Shiva Shankar Bugude, Mohan Sasankoti
Periodontology, Qassim University, Buraydah, SAU.
Oral Medicine and Radiology, Hope Health Inc, Florence, USA.
Cureus. 2025 Feb 24;17(2):e79556. doi: 10.7759/cureus.79556. eCollection 2025 Feb.
Background Evidence-based decision-making (EBDM) is essential in contemporary dentistry. However, navigating the extensive and constantly evolving scientific literature can be challenging. Large language models (LLMs), such as ChatGPT-4, have the potential to transform EBDM by analyzing vast datasets and extracting critical information, thereby significantly reducing the time required to find evidence. This observational feasibility study investigates ChatGPT-4's potential in dental EBDM, focusing on its capabilities, strengths, and limitations. Materials and methods In this observational feasibility study, two independent examiners conducted interactive sessions with ChatGPT-4. Five clinical scenarios were explored using the Google Chrome web browser, accessing publicly available scientific evidence from Cochrane, ADA, and PubMed. This approach ensured compliance with the Cochrane guidelines for EBDM. Two independent dentists engaged with ChatGPT-4 in simulated real-life clinical scenarios to seek scientific information. The output from ChatGPT-4 for each scenario was assessed based on predetermined criteria. Its responses were evaluated for accuracy, relevance, efficiency, actionability, and ethical considerations using the ChatGPT-4 Response Scoring System (CRSS) and the ChatGPT-4 Generative Ability Matrix (C-GAM). Results ChatGPT-4 demonstrated consistent performance across all five clinical scenarios, achieving a C-GAM score of 46.4% and a CRSS score of 12 out of 28. It effectively identified relevant sources of evidence and provided concise summaries, potentially saving valuable time and enhancing access to information. No significant differences in scores were found when the responses to all clinical scenarios were analyzed independently by the two researchers. However, a notable limitation was its inability to provide specific web links directing users to relevant scientific articles. Additionally, while ChatGPT-4 offered suggestions for incorporating the latest scientific publications into decision-making, it could not generate direct links to these articles. Conclusion Despite its current limitations, ChatGPT-4, as a generative AI, can assist clinicians in making evidence-based decisions. It can save time compared to conventional search engines. Ethical considerations must be prioritized in training these models to ensure that clinicians make responsible, evidence-based decisions rather than relying solely on specific evidence statements provided by ChatGPT-4. This model shows its potential as an AI tool for EBDM in dentistry. Further development and training could address existing limitations and enhance its effectiveness; however, clinicians must retain ultimate responsibility for informed decisions, necessitating expertise and critical evaluation of the evidence presented.
背景 循证决策（EBDM）在当代牙科领域至关重要。然而，在浩如烟海且不断发展的科学文献中查找信息可能具有挑战性。诸如ChatGPT-4之类的大型语言模型（LLMs）有潜力通过分析海量数据集并提取关键信息来改变循证决策，从而显著减少查找证据所需的时间。这项观察性可行性研究调查了ChatGPT-4在牙科循证决策中的潜力，重点关注其能力、优势和局限性。
材料与方法 在这项观察性可行性研究中，两名独立的审查员与ChatGPT-4进行了交互会话。使用谷歌浏览器探索了五个临床场景，从考科蓝图书馆（Cochrane）、美国牙科协会（ADA）和美国国立医学图书馆（PubMed）获取公开可用的科学证据。这种方法确保符合考科蓝循证决策指南。两名独立的牙医在模拟的现实临床场景中与ChatGPT-4互动，以寻求科学信息。根据预先确定的标准评估ChatGPT-4对每个场景的输出。使用ChatGPT-4响应评分系统（CRSS）和ChatGPT-4生成能力矩阵（C-GAM）评估其回答的准确性、相关性、效率、可操作性和伦理考量。
结果 ChatGPT-4在所有五个临床场景中表现一致，C-GAM得分为46.4%，CRSS得分为28分中的12分。它有效地识别了相关证据来源并提供了简洁的总结，可能节省宝贵的时间并增加信息获取途径。两位研究人员独立分析对所有临床场景的回答时，得分没有显著差异。然而，一个明显的局限性是它无法提供指向相关科学文章的具体网页链接。此外，虽然ChatGPT-4提供了将最新科学出版物纳入决策的建议，但它无法生成这些文章的直接链接。
结论 尽管ChatGPT-4目前存在局限性，但作为一种生成式人工智能，它可以帮助临床医生做出循证决策。与传统搜索引擎相比，它可以节省时间。在训练这些模型时，必须优先考虑伦理考量，以确保临床医生做出负责任的循证决策，而不是仅仅依赖ChatGPT-4提供的特定证据陈述。该模型显示出其作为牙科循证决策人工智能工具的潜力。进一步的开发和训练可以解决现有局限性并提高其有效性；然而，临床医生必须对明智的决策承担最终责任，这需要专业知识和对所提供证据的批判性评估。
