Page 20 - 《南京医科大学学报》2026年第1期
P. 20
南京医科大学学报(自然科学版) 第46卷第1期
· 14 · Journal of Nanjing Medical University(Natural Sciences) 2026年1月
·专题研究:肿瘤·
基于多模态数据的甲状腺髓样癌中央区淋巴结转移术前预测
模型的构建与分析
张 祥 ,柳 卫 ,杨倩倩 ,张 燕 1*
3
2
1
南京医科大学第一附属医院检验学部,核医学科,江苏 南京 210029;南京梅山医院检验科,江苏 南京 210039
1 2 3
[摘 要] 目的:基于多模态数据构建甲状腺髓样癌(medullary thyroid carcinoma,MTC)中央区淋巴结转移(central lymph node
metastasis,CLNM)的预测模型,并分析其临床意义。方法:回顾性分析2017年1月—2025年5月南京医科大学第一附属医院收
治的104例MTC患者的临床病理资料、术前影像学特征及实验室指标。通过单因素Logistic回归初步筛选变量(P < 0.1),进一
步采用向后逐步回归法进行多因素Logistic回归分析筛选CLNM的独立危险因素,构建预测模型并绘制列线图,受试者工作特
征(receiver operating characteristic,ROC)曲线、校准曲线和决策曲线分析(decision curve analysis,DCA)评估模型的区分度、校准度
和临床适用性,使用Bootstrap法进行内部验证。结果:根据病理结果是否存在CLNM将104例MTC患者分成转移组(55例)与非转
移组(49 例)。与非转移组患者相比,转移组患者性别(P=0.001)、超声形态是否规则(P < 0.001)、超声边缘是否光整(P <
0.001)、血清癌胚抗原(carcinoembryonic antigen,CEA)水平(P=0.006)、血清降钙素(calcitonin,CT)水平(P < 0.001)等差异有统
计学意义。多因素Logistic 回归分析显示,患者的性别(OR=6.50,95%CI:2.03~20.81,P=0.002)、超声边缘是否光整(OR=9.77,
95%CI:3.12~30.59,P < 0.001)以及血清CT水平(OR=1.25,95%CI:1.10~1.42,P < 0.001)是CLNM的独立危险因素。联合三者
建立的列线图模型可良好地识别CLNM,ROC曲线下面积(area under the curve,AUC)为0.873,95%CI:0.808~0.939,校准曲线和
DCA均表明该模型具有良好的性能及临床适用性。使用Bootstrap法进行内部验证也显示该模型具有良好的稳定性和可靠性
(AUC=0.874,95%CI:0.865~0.879)。结论:结合患者的性别、超声边缘是否光整以及血清CT水平的多模态数据模型能有效预
测MTC患者CLNM风险,为临床决策提供依据。
[关键词] 甲状腺髓样癌;中央区淋巴结转移;多模态数据;预测模型
[中图分类号] R736.1 [文献标志码] A [文章编号] 1007⁃4368(2026)01⁃14⁃07
doi:10.7655/NYDXBNSN250910
Construction and analysis of a preoperative prediction model for central lymph node
metastasis in medullary thyroid carcinoma based on multimodal data
1 2 3 1*
ZHANG Xiang ,LIU Wei ,YANG Qianqian ,ZHANG Yan
2
1 Department of Laboratory Medicine,Department of Nuclear Medicine,the First Affiliated Hospital of Nanjing
3
Medical University,Nanjing 210029;Department of Laboratory Medicine,Nanjing Meishan Hospital,Nanjing
210039,China
[Abstract] Objective:To develop and validate a multimodal data⁃based predictive model for central lymph node metastasis(CLNM)
in patients with medullary thyroid carcinoma(MTC)and evaluate its clinical significance. Methods:We retrospectively analyzed
clinical⁃pathological data,preoperative imaging features,and laboratory parameters of 104 MTC patients treated at the First Affiliated
Hospital of Nanjing Medical University between January 2017 and May 2025. Potential predictors(P < 0.1 in univariate analysis)were
included in a multivariate logistic regression model with backward stepwise selection to identify independent risk factors for CLNM. A
prediction model was constructed and a nomogram was drawn. The receiver operating characteristic(ROC)curve,calibration curve and
decision curve analysis(DCA)were used to evaluate the discrimination,calibration,and clinical applicability of the model. Internal
validation was performed via bootstrap resampling. Results:According to the presence or absence of CLNM in the pathological results,
[基金项目] 江苏省医学重点学科(ZDXK202239)
通信作者(Corresponding author),E⁃mail:zhangyanzd@njmu.edu.cn(ORCID:0009⁃0004⁃1213⁃9103)
∗

