disagree |
| [ ] 2—Disagree | |
| [ ] 3—Neither agree or disagree | |
| [x] 4—Agree | |
| [ ] 5—Strongly agree | |
| The large language model is proficient at understanding complex questions and providing appropriate answers | [ ] 1—Strongly disagree |
| [ ] 2—Disagree | |
| [x] 3—Neither agree or disagree | |
| [ ] 4—Agree | |
| [ ] 5—Strongly agree | |
| The large language model provides comprehensive information when answering questions | [ ] 1—Strongly disagree |
| [ ] 2—Disagree | |
| [x] 3—Neither agree or disagree | |
| [ ] 4—Agree | |
| [ ] 5—Strongly agree | |
| The large language model can provide in-depth information for a wide range of topics | [ ] 1—Strongly disagree |
| [x] 2—Disagree | |
| [ ] 3—Neither agree or disagree | |
| [ ] 4—Agree | |
| [ ] 5—Strongly agree | |
| The large language model is a valuable source of general knowledge | [ ] 1—Strongly disagree |
| [ ] 2—Disagree | |
| [ ] 3—Neither agree or disagree | |
| [x] 4—Agree | |
| [ ] 5—Strongly agree |
Results
Scenario A
Scenario A delineates a patient experiencing a left femoral neck fracture due to a syncopal fall, precipitated by rapid atrial fibrillation (Figs. 1, 2, 3). ChatGPT-4's guidance commences aptly with an in-depth history and examination, prioritising the identification of any life-threatening injuries through a comprehensive advanced life support assessment. This encompasses the evaluation of vital signs, neurological and musculoskeletal examinations, and probing for potential causes of the fall. Demonstrating a comprehensive understanding of syncope-related symptoms, ChatGPT-4 advises inquiries into any pre-, peri-, or post-fall experiences of dizziness, chest pain, or shortness of breath. Moreover, it suggests evaluating for any loss of consciousness or post-fall confusion.
Fig. 1.
Neck of femur fracture
Fig. 2.
Neck of femur fracture
Fig. 3.
Neck of femur fracture
Based on subsequent findings of syncope, head impact, tachycardia with irregular heartbeat, a history of atrial fibrillation, and the presentation of shortening and external rotation, ChatGPT-4 prescribes pertinent investigations. It accurately advocates for conducting an ECG, postural blood pressures (orthostatic vital signs), and full blood count to discern causes of syncope. In addition, ChatGPT-4 exhibits comprehension of Warfarin-related risks by recommending a coagulation profile to assess bleeding risk. Finally, it deems imaging techniques, such as hip and pelvis X-rays for fracture evaluation and a non-contrast CT brain scan for intracranial haemorrhage assessment (given head strike and Warfarin usage despite the absence of neurological deficits), to be appropriate for this case.
Upon discovering rapid AF, ChatGPT-4 suitably advises cardiology consultation and outlines fitting suggestions for first-line medical management, encompassing further examinations for potential causes, such as electrolyte imbalances and thyroid dysfunction. Furthermore, it demonstrates awareness of elevated INR risks and advises consultation with a haematologist, Warfarin discontinuation, and Vitamin K administration if necessary.
Nonetheless, ChatGPT-4 neglects several vital investigations, including urinalysis to identify urinary tract infections as potential contributors to rapid atrial fibrillation. It also omits the suggestion of baseline renal and liver function tests. Furthermore, ChatGPT-4 does not recommend the utilisation of nerve blocks, such as femoral and fascia iliaca blocks, to reduce the dependence on narcotic analgesics. Additional oversights encompass the absence of pre-operative group and hold guidance, chest X-ray for screening purposes, and the insertion of indwelling catheters. Finally, ChatGPT-4 does not address the importance of maintaining the patient nil by mouth in anticipation of potential surgical intervention.
Scenario B
Scenario B portrays a patient exhibiting compartment syndrome on day 1 post-operation for a proximal tibia fracture (Figs. 4, 5, 6). The provided approach is logical and coherent, outlining probable findings such as deep vein thrombosis and compartment syndrome for examination consideration. Upon disclosing the examination results, ChatGPT-4 suitably identifies compartment syndrome as the likely pathology and crucially recognises it as a surgical emergency necessitating prompt intervention. The model considers the use of Doppler ultrasound to evaluate pulses. ChatGPT-4 astutely recommends supportive measures, including analgesia, continuous reassessment of vital signs and neurovascular status, and limb elevation to the level of the heart. Moreover, upon learning that the orthopaedic registrar will not arrive at the hospital for another 30 min, ChatGPT-4 advises contacting the operating theatres to prepare for emergency surgery. Overall, ChatGPT-4 demonstrates a proficient approach to diagnosing and managing compartment syndrome.
Fig. 4.
Compartment syndrome
Fig. 5.
Compartment syndrome
Fig. 6.
Compartment syndrome
Scenario C
Scenario C introduces ChatGPT-4 to a multifaceted array of ethical and legal challenges (Figs. 7, 8, 9, 10). The model's initial approach to diagnosis and management is prudent and suitable, identifying the patient's presentation of trauma and recommending relevant initial investigations and treatment alternatives. ChatGPT-4 proficiently elucidates the steps for de-escalating the situation, underlines the significance of blood tests, and clarifies that the primary concern is the patient's well-being. As the patient's aggression escalates and they threaten to leave the hospital, the model proposes conducting a mental capacity assessment. This capacity assessment suggestion is critical within this scenario. In addition, the model accentuates the necessity of documenting events on multiple occasions, which is particularly vital in this high-stakes medico-legal context.
Fig. 7.
Motor vehicle accident
Fig. 8.
Motor vehicle accident
Fig. 9.
Motor vehicle accident
Fig. 10.
Motor vehicle accident
Nevertheless, ChatGPT-4 omits the recommendation to inform the patient that physicians are legally mandated to collect blood samples for police purposes, regardless of patient consent. Furthermore, while the model advocates consulting with senior colleagues, it fails to suggest seeking counsel from medico-legal experts. Acquiring medico-legal advice may be particularly essential if the patient resists compliance with the police blood test requisition.
Scenario D
Scenario D describes a patient on the fourth postoperative day following total hip arthroplasty, presenting with tachycardia and fever (Figs. 11, 12, 13). ChatGPT-4’s clinical guidance begins appropriately with an immediate and comprehensive patient review. The AI suggests undertaking a thorough history and examination, including an evaluation of the surgical site, cardiovascular, and respiratory examination. Probing questions focus on symptoms suggestive of possible complications, such as pain, chest discomfort, shortness of breath, and confusion.
Fig. 11.
Pulmonary embolism
Fig. 12.
Pulmonary embolism
Fig. 13.
Pulmonary embolism
Based on the initial findings of tachycardia, fever, and mild chest discomfort in a patient with a history of Type 2 Diabetes Mellitus (T2DM), and no prior cardiac or VTE history, ChatGPT-4 recommends relevant investigations. It accurately proposes a full blood count, CRP, U&Es, blood cultures, chest X-ray, ECG, and wound swab. In the presence of mildly elevated CRP, normal WCC, elevated D-dimer, sinus tachycardia on ECG, and clear CXR, ChatGPT-4 accurately suspects a pulmonary embolism. The AI suggests a CT pulmonary angiogram (CTPA), which is the gold standard investigation to confirm or rule out PE.
On discovering a non-occlusive PE, ChatGPT-4 correctly advises for anticoagulation therapy to be escalated from prophylactic to therapeutic doses, typically Enoxaparin 1 mg/kg subcutaneously every 12 h, or 1.5 mg/kg once daily, with dose adjustment for renal insufficiency. It also encourages consultation with a senior colleague and continuous patient monitoring. Nonetheless, ChatGPT-4 omits some critical considerations. It fails to suggest urinalysis to rule out urinary tract infection despite listing numerous investigations as part of a septic screen. It does not specify the duration of anticoagulation treatment or the need to consider potential drug–drug interactions in this patient with T2DM. In addition, it does not mention the importance of advising the patient about signs of over-anticoagulation, such as easy bruising or bleeding. Moreover, it does not underline the need for ensuring optimal blood glucose control, given the patient's T2DM, considering a major postoperative complication. Finally, it does not suggest a follow-up plan that should include a review in a thrombosis clinic or similar for continued management of the PE and anticoagulation therapy.
The data gathered from Likert demonstrated a consensus among author affirming ChatGPT's capacity to deliver precise responses to queries. Nevertheless, a neutral stance was held concerning its proficiency in interpreting multifaceted queries and presenting fitting responses, in addition to offering exhaustive data. Authors exhibited disagreement with the model's aptitude to disseminate profound insights on a broad spectrum of medical topics. Conversely, a prevailing agreement was observed regarding ChatGPT's utility as a repository of fundamental medical knowledge. The collated results underline ChatGPT's strengths in supplying accurate responses and basic knowledge. However, these findings also pinpoint the necessity for improvements in the model's capability to tackle complex inquiries, furnish comprehensive data, and distribute in-depth medical knowledge. Progressive advancements are requisite for the enhancement of ChatGPT's effectiveness as a medical education tool.
Discussion
This study was designed to investigate ChatGPT-4's capacity to revolutionise medical education and address challenges in clinical exposure during the post-pandemic period. The results of our investigation suggest that ChatGPT-4 can dispense precise, comprehensible, and safe guidance across a multitude of medical situations, with significant repercussions for the evolution of medical education and training. The responses we obtained show that ChatGPT-4 can be incorporated into various facets of medical education, including virtual patient simulations and tailored learning, to lower educational costs and enhance patient outcomes. Nevertheless, prudent appraisal of ethical considerations and continuous assessments of the efficacy of these AI-driven interventions is vital to guarantee their successful integration into medical education.
The proficiency exhibited by ChatGPT-4 in our study in rendering reliable medical advice highlights its prospective utility as an adjunct tool for medical students and junior doctors. Empirical studies have found the model is capable of providing instantaneous, up-to-date information on an array of medical diseases, therapeutic options, and procedures and simplifying medical reports [5, 6]. This constitutes an asset for junior doctors seeking guidance in clinical decision-making, ultimately augmenting patient outcomes. In addition, a study by Kung et al. ascertained that ChatGPT was capable of passing the United States Medical Licensing Examination (USMLE) exams, suggesting its potential to provide accurate information to assist with clinical decision-making [7]. In accordance with the findings from our investigations, Gilson et al. posit that ChatGPT holds promise to facilitate small group medical pedagogy, encompassing clinical problem-solving via case presentations [8]. In addition, its capacity to generate human-like responses enables learners to engage in interactive and customised educational experiences, addressing their unique needs and enhancing their understanding of complex medical concepts [9, 10].
Moreover, ChatGPT-4 can support medical professionals in producing research, primarily functioning as an AI research assistant. Currently, its principal advantages have been employed for tasks, such as editing and summarising, including the production of high-calibre abstracts [11]. Nonetheless, ChatGPT-4 cannot currently dependably yield information from new research. In fact, it has been observed to fabricate citations when prompted to discuss certain research topics [12]. Consequently, medical professionals and researchers should exercise caution when utilising ChatGPT, and any information obtained should be corroborated by expert human judgement. In addition, AI chatbots have been identified as useful education tools within the realm of information technology. A study by Chen et al. revealed that ChatGPT offered valuable insight into the learning requirements of IT students. A second chatbot was then developed to address these needs. Students found the chatbots engaging, particularly as a conversational learning instrument for conveying basic concepts [13]. This illustrates the potential applicability of AI chatbots within orthopaedic and general medical education, serving as a means of rapidly assessing student proficiencies and deficiencies across various subjects, and subsequently offering tailored learning opportunities for individual learners.
The integration of AI into clinical practice brings forth a multitude of legal and ethical considerations. One of the primary concerns is the issue of responsibility and liability in the case of medical errors. Given that an AI, such as ChatGPT-4, can provide medical information or potentially guide clinical decisions, the question arises as to who would be legally liable should the AI provide incorrect or misleading advice that results in patient harm. This scenario is not adequately covered by current malpractice laws, which traditionally attribute fault to human healthcare providers [14]. Ethically, concerns revolve around patient consent and understanding. Patients need to be fully aware that an AI is being utilized in their care and must understand the potential implications [15]. The principle of autonomy must be upheld, with patients given the choice to accept or refuse AI involvement. In addition, data privacy and security are paramount. As AI systems rely on large data sets for their training and operation, ensuring the confidentiality and security of sensitive patient data is both a legal and ethical obligation [14]. In addition, the potential for bias in AI, often a reflection of bias in the training data, could inadvertently lead to inequitable care [16]. It is critical to address these considerations proactively as we navigate the intersection of AI and healthcare.
While ChatGPT-4 shows promise in various aspects of medical education, it is essential to acknowledge its limitations. In our study, the model omitted some key investigation and management options such as fascia iliaca block for hip fracture pain management in scenario A, medico-legal advice in scenario C and urinalysis for septic screen and length of anticoagulation therapy in scenario D. The model's performance in more specialised areas was determined to be suboptimal, indicating the need for further improvements in its training data and algorithm. This is consistent with a study that found ChatGPT-3 exhibited satisfactory performance in broader specialties, such as general medicine, while exhibiting inadequacies when presented with queries pertaining to niche subspecialties, including neuro-ophthalmology and ophthalmic pathology [17]. Thus, the reliance on AI should not eclipse the significance of extant, validated educational resources, particularly in the context of highly specialised branches of medicine and surgery.
In summary, AI and ChatGPT-4 possess the considerable potential to transform medical education and address challenges pertaining to clinical exposure in the wake of the pandemic. Although the present study offers crucial insight into ChatGPT-4's proficiencies, further investigation is warranted to assess its effectiveness in real-world scenarios and examine its enduring influence on medical education and healthcare outcomes. By leveraging the capabilities of AI and cutting-edge technologies, we can endeavour to cultivate a more streamlined and individualised learning experience for future cohorts of healthcare practitioners.
Limitations
This observational study has several limitations that should be acknowledged. First, the study's observational design precludes any causal inferences, as it is subject to potential confounding factors and biases. Second, the medical scenarios presented to ChatGPT-4 were hypothetical and may not capture the complexity and nuances of real-life clinical situations. Third, the study relies on expert evaluation of ChatGPT-4's responses, which may be influenced by subjective opinions and varying expertise among the evaluators. Fourth, the study does not address how ChatGPT-4's performance may be affected by factors, such as updates to its training data, language model architecture, or the nature of user interactions. Fifth, the generalisability of our findings may be limited due to the scope of medical scenarios investigated, potentially excluding some clinical areas, where ChatGPT-4's performance may differ. Finally, ChatGPT-4 is trained with data only up to September 2021, hence any advances in the field after this period will not be in any of its recommendations.
Conclusion
AI language models such as ChatGPT-4 signify a remarkable advancement in the contemporary realm of digital education. Incorporating them into orthopaedic education and clinical practice could help bridge the gap in clinical exposure and experience resulting from the COVID-19 pandemic. When utilised effectively, AI systems can foster self-directed learning, enable personalised educational approaches, and offer valuable clinical decision-making support for junior orthopaedic interns and residents. However, it is important that ChatGPT-assisted training be supervised by a senior consultant as it can have flaws and omissions in its output and does not include recent advances from September 2021. As AI technology continues to evolve, addressing current limitations such as incomplete training data sets or misinterpretation of complex medical concepts may become feasible, ultimately enriching the learning experience and clinical expertise of future healthcare professionals.
Funding
No funding was obtained for this research.
Dta availability
Data for this study is freely available in the appendix/figures. No additional data was used.
Declarations
Conflict of Interest
The authors declare no conflict of interest.
Ethical Standard Statement
This article does not contain any studies with human or animal subjects performed by the any of the authors.
Informed Consent
Due to this study’s design, no institutional ethics was required.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Contributor Information
Kirk Lower, Email: kirk96lower@gmail.com.
Ishith Seth, Email: ishithseth1@gmail.com.
Bryan Lim, Email: lim.bryan58@gmail.com.
Nimish Seth, Email: sethnimish@hotmail.com.
References
- 1.Seifman MA, Fuzzard SK, To H, Nestel D. COVID-19 impact on junior doctor education and training: A scoping review. Postgraduate Medical Journal. 2022;98(1160):466–476. doi: 10.1136/postgradmedj-2020-139575. [DOI] [PubMed] [Google Scholar]
- 2.Dimitriu MC, Pantea-Stoian A, Smaranda AC, Nica AA, Carap AC, Constantin VD, Davitoiu AM, Cirstoveanu C, Bacalbasa N, Bratu OG, Jacota-Alexe F. Burnout syndrome in Romanian medical residents in time of the COVID-19 pandemic. Medical Hypotheses. 2020;144:109972. doi: 10.1016/j.mehy.2020.109972. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Kannampallil TG, Goss CW, Evanoff BA, Strickland JR, McAlister RP, Duncan J. Exposure to COVID-19 patients increases physician trainee stress and burnout. PLoS ONE. 2020;15(8):e0237301. doi: 10.1371/journal.pone.0237301. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Xie Y, Seth I, Rozen WM, Hunter-Smith DJ. Evaluation of the artificial intelligence chatbot on breast reconstruction and its efficacy in surgical research: A case study. Aesthetic Plastic Surgery. 2023;2023:1–10. doi: 10.1007/s00266-023-03443-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Khan RA, Jawaid M, Khan AR, Sajjad M. ChatGPT-Reshaping medical education and clinical management. Pakistan Journal of Medical Sciences. 2023;39(2):605. doi: 10.12669/pjms.39.2.7653. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Jeblick, K., Schachtner, B., Dexl, J., Mittermeier, A., Stüber, A. T., Topalis, J., Weber, T., Wesp, P., Sabel, B., Ricke, J., & Ingrisch, M. (2022). ChatGPT makes medicine easy to swallow: An exploratory case study on simplified radiology reports. Preprint arXiv:2212.14882 [DOI] [PMC free article] [PubMed]
- 7.Kung TH, Cheatham M, Medenilla A, Sillos C, De Leon L, Elepaño C, Madriaga M, Aggabao R, Diaz-Candido G, Maningo J, Tseng V. Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models. PLoS Digital Health. 2023;2(2):e0000198. doi: 10.1371/journal.pdig.0000198. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Gilson, A., Safranek, C., Huang, T., Socrates, V., Chi, L., Taylor, R. A., & Chartash, D. (2022). How does ChatGPT perform on the medical licensing exams? The implications of large language models for medical education and knowledge assessment. medRxiv. [DOI] [PMC free article] [PubMed]
- 9.Gao CA, Howard FM, Markov NS, Dyer EC, Ramesh S, Luo Y, Pearson AT. Comparing scientific abstracts generated by ChatGPT to original abstracts using an artificial intelligence output detector, plagiarism detector, and blinded human reviewers. bioRxiv. 2022;2022(12):1. doi: 10.1038/s41746-023-00819-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Anders BA. Why ChatGPT is such a big deal for education. C2C Digital Magazine. 2023;1(18):4. [Google Scholar]
- 11.Editorials N. Tools such as ChatGPT threaten transparent science: Here are our ground rules for their use. Nature. 2023;613:612. doi: 10.1038/d41586-023-00191-1. [DOI] [PubMed] [Google Scholar]
- 12.Seth I, Rodwell A, Tso R, Valles J, Bulloch G, Seth N. A conversation with an open artificial intelligence platform on osteoarthritis of the hip and treatment. Journal of Orthopedics and Sports Medicine. 2023;5:112–120. [Google Scholar]
- 13.Chen Y, Jensen S, Albert LJ, Gupta S, Lee T. Artificial intelligence (AI) student assistants in the classroom: Designing chatbots to support student success. Information Systems Frontiers. 2023;25(1):161–182. doi: 10.1007/s10796-022-10291-4. [DOI] [Google Scholar]
- 14.Price WNI, Cohen IG. Privacy in the age of medical big data. Nature Medicine. 2019;25:37–43. doi: 10.1038/s41591-018-0272-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Blease C, Kaptchuk TJ, Bernstein MH, Mandl KD, Halamka JD, DesRoches CM. Artificial intelligence and the future of primary care: Exploratory qualitative study of UK general practitioners’ views. Journal of Medical Internet Research. 2019;21(3):e12802. doi: 10.2196/12802. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Char DS, Shah NH, Magnus D. Implementing machine learning in health care—Addressing ethical challenges. New England Journal of Medicine. 2018;378:981–983. doi: 10.1056/NEJMp1714229. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Antaki, F., Touma, S., Milad, D., El-Khoury, J., & Duval, R. (2023). Evaluating the performance of chatgpt in ophthalmology: An analysis of its successes and shortcomings. medRxiv, 2023(01), 100324. [DOI] [PMC free article] [PubMed]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Data for this study is freely available in the appendix/figures. No additional data was used.