The application of large language models in orthopedic postgraduate education: potentials, challenges, and future prospects

Abstract

With the widespread integration of artificial intelligence (AI), orthopedics postgraduate education is transitioning into the intelligent era. Large language models (LLMs), which leverage deep learning and natural language processing (NLP), have profoundly influenced orthopedic postgraduate education through their sophisticated capabilities in coherent comprehension, contextual response and content generation. These models encompass both general-purpose tools (e.g., ChatGPT) and specialized orthopedic applications (e.g., DocOA, BioinspiredLLM, MechGPT, DrSR, and AmbossGPT). They can offer interdisciplinary research training, targeted academic guidance, real-time resource access, interactive case exercise and immersive simulation practice in orthopedics. Most models exhibited promising performance: for instance, ChatGPT-4 achieved a 61.2% accuracy on the Orthopedic In-Training Examination (OITE) comparable to orthopedic residents, while DocOA significantly outperformed ChatGPT-4 with more than 142% improvement in orthopedic benchmark evaluations. The LLMs’ capabilities could promote a personalized, interactive, and adaptive transformation of orthopedic postgraduate education. However, the application of LLMs faces significant challenges such as over-reliance, delayed updates and inconsistent outputs, sparking ongoing controversy in the medical education. Moving forward, establishing a comprehensive human-AI collaborative framework is imperative to optimize the application of LLMs in orthopedic postgraduate education. This holistic framework integrates learner-centered perspectives, multidimensional governance, phased implementation strategies, and geographic diversity. Together, adopting this innovative human-AI approach will strengthen the cultivation of high-level orthopedic talents for orthopedic postgraduate education.

Introduction

Orthopedics is undergoing a profound transformation toward the era of intelligent healthcare, driven by the expanding adoption of artificial intelligence (AI) [1–3]. This evolution is fundamentally reshaping traditional orthopedic postgraduate education, shifting toward comprehensive cultivation of medical students’ comprehensive capabilities [3–5]. Specifically, AI is fostering interdisciplinary knowledge integration, redefining disciplinary boundaries, and accelerating cross-disciplinary convergence [6, 7]. Moreover, AI technologies are revolutionizing the generation and dissemination of medical knowledge, updating the ways in knowledge acquirement, and transforming pedagogical models [8, 9]. For example, more than 60% of students expressed interest in AI-based education application, approximately 50% intended to incorporate AI into their future teaching practices, and over 40% indicated a willingness to recommend AI tools to their colleagues [10].

Large language models (LLMs), characterized with deep learning and natural language processing (NLP) technologies, are pretrained on vast amounts of data with architectures containing billions of parameters [5, 11, 12]. As one of the most significant breakthroughs for AI, LLMs are capable of participating in conversational exchanges with users while generating rapid, accurate, and contextual responses [5, 11, 13]. The most prevalent LLMs, Generative Pre-trained Transformer (ChatGPT), without further training or reinforcement, have exhibited considerable medical knowledge and reasoning abilities, achieving commendable performance in several medical licensing examinations [14–17]. More advanced medical-specific LLMs, such as Google’s Med-PaLM 2, have often been preferred over physicians’ answers for scientific accuracy, reasoning, and precision in pairwise evaluations [18]. These capabilities of LLMs provide innovative solutions to promote the development of orthopedic postgraduate education. Integrating LLMs into orthopedic postgraduate education, can enable multi-disciplinary data analysis, and offer personalized ability guidance and simulate interactive clinical practice [3, 19, 20]. Therefore, LLMs will propel orthopedic postgraduate education toward a more personalized, interactive, and adaptive paradigm.

Despite the significant potential of LLMs in orthopedic postgraduate education, concerns related to reliability and security require careful attention in the medical context [21–23]. A prominent limitation is that LLMs generate responses based on probabilistic predictions, which may generate inaccurate information and thereby undermine outputs’ reliability [24, 25]. On the other hand, LLMs can impede the development of critical thinking, and even cause a series of ethical and legal issues, which necessitates stringent oversight in the future application [26, 27].

However, there is currently a lack of systematic review on the application of LLMs in orthopedic postgraduate education. This review aims to comprehensively explore the potentials, challenges, and future prospects of LLMs application in orthopedic postgraduate education, for proposing a human-AI collaborative education framework and innovative cultivation of orthopedic talents in the era of intelligent healthcare.

The potentials of LLMs application in orthopedic postgraduate education

A summary of the application potentials of LLMs in orthopedic postgraduate education is presented in Fig. 1. These applications can be categorized into five major domains: training of interdisciplinary research collaboration, provision of targeted academic guidance, facilitation of real-time access to diverse educational resources, engagement in interactive case exercises, and implementation of immersive simulation practices. Collectively, these domains delineate a systematic framework through which LLMs may foster a more personalized, interactive, and adaptive paradigm of orthopedic postgraduate education.

Fig. 1.

The potential of LLMs application in orthopedic postgraduate education

Interdisciplinary collaboration research

Orthopaedics is at a historic juncture of disciplinary integration, which is essential pathway for future development [28]. A central strand of this interdisciplinary collaboration research is the development of medical biomaterials, such as bone-repair materials and devices [29, 30]. However, orthopedic postgraduates’ innovative research on medical biomaterials encounters challenges due to the cross-disciplinary convergence and complex material properties [31, 32]. LLMs provide the best chance to break these barriers by their abilities to rapidly process vast information, extract contextual knowledge, and generate predictive responses. Specifically, LLMs can empower orthopedic postgraduates by intelligently filtering parameters, simulating biomimetic microstructures, and performing biomechanical predictions, which substantially enhances the research efficiency in biomaterials. For example, in designing medical biomaterials for bone scaffolds, implants, and prosthetics, Luu and Buehler’s BioinspiredLLMS could accurately retrieve critical parameters such as titanium alloy porosity, elasticity modulus, and degradation rate, and generate predictions for the optimal combination based on compatibility [33]. In addition, Lang2Sim platforms, extended LLMs capabilities by simulating biomaterial properties under specific conditions and thereby evaluating the applicability in real-world practice [34]. Another multiscale LLMs, MechGPT, from Buehler’s laboratory, could effectively accelerate the construction and validation of knee joint biomechanical models through high-dimensional dynamic simulations, including damping and oscillatory equations. Moreover, DrSR framework, built on the further integration of LLMs, had markedly improved the efficiency and interpretability of biomechanical modeling [35]. It can achieve the derivation of damping oscillation equations approximating real physical laws within only 1000 iterations (error < 10⁻⁵) [35]. By converting interdisciplinary knowledge across clinical medicine, materials science, and biomechanics into comprehensive learning tools, LLMs are reshaping the research paradigm and driving innovation in orthopedic postgraduate education.

Targeted learning support

Orthopedic postgraduate education requires high standards in the cultivation of both clinical practice competence and research-oriented thinking [36]. LLMs meet this need by serving as readily accessible, delivering targeted support in complex clinical cases analysis, multimodal imaging interpretation, and academic manuscript preparation. Actually, ChatGPT-4 has demonstrated knowledge performance comparable to orthopedic residents for about 61.2% accuracy on the Orthopedic In-Training Examination (OITE), which highlights the potentials of LLMs for education support [37]. The specialized orthopedic LLMs, DocOA, integrated Retrieval-Augmented Generation (RAG) technology and interfaced with osteoarthritis (OA) guidelines, could further outperform ChatGPT-4 significantly with at least 142% improvement in OA benchmark evaluations [38]. For image-based learning, AmbossGPT, fine-tuned on the Amboss medical library, excelled in fracture subtype recognition, achieving 100% accuracy in spinal fractures compared with 79.2% of ChatGPT-4o [39].

In addition, LLMs have proven beneficial for academic research and lesson preparation. A recent study revealed that implementing ChatGPT into academic writing led to notable increases in both AI and manual scoring [40]. About 92% of students acknowledged advancements in their writing quality and 76% recognized the contribution of ChatGPT [40]. Another study found that orthopedic fellows spent an average of 94 min creating joint arthroplasty-related teaching materials, whereas ChatGPT generated better orthopedic contents in just 16 s, which demonstrated the substantial efficiency advantages of AI-driven methods over traditional teacher-centered education [41]. Through effective Human-AI collaboration, LLMs are poised to build more intelligent and personalized learning environment for orthopedic postgraduates beyond traditional boundaries.

Overall, LLMs’ performance varies by technical enhancements and specific adaptations. As ChatGPT-4 establishes a solid baseline for general tasks, RAG technology enables general-purpose models to integrate specialized knowledge, reaching near-expert accuracy in literature retrieval and academic tasks [42]. In addition, domain-specific fine-tuning yields superior results for specialized tasks: compared to ChatGPT-4, DocOA excelled in osteoarthritis benchmarks linked to OA guidelines, and AmbossGPT, trained on the AMBOSS medical library, achieved higher accuracy in spinal fracture recognition [38, 39]. This suggests that while general-purpose LLMs handle broad queries, RAG-enhanced architectures and specialized training are essential for complex medical contexts and precise clinical scenarios.

Efficient knowledge access

The rapid development of orthopedic knowledge requires us to promptly acquire the latest clinical guidelines, literature materials, and surgical techniques [43]. LLMs can offer immediate access to up-to-date information and generate relevant learning materials, which markedly enhances the efficiency of resource retrieval. For example, a previous study, involving 100 real-world cases of common orthopedic conditions, assessed the concordance between ChatGPT-4-generated treatment recommendations with both American Academy of Orthopaedic Surgeons (AAOS) clinical practice guidelines (CPGs) and the treatment plans of orthopedic attending physicians [44]. It found a high degree of alignment with AAOS CPG in 90% of cases and with physicians’ plans in 78%, suggesting the potentials of LLMs to serve as reliable sources of best practice for most standardized orthopedic clinical scenarios [44]. When combined with RAG technology, ChatGPT-4 + RAG further outperformed ChatGPT-4 (73.80% vs. 64.91%) in answering 1,023 questions from the AAOS OrthoWizard database, nearly matching the 73.97% accuracy of orthopedic professionals [42]. The credibility of ChatGPT-4 + RAG in orthopedic postgraduate education is enhanced by the ability to trace answers to their source materials [42]. With efficient accessing to mostly authoritative material sources, LLMs can optimizing time allocation for orthopedic postgraduates and enable greater focus on knowledge application rather than information collection.

Interactive case practice

The cultivation of clinical competence is a core task in medical education and medical practice [45]. One of the core challenges in orthopedic postgraduate education is how to provide realistic and interactive clinical case training within a controlled environment. While standardized patients (SPs) provide valuable opportunities to simulate clinical scenarios, limitations have persistently remained due to substantial cost and resource constraints [46, 47]. The interactive virtual SPs, powered by LLMs’ dynamic generation, present an innovative augmentation to traditional SPs [48]. The virtual SPs can simulate the full spectrum of orthopedic conditions with visualized anatomical structure and dynamic adjustments, which enables orthopedic postgraduates to practice decision-making in a more realistic, controlled and progressive environment. One randomized controlled trial had found that virtual SPs could significantly improve clinical decision-making scores compared to traditional SPs [49]. Additionally, LLMs-driven simulations had been shown to alleviate clinical performance anxiety and provide positive feedback for medical students, increasing their willingness to engage in similar practices [50]. Furthermore, LLMs can deliver higher-quality assessment than traditional SPs. For instance, the error rate of ChatGPT was significantly lower than that of traditional SPs (1.0% vs. 7.2%) in scoring medical students’ notes, which helped students accurately identify their competency gaps [51]. LLMs-driven interactive case practice has approximated traditional SPs in quality and even surpassed them in terms of realism and education feedback, allowing orthopedic postgraduates to continuously develop their clinical skills in a relaxed and flexible environment.

Immersive simulation assistance

As orthopedics relies heavily on precise procedures, a critical task is learning complex surgical techniques during postgraduate stage [52]. Virtual Reality (VR) and Augmented Reality (AR) have proven effective in orthopedic surgical training, and LLMs can further enrich immersive simulations with personalization, interactivity and adaptability [53–57]. For example, Osso VR, which utilizes LLMs to deliver simulation and quantization across various surgical procedures, offers trajectory analysis, personalized feedback, and objective assessments [57]. A study had further shown that medical students trained with Osso VR performed significantly better in intramedullary nail fixation, with 130% higher total skill scores, 38% more correct steps and 39% more knowledge mastery compared to traditional training [58]. Although the integration of LLMs with VR/AR is still in exploration stage, the intelligent transformation of orthopedic postgraduate surgical training represents an emerging trend in the future.

The challenges of LLMs application in orthopedic postgraduate education

The promising potentials of LLMs in orthopedic postgraduate education has been demonstrated, however, several challenges occur in the application (Table 1). First, prolonged immersion in LLMs-driven mechanical extraction and virtual simulations may lead to over-reliance [59]. Through reducing face-to-face interactions and reflective analysis in traditional education, LLMs could hinder the development of critical thinking and decision-making abilities, which is essential for medical practice. Another challenge is the issue of delayed updates in training data and insufficient regional adaptability [11]. Given the rapid iteration in orthopedic technologies, like medical biomaterials and devices, LLMs must ensure timely synchronization with the latest advancements. Furthermore, LLMs inherently predict the most likely continuation of given inputs based on training data. This “black-box” algorithm may result in “hallucination”, a phenomenon where fabricating content is generated without supporting evidence [24]. This phenomenon may even pose potential medical risks in orthopedics. Most importantly, the exceptional convenience of LLMs may facilitate academic misconduct, thereby sparking a new wave of academic integrity crises [60]. Orthopedic postgraduates may exploit the LLMs’ auto-generation capabilities as academic cheating tools. In response to these challenges, an effective governance framework for LLMs in orthopedic postgraduate education must be built to ensure compliant application while mitigating potentials risks.

Table 1.

The challenges of LLMs application in orthopedic postgraduate education

Challenges	Description	Evidence
Over-reliance and diminished critical thinking	Excessive immersion in LLMs may diminish critical thinking and clinical practice.	1. A study of 370 medical students in India found that while 67.4% exhibit openness toward incorporating LLMs in medical education, over 53% expressed concerns that overreliance on LLMs could hinder the development of clinical reasoning skills [61]. 2. Compared to writing with no external tools, students using ChatGPT showed lower activation in brain regions associated with memory, critical thinking and creativity in EEG [62]. 3. A study of 666 participants in the United Kingdom showed that the frequent use of AI tool negatively correlated with critical thinking abilities, with younger participants (ages 17–25) exhibiting higher dependence on AI tools and lower critical thinking scores [63].
Insufficient timeliness	LLM-generated knowledge is relatively limited due to outdated training data and regional differences.	1. GPT-4 was trained on data up to March 2023, whereas later versions of ChatGPT incorporate more recent training updates [11]. 2. While ChatGPT could pass most exams such as the USML, it had failed to pass the Chinese medical licensure assessments, partly considering the model’s lack of training on local syllabus and terminology [64].
Hallucination	LLMs’ generation lacks transparency and may result in misleading information.	1. GPT-3.5 exhibited a hallucination rate of 39.6% when retrieving references based on the same inclusion criteria as original systematic reviews, whereas the more advanced GPT-4 showed a hallucination rate of 28.6% with only 9–14% correct references [65]. 2. The high reliance on precision and accuracy in orthopedic procedures makes hallucinations particularly prominent in areas such as hip, knee and hand joints, requiring LLMs trained specifically on substantial musculoskeletal data [66].
Academic misconduct	Powerful generation capability makes LLMs prone to misuse for plagiarism, ghostwriting, and cheating.	1. A study of 1,133 medical students in China revealed that while 65.4% were concerned about the risk of ChatGPT potentially facilitating plagiarism, 62.9% still used this technology in their medical research, and 16.5% incorporated the technology into their published articles [67]. 2. The reviewers with at least five years of experience could correctly identify only 62% of AI-generated abstracts, predominantly due to unrealistic data [68]. 3. A Science editorial emphasized that LLMs should not be held accountable as authors of scientific studies and LLM-generated content without adequate citations may be considered plagiarism [69].

AI Artificial intelligence, EEG Electroencephalography, GPT Generative Pre-trained Transformer, LLMs Large Language Models, USML United States Medical Licensing Examination

The future prospects of LLMs application in orthopedic postgraduate education

To address the multifaceted challenges and fully maximize potentials of LLMs in orthopedic postgraduate education, it is essential to establish a comprehensive human-AI collaborative framework.

Learner-centered perspectives: A learner-centered perspective is essential for constructing this governance framework, as postgraduate students are core participants in medical education and their attitudes toward LLMs play a crucial role in integration [70, 71]. A survey of 586 medical students indicated a strong willingness to embrace AI, with approximately 60% expressing interest in AI-based education and nearly half intending to incorporate it into future teaching practices [10]. However, this enthusiasm was tempered by prevalent concerns, with more than 80% expressing worries about accuracy of AI-generated information [10]. Similarly, an Indian survey of 370 medical students found that while 67.4% of students were open to LLMs’ incorporation, over 53% feared that over-reliance might erode their clinical reasoning skills [61]. Ethical and academic integrity is another prominent concern: although 62.9% of students reportedly utilized tools like ChatGPT for research, a significant majority remained apprehensive about the risks of spreading misinformation (76.9%), plagiarism and academic misconduct (65.4%) and falsifying research (63.2%) [67]. Collectively, while many participants recognized the potential of LLMs and expressed enthusiasm for their use, a lack of prescriptive guidance and regulations contributed to their hesitancy in actual application [70, 72].

Multidimensional governance framework: A multidimensional framework integrating technological, pedagogical, and ethical-legal dimensions remains essential for the application of LLMs in orthopedic postgraduate education. Technologically, implementation prioritizes the integration of RAG with orthopedic data repositories to develop specialized LLMs with high clinical precision [73]. This process entails rigorous protocols to safeguard patient privacy and institutional intellectual property during model execution. Pedagogically, the framework emphasizes integrating AI literacy into curricula through practice-oriented simulations, such as AI-assisted preoperative planning [74]. These formats facilitate a transition from passive knowledge acquisition to active critical appraisal of AI-generated outputs. Centrally, a multidisciplinary AI ethics committee is required, comprising senior clinicians, medical ethicists, data scientists, and student representatives. This body evaluates usage contexts based on patient privacy, algorithmic transparency, and educational equity, with reviews triggered by new tool adoption or adverse event reporting [75]. Such oversight aligns with China’s ethical mandates and the EU’s Artificial Intelligence Act, ensuring traceability and reinforcing shared responsibility between developers and users [76, 77].

Phased implementation roadmap: To transition from theory to practice, we propose a three-phase roadmap for integrating LLMs into orthopedic postgraduate education: (1) Phase 1 (Year 1): Focus centers on establishing an AI ethics committee to define usage boundaries and privacy protocols. Simultaneously, pilot programs and faculty training will be launched to build institutional trust and mentor-led responsible adoption. (2) Phase 2 (Year 2–3): LLMs will be formally integrated into orthopedic curricula via personalized instruction and intelligent learning systems. Assessment frameworks will shift toward critical appraisal, ensuring students can evaluate AI outputs while clarifying the technology’s practical value. (3) Phase 3 (Year 4+): LLMs will be fully implemented across all clinical rotations and research training. A dynamic feedback loop will be established to facilitate continuous evaluation, ensuring that educational contents and technical protocols undergo iterative refinement to remain aligned with rapid AI advancements and evolving clinical standards [78].

However, successful implementation of this roadmap requires realistic institutional planning regarding cost considerations and resource implications [78]. Administrators need prioritize funding for initial licensing and infrastructure costs, alongside ongoing system maintenance and updating. In addition, sustained investment in faculty development resources is essential to support effective mentorship in the intelligent era.

Geographic diversity: Global integration of LLMs requires addressing substantial disparities across diverse healthcare systems [79]. To ensure inclusivity, particularly in resource-limited settings, LLMs adoption aligns with the concept of “appropriate technology” [80]. This approach prioritizes cost-effective, accessible tools tailored to local infrastructural capacities, rather than relying solely on resource-intensive models. For instance, lightweight, mobile-based LLMs applications can democratize access to high-quality orthopedic knowledge, bridging the educational gap between major medical centers and developing regions [81]. Furthermore, the establishment of open educational resources tailored to local clinical contexts and data diversity promotes equitable learning opportunities. By fostering course designs that respect regional needs, institutions facilitate a balanced and inclusive evolution of AI-driven medical education globally [82].

Ultimately, LLMs are best regarded as assistants rather than leaders in medical practice. Through this comprehensive human-AI collaborative framework, orthopedic postgraduates can develop the competencies to responsibly maximize the positive impact of LLMs, which fosters the higher-quality and more sustainable progress in medical education.

Conclusion

The LLMs are changing orthopedic postgraduate education in unprecedented ways. The integration of LLMs drives personalization, interactivity, and adaptability in education, enabling them to serve as comprehensive, intelligent assistants across research training and clinical practice. It will not only help address the transformation challenges in the era of intelligent healthcare, but also provide new momentum for sustained innovation and high-quality development. However, the application of LLMs also introduces a series of challenges, particularly concerning reliability and security. It is crucial to further establish a comprehensive human-AI collaborative framework encompassing learner-centered perspectives, multidimensional governance, phased implementation strategies, and geographic diversity, which regulates the application boundaries and thereby maximizes the beneficial impacts in medical contexts. In the future, orthopedic postgraduate education will be characterized by human-AI collaboration and complementary strengths, contributing to the cultivation of high-level, multidisciplinary professionals.

Abbreviations

AAOS

American Academy of Orthopaedic Surgeons

AI

Artificial intelligence

AR

Augmented Reality

CPGs

Clinical practice guidelines

EEG

Electroencephalography

GPT

Generative Pre-trained Transformer

LLMs

Large language models

NLP

Natural language processing

OITE

Orthopedic In-Training Examination

RAG

Retrieval-Augmented Generation

SPs

Standardized patients

USML

United States Medical Licensing Examination

VR

Virtual Reality

Author contributions

CZ, YW, JW, and GL conceived and designed the study. All authors have read, provided critical feedback on intellectual content, and approved the final manuscript. KR and QW designed and coded figures and tables. KR, QW, and QC drafted the manuscript.

Funding

This work was supported by the Degree and Postgraduate Education Reform Research Project of Hunan province (2024JGZD011), Degree and Postgraduate Education Reform Research Project of Central South University (2024JGA009, 2025JC016) and by grant from The Scientific Research Program of FuRong Laboratory (JW: 2023SK2100, GL: 2024PT5108).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.