Bertolli Eduardo, Micheletti Sara B, de Camargo Veridiana P, da Silva Tiago V, Bacchi Carlos E, Buzaid Antonio C
Surgical Oncology.
Oncology, BP Hospital, São Paulo.
Melanoma Res. 2025 Aug 1;35(4):227-231. doi: 10.1097/CMR.0000000000001047. Epub 2025 May 27.
Nomograms are commonly used in oncology to assist clinicians in individualized decision-making processes, such as considering sentinel node biopsy (SNB) for melanoma patients. Concurrently, artificial intelligence (AI) is increasingly being utilized in medical predictions. This study aims to compare the predictive accuracy of nomograms and AI platforms for SNB positivity in a real-world cohort of melanoma patients. A retrospective analysis of melanoma patients who underwent SNB from 2020 to 2024 in a single institution was performed. Three open-access nomograms and three public AI platforms were employed to assess SNB positivity based on comprehensive clinical and pathological characteristics. Our cohort comprised 62 melanoma patients who have undergone SNB, of whom 12 (19.4%) were positive. There was no concordance among the three nomograms, nor among AI platforms ( P < 0.001). Only the Memorial Sloan Kettering Cancer Center (MSKCC) nomogram scored statistically different between positive and negative SNB ( P = 0.04), and ChatGPT was the only AI platform that was also statistically significant ( P = 0.02). Only ChatGPT score was statistically significant for SNB positivity after univariate logistic regression (odds ratio: 1.05; 95% confidence interval: 1.004-1.108; P = 0.03). A receiver operating characteristic curve based on ChatGPT predictions generated a model with an area under the curve (AUC) of 0.702. Integrating MSKCC predictions marginally improved the model's predictive performance, enhancing the AUC to 0.715. In conclusion, SNB positivity could be better performed by an AI platform in this cohort of patients. Enhancing AI platforms could provide better populations for nomogram validation, which would lead to better predictive models.
列线图在肿瘤学中常用于协助临床医生进行个体化决策，比如考虑对黑色素瘤患者进行前哨淋巴结活检（SNB）。与此同时，人工智能（AI）在医学预测中的应用越来越广泛。本研究旨在比较列线图和AI平台在一组真实世界黑色素瘤患者中预测SNB阳性的准确性。对2020年至2024年在单一机构接受SNB的黑色素瘤患者进行了回顾性分析。使用三个开放获取的列线图和三个公共AI平台，根据全面的临床和病理特征评估SNB阳性情况。我们的队列包括62例接受SNB的黑色素瘤患者，其中12例（19.4%）为阳性。三个列线图之间以及AI平台之间均无一致性（P < 0.001）。只有纪念斯隆凯特琳癌症中心（MSKCC）列线图在SNB阳性和阴性之间的得分有统计学差异（P = 0.04），ChatGPT是唯一具有统计学意义的AI平台（P = 0.02）。单因素逻辑回归后，只有ChatGPT评分对SNB阳性具有统计学意义（优势比：1.05；95%置信区间：1.004 - 1.108；P = 0.03）。基于ChatGPT预测生成的受试者工作特征曲线模型的曲线下面积（AUC）为0.702。整合MSKCC预测略微提高了模型的预测性能，将AUC提高到0.715。总之，在这组患者中，AI平台对SNB阳性的预测表现更佳。增强AI平台可为列线图验证提供更好的人群，从而产生更好的预测模型。
