关键词: 甲状腺癌, 乳头状, 淋巴转移, 影响因素分析

Abstract: Objective To identify the influencing factors for cervical lymph node metastasis (LNM) and lateral lymph node metastasis (LLNM) in papillary thyroid carcinoma (PTC) by combining key CT imaging features, preoperative thyroid function tests, and clinical baseline information, and to construct a visualized predictive model for metastasis. 
 Methods We selected 122 PTC patients treated from January 2021 to Apirl 2024 at the First Affiliated Hospital of Anhui University of Science and Technology, and divided them into the metastasis group (n=51), including 30 patients with LLNM, and non-metastasis group (n=71). Clinical and pathological features of these patients were analyzed retrospectively. Logistic regression and random forest models were constructed to identify and validate variables potentially associated with LNM or LLNM. Visualized nomograms were developed to display the independent risk values related to lymphatic metastasis in PTC patients. 
 Results Significant differences were observed between the metastasis and non-metastasis groups in terms of multifocality, microcalcification, capsular invasion, tumor location, tumor size, serum thyroid-stimulating hormone (TSH), and free triiodothyronine (FT3) levels (P＜0.05). Multivariate Logistic regression analysis identified microcalcification, tumor location, tumor size, serum TSH, and FT3 levels as risk factors for LNM (P＜0.05). Significant differences in tumor location, tumor size, and serum TSH levels were found between the non-LLNM and LLNM groups (P＜0.05). Multivariate Logistic regression analysis indicated that tumor size and serum TSH levels were risk factors for LLNM in the metastasis group (P＜0.05). The receiver operating characteristic (ROC) curves ［the area under curve (AUC) values of 0.884 and 0.894］ demonstrated good predictive performance of the models. Clinical decision curves and calibration curves showed favorable clinical guidance functions. The random forest model validated the above models with consistent conclusions. Analysis of the training sets revealed that multifocality, TSH, FT3, tumor size, and tumor location were the highest-ranked factors associated with LNM, while TSH and tumor size were the highest-ranked factors associated with LLNM. The validation set ROC curves (AUC values of 0.845 and 0.862) further confirmed the good predictive ability of the models. 
 Conclusion The visualized nomograms reveal that clinical features such as multifocality, TSH, FT3, tumor size, and tumor location are closely associated with LNM, while TSH and tumor size are risk factors for LLNM. These findings could assist clinicians in developing personalized treatment plans for PTC patients. 

Key words: thyroid cancer, papillary, lymphatic metastasis, influencing factor analysis