Jun Hye Won, Ryu Sun Young, Yoo Tae Keun
Department of Ophthalmology, Hangil Eye Hospital, Incheon, Republic of Korea.
Department of Refractive Surgery, B&VIIT Eye Center, Seoul, Republic of Korea.
J Ophthalmol. 2025 Jul 15;2025:9930116. doi: 10.1155/joph/9930116. eCollection 2025.
To evaluate the feasibility of using ChatGPT-4, a large language model (LLM), to develop a customized nomogram calculator for small-incision lenticule extraction (SMILE) surgery based on institution-specific data, without requiring any coding expertise. Customized nomograms are essential due to variations in surgical practices, patient populations, and diagnostic equipment across vision correction centers. A retrospective analysis of consecutive patients was performed on data of 1268 eyes that underwent SMILE. Preoperative measurements and postoperative refractive errors at 6 months were collected and analyzed. The entire dataset was divided into a training set and validation set at a ratio of 3:1. After data anonymization, ChatGPT-4 was instructed to perform a linear regression analysis to predict postoperative refractive errors using preoperative data. Subsequently, we instructed ChatGPT-4 to generate HTML code for a webpage-based nomogram calculator that inputs preoperative data and calculates surgical parameters using the derived formulas. The results of the regression analysis performed using ChatGPT-4 were compared with those obtained using two conventional statistical software programs, R and SPSS. ChatGPT-4 successfully performed SMILE nomogram regression analysis. The predicted SMILE parameters were not significantly different from those obtained using the statistical software. The nomogram showed a higher predictive ability for postoperative refractive error than the simple empirical nomogram ( < 0.001). We successfully created a webpage-based calculator using ChatGPT-4 through multiple prompt instructions without coding. ChatGPT-4 not only provides a statistical model for SMILE nomograms but also creates a calculator for user convenience. Clinicians can easily build their own nomogram calculators using only the collected data without coding. The advanced LLM will allow clinicians to conveniently create customized nomogram tools.
为评估使用大型语言模型ChatGPT-4基于特定机构数据开发用于小切口透镜切除术（SMILE）手术的定制列线图计算器的可行性，且无需任何编码专业知识。由于不同视力矫正中心的手术操作、患者群体和诊断设备存在差异，定制列线图至关重要。 对1268只接受SMILE手术的眼睛的数据进行了连续患者的回顾性分析。收集并分析了术前测量数据和术后6个月的屈光不正情况。整个数据集按3:1的比例分为训练集和验证集。在对数据进行匿名化处理后，指示ChatGPT-4进行线性回归分析，以使用术前数据预测术后屈光不正。随后，我们指示ChatGPT-4生成基于网页的列线图计算器的HTML代码，该计算器输入术前数据并使用推导公式计算手术参数。将使用ChatGPT-4进行的回归分析结果与使用两种传统统计软件程序R和SPSS获得的结果进行比较。 ChatGPT-4成功进行了SMILE列线图回归分析。预测的SMILE参数与使用统计软件获得的参数无显著差异。该列线图对术后屈光不正的预测能力高于简单的经验列线图（<0.001）。我们通过多次提示指令，在不进行编码的情况下使用ChatGPT-4成功创建了一个基于网页的计算器。ChatGPT-4不仅为SMILE列线图提供了统计模型，还创建了一个方便用户使用的计算器。临床医生仅使用收集到的数据，无需编码即可轻松构建自己的列线图计算器。先进的大型语言模型将使临床医生能够方便地创建定制的列线图工具。
