Prediction model of lateral cervical lymph node metastasis in papillary thyroid carcinoma based on SEER database
Highlight box
Key findings
• A predictive model for lateral lymph node metastasis (LLNM) in papillary thyroid carcinoma (PTC) was developed using data from the Surveillance, Epidemiology, and End Results (SEER) database.
What is known and what is new?
• PTC is prone to LLNM, which impacts prognosis. Current diagnostic methods like ultrasonography have limited sensitivity, and prophylactic lateral neck dissection (LND) is not recommended for patients without LLNM evidence due to risks.
• This study developed a predictive model for LLNM using SEER data, integrating demographic and clinicopathological factors. It identified specific risk factors [sex, age, race, tumor (T) stage, metastasis (M) stage] and provided a precise preoperative LLNM risk assessment, improving surgical decision-making.
What is the implication, and what should change now?
• The model helps identify high-risk LLNM patients, enabling targeted surgical interventions and reducing unnecessary prophylactic LNDs, thus minimizing complications.
• Clinicians should integrate this model into preoperative PTC assessments. Further research should validate the model with external datasets and include additional variables (e.g., BRAF mutation) to enhance accuracy.
Introduction
Background
Thyroid carcinoma (TC) is a common malignant endocrine tumor. According to global cancer statistics for 2020, TC ranks seventh in cancer incidence (1). Histopathologically, TC can be classified into papillary, follicular, medullary, and anaplastic (undifferentiated) carcinomas. Papillary thyroid carcinoma (PTC) is the most common subtype, accounting for approximately 80% of all cases (2). Most PTCs have low malignant potential and exhibit a favorable prognosis following surgical resection and radioactive iodine treatment (3). However, lymph node metastasis (LNM) is common in PTC, occurring in approximately 20–90% of patients (4). LNM can be further categorized as central lymph node metastasis (CLNM) or lateral lymph node metastasis (LLNM) based on neck classification standards. LLNM occurs in 3.1% to 65.4% of cases and significantly impacts the prognosis of patients with PTC (5). Although neck dissection using a compartment-oriented approach is considered essential, the debate regarding which patients with PTC actually require this procedure is ongoing (6).
Rationale and knowledge gap
High-resolution ultrasonography is the preferred imaging method used by clinicians for preoperative assessment of cervical lymph nodes. However, its sensitivity for preoperatively detecting LMN is limited to ultrasonography, ranging from 38% to 59% (7). Additionally, the accuracy of preoperative ultrasonography in diagnosing LLNM varies significantly among physicians depending on their experience levels, with a positive predictive value of only 38.5% for those with insufficient experience (8). For individuals with clinically negative lateral neck, prophylactic lateral neck dissection (LND) is not recommended unless preoperative fine-needle aspiration cytology validates the suspected LLNM (9,10).
Objective
This study aimed to develop a predictive model for the early detection of LLNM and residual disease risk. A more precise preoperative assessment would enable targeted surgical intervention and improve patient prognosis. In addition, preoperative assessment can reduce the risk of reoperation and postoperative recurrence while sparing low-risk patients from the psychological and financial burden associated with overtreatment. We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-aw-472/rc).
Methods
Data selection from the Surveillance, Epidemiology, and End Results (SEER) database
Data were retrieved from the SEER database (www.seer.cancer.gov) regarding individuals diagnosed with PTC during 2016 to 2020. The process of choosing patients utilized SEER Stat 8.4.3, sourced from the “Incidence-SEER Research data, 17 Registries, Nov 2022 Sub (2000–2020)” SEER database.
Criteria for inclusion included: diagnoses between 2016 and 2020 using histopathological codes from the International Classification of Diseases for Oncology, Third Edition (ICD-O-3), 8050, 8260, 8340, 8341, and 8344; individuals who received a full thyroidectomy for PTC; and those who had both lateral lymph node dissection (LLND) and central lymph node dissection (CLND). Criteria for exclusion included: patients without clear clinicopathological profiles or insufficient clinical information; those under 18 and over 85 years old; and individuals with a concurrent history of malignant tumors.
Collected clinical and demographic data included: extrathyroidal extension (ETE), tumor dimensions, tumor (T) stage, metastasis (M) stage, age, ethnicity, and marital status. The LNM data for patients were gathered concurrently. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.
Regarding the scope and standards of lymph node dissection during thyroid cancer surgery, the clinical practice in the United States mainly follows the guidelines issued by authoritative institutions such as the American Thyroid Association (ATA). Total lymph node calculations encompassed both the central and lateral sections, while simultaneous operations involved cervical lymph node dissection and thyroidectomy. The surgical procedure for lymph node dissection encompassed either LLND or CLND. The initial procedure involved dissection of the central lymph nodes. Enhanced cervical computed tomography (CT) and ultrasonography were employed for preoperative neck mapping. When imaging suggested or fine-needle aspiration biopsy verified lateral compartment LNM, patients were subjected to dissection in areas II, III, IV, and Vb.
Statistical analysis
After dividing the dataset from the SEER database randomly into training and validation groups in a 7:3 ratio, the variables underwent comparison. The univariate analysis utilized either the Student’s t-test or the rank-sum test for examining continuous variables, while the Chi-squared test or Fisher’s exact test was applied for analyzing categorical variables. For the creation of a nomogram forecasting model and pinpointing distinct risk elements for lateral cervical LNM in the training group, multivariate analysis was conducted using the least absolute shrinkage and selection operator (LASSO) logistic regression method. The nomogram’s effectiveness was assessed using receiver operating characteristic (ROC) and Calibration curves. A decision curve analysis (DCA) was employed to determine the prediction’s net benefit limit. Results where the P value fell below 0.05 were deemed to hold statistical significance. All statistical evaluations were performed using the R software (version 4.2.2).
Results
Baseline patient’s characteristics
A comprehensive analysis of the clinical and demographic characteristics of the 18,342 participants in the prediction study was conducted using information from the SEER database, a thorough examination. Participants were randomly selected into a training cohort (12,839 participants) and an internal testing cohort (5,503 participants). These findings provide a robust foundation for the characteristics required for classification and predictive modeling in cancer research. The baseline characteristics of the enrolled patients are presented in Table 1.
Table 1
| Characteristics | Training cohort | Validation cohort | |||||
|---|---|---|---|---|---|---|---|
| Non-LLNM (n=5,960) | LLNM (n=6,879) | P value† | Non-LLNM (n=2,599) | LLNM (n=2,904) | P value† | ||
| Sex | <0.001 | <0.001 | |||||
| Male | 1,036 (17%) | 2,115 (31%) | 440 (17%) | 921 (32%) | |||
| Female | 4,924 (83%) | 4,764 (69%) | 2,159 (83%) | 1,983 (68%) | |||
| Age, years | <0.001 | <0.001 | |||||
| Median [IQR] | 48 [37, 59] | 43 [32, 55] | 49 [38, 59] | 43 [33, 55] | |||
| Race | <0.001 | <0.001 | |||||
| White | 4,944 (83%) | 5,510 (80%) | 2,121 (82%) | 2,344 (81%) | |||
| Black | 288 (5%) | 211 (3%) | 130 (5%) | 83 (3%) | |||
| Asian or Pacific Islander | 681 (11%) | 1,098 (16%) | 328 (13%) | 452 (16%) | |||
| American Indian/Alaska Native | 47 (1%) | 60 (1%) | 20 (1%) | 25 (1%) | |||
| Marital status | <0.001 | <0.001 | |||||
| Married | 3,847 (65%) | 4,086 (59%) | 1,662 (64%) | 1,757 (61%) | |||
| Single | 1,371 (23%) | 2,120 (31%) | 605 (23%) | 846 (29%) | |||
| Other | 742 (12%) | 673 (10%) | 332 (13%) | 301 (10%) | |||
| Tumor size, mm | <0.001 | <0.001 | |||||
| Median [IQR] | 13 [8, 23] | 20 [13, 31] | 13 [8, 22] | 20 [12, 31] | |||
| T | <0.001 | <0.001 | |||||
| T1a | 2,167 (36%) | 1,039 (15%) | 945 (36%) | 439 (15%) | |||
| T1b | 1,804 (30%) | 1,901 (28%) | 825 (32%) | 793 (27%) | |||
| T2 | 1,110 (19%) | 1,563 (23%) | 446 (17%) | 635 (22%) | |||
| T3 | 806 (14%) | 1,936 (28%) | 346 (13%) | 859 (30%) | |||
| T4a | 58 (1%) | 366 (5%) | 34 (1%) | 144 (5%) | |||
| T4b | 15 (0%) | 74 (1%) | 3 (0%) | 34 (1%) | |||
| M | <0.001 | <0.001 | |||||
| M0 | 5,937 (100%) | 6,750 (98%) | 2,586 (99%) | 2,833 (98%) | |||
| M1 | 23 (0%) | 129 (2%) | 13 (1%) | 71 (2%) | |||
†, Pearson’s Chi-squared test, Wilcoxon rank sum test. IQR, interquartile range; LLNM, lateral lymph node metastasis; M, metastasis; T, tumor.
Univariate and multivariate analysis of risk factors for LLNM
Independent risk factors associated with LLNM were analyzed using logistic regression, with a backward stepwise elimination approach to identify the relevant risk factors. Univariate analyses were conducted to compare the indices between different outcome groups. As shown in Table 2, LLNM significantly correlated with sex, age, race, T stage, and M stage in the univariate analysis (all P<0.05).
Table 2
| Variables | Univariate | Multivariate | |||
|---|---|---|---|---|---|
| P | OR (95% CI) | P | OR (95% CI) | ||
| Sex | |||||
| Male | 1.00 (reference) | 1.00 (reference) | |||
| Female | <0.001* | 2.19 (2.01–2.38) | <0.001* | 2.18 (1.99–2.39) | |
| Age | <0.001* | 0.98 (0.98–0.98) | <0.001* | 0.97 (0.97–0.98) | |
| Race | |||||
| White | 1.00 (reference) | 1.00 (reference) | |||
| Black | 0.64 | 1.10 (0.74–1.65) | 0.50 | 1.16 (0.75–1.79) | |
| Asian or Pacific Islander | 0.001* | 0.50 (0.32–0.77) | 0.029* | 0.60 (0.38–0.95) | |
| American Indian/Alaska Native | 0.23 | 0.79 (0.53–1.16) | 0.51 | 0.87 (0.57–1.33) | |
| T | |||||
| T1a | 1.00 (reference) | 1.00 (reference) | |||
| T1b | <0.001* | 2.14 (1.94–2.35) | <0.001* | 2.02 (1.83–2.23) | |
| T2 | <0.001* | 3.01 (2.70–3.35) | <0.001* | 2.67 (2.39–2.98) | |
| T3 | <0.001* | 5.16 (4.62–5.77) | <0.001* | 4.74 (4.23–5.31) | |
| T4a | <0.001* | 11.46 (8.75–14.99) | <0.001* | 12.58 (9.53–16.60) | |
| T4b | <0.001* | 10.38 (6.03–17.86) | <0.001* | 11.00 (6.26–19.32) | |
| M | |||||
| M0 | 1.00 (reference) | 1.00 (reference) | |||
| M1 | <0.001* | 4.53 (3.06–6.72) | <0.001* | 2.89 (1.89–4.42) | |
*, P<0.05. CI, confidence interval; M, metastasis; OR, odds ratio; T, tumor.
Variables with a P value <0.05 in univariate analysis were included in the multivariate logistic regression model. The results are presented in Table 2 and visualized in a forest map (Figure 1). We created a multivariate logistic regression model with five independent predictors (sex, age at diagnosis, race, T stage, and M stage) to predict LLNM based on the aforementioned results.
Development of a nomogram for predicting LLNM in PTC patients
To further evaluate the percentage of each independent risk factor for LLNM metastasis, a nomogram was developed, with each independent risk factor represented as a broken line. The five key predictors identified through multivariate logistic regression were incorporated into the model. Additionally, each predictor was integrated and displayed in the form of a nomogram. The regression coefficients were determined for their corresponding scores (Figure 2).
The nomogram demonstrated strong predictive performance for LLNM. In the ROC analysis of the development cohort, the area under the curve (AUC) was 0.715 [95% confidence interval (CI): 0.706–0.724], as shown in Figure 3A. The model was further validated using an internal validation cohort, where the AUC was 0.707 (95% CI: 0.693–0.720), as shown in Figure 3B.
Model calibration was assessed through bootstrap resampling (1,000 runs). The calibration curve of the development cohort closely aligned with the ideal 45° line, indicating good agreement between predicted and actual probabilities, as illustrated in Figure 4A. Similarly, the calibration curve of the validation cohort confirmed the model’s reliability and validity, as shown in Figure 4B.
To further investigate the nomogram model’s clinical use, we created clinical models to test its validity. Figure 5 depicts how the decision curve reveals the superiority of the predictive model in clinical decision-making. Risk thresholds were dynamically adjusted based on individual patient clinicopathological characteristics, highlighting the nomogram’s practical utility in guiding treatment decisions.
Discussion
Although PTC generally has a favorable prognosis, it is highly prone to LNM. PTC patients with LLNM tend to have a poor prognosis, necessitating more aggressive surgical treatment, particularly for high-risk individuals who are prone to LLNM (11). Currently, ultrasonography is the most commonly used method for detecting metastatic lymph nodes in the neck; however, its diagnostic sensitivity for LLNM is relatively low and limited by the level of experience of the imaging physician (12). However, LND is a complex procedure associated with risks such as unintentional injuries and complications, including bleeding, chyle leak, nerve dysfunction, and surgical site infection (13,14). Clinical guidelines do not recommend prophylactic LND in patients without evidence of negative lymph nodes in the lateral neck region (15-17). Therefore, developing a predictive model capable of early and accurate diagnosis of the occurrence of LLNM in PTC is important for patients with PTC and can assist clinicians in deciding surgical options.
In a cross-sectional study by Huang et al., risk factors for LLNM were evaluated in 1,066 patients with papillary thyroid microcarcinoma (PTMC). ETE, multifocality, and CLNM were identified as significant factors; however, a model that would allow for individualized and accurate prediction was not constructed (3). In a retrospective study by Zhuo et al., involving 477 patients with PTC who underwent surgery, a nomogram was developed for the noninvasive prediction of LLMN based on ultrasonography and clinical features. Although the nomogram achieved a high C-index of 0.880, its applicability was limited to Asian populations due to the single-center design (18). Wang et al. established a nomogram model for predicting LLNM based on 355 patients with PTC, incorporating clinicopathological features, such as serum iodine concentration. The model demonstrated AUCs of 0.795 and 0.792 for the training and validation cohorts, respectively. Nevertheless, the inherent bias that exists as one of the limitations of single-center studies could not be eliminated (19). In this study, we included 1,106 patients in the training cohort, identified five demographic and clinicopathological characteristics as predictors, and constructed a nomogram to evaluate the probability of LLNM in patients with PTC using logistic regression analysis. During model evaluation, the ROC and calibration curves of both the training and validation cohorts indicated that this model had high predictive accuracy. Additionally, the DCA has shown good clinical utility. This predictive model can assist clinicians in accurately assessing the likelihood of a patient presenting with LLNM, and making appropriate surgical decisions.
The demographic factors included in our models were sex, age, and race. Our findings indicate that male sex and younger age were significantly associated with the probability of developing LLNM in patients with PTC. A meta-analysis that included 16 studies showed that male sex was a significant risk factor for LLNM (5). Similarly, a Korean study that analyzed 5,656 patients with PTMC found that being male and young were independent predictors of LLNM (20). This finding is consistent with the results of our study. Race was also strongly associated with the occurrence of LLNM. Our study showed that Black patients had the lowest incidence of LLNM, whereas Asian populations were most likely to develop LLNM. An analysis of 2,737 patients showed that LLNM occurred in 4.4% of Black patients compared to 13.2% of Asian patients (21). Therefore, these findings highlight the importance of clinicians carefully evaluating the possibility of LLNM risk in both male and younger Asian populations.
The TNM staging of PTC, which includes tumor diameter and ETE, is also an important part of LLNM evaluation (22). In our study, the T stage was an extremely important risk factor. In previous studies by Caliskan et al. and Huang et al., both showed that tumor diameter and ETE had a significant impact on the occurrence of LLNM in patients with PTC, which is consistent with the results of our study (22,23). Additionally, Khan et al. reported a significantly higher rate of LNM in patients with PTC and distant metastases (24). This also suggests that the M stage has a significant effect on the occurrence of LLNM.
However, our study has some limitations. Firstly, as this is a retrospective study based on the SEER database, it is susceptible to potential selection biases, such as the surgical proficiency of each surgeon in grasping the anatomical location. Secondly, the SEER database has its inherent limitations, lacking some other key prognostic factors such as specific treatment details, auxiliary examination results, disease recurrence information, and molecular markers (BRAF, RAS, TERT mutations), as well as information on multiple nodules and extrathyroidal invasion, which have been proven to be related to LNM. Finally, this predictive model has not yet been externally validated through an independent database. In the future, it is necessary to conduct large-scale prospective studies to integrate more data on relevant variables in the real world, thereby further improving the diagnostic performance of this predictive model.
Conclusions
In summary, we successfully developed a diagnostic model for predicting the risk of LLNM in patients with PTC. This model provides an intuitive risk assessment tool for clinicians by integrating multiple demographic and clinicopathological predictors. The findings revealed that the model has high prediction accuracy and can effectively identify high-risk groups of patients with LLNM, thus providing a scientific basis for the formulation of individualized treatment plans and assisting in the optimization of clinical decision-making.
Acknowledgments
The authors extend their sincere gratitude to the Surveillance, Epidemiology, and End Results (SEER) Program for providing the valuable data used in this study.
Footnote
Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-aw-472/rc
Peer Review File: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-aw-472/prf
Funding: None.
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-aw-472/coif). The authors have no conflicts of interest to declare.
Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.
Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.
References
- Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
- Rosenbaum MA, McHenry CR. Contemporary management of papillary carcinoma of the thyroid gland. Expert Rev Anticancer Ther 2009;9:317-29. [Crossref] [PubMed]
- Huang Y, Yin Y, Zhou W. Risk Factors for Central and Lateral Lymph Node Metastases in Patients With Papillary Thyroid Micro-Carcinoma: Retrospective Analysis on 484 Cases. Front Endocrinol (Lausanne) 2021;12:640565. [Crossref] [PubMed]
- Hay ID. Papillary thyroid carcinoma. Endocrinol Metab Clin North Am 1990;19:545-76.
- So YK, Kim MJ, Kim S, et al. Lateral lymph node metastasis in papillary thyroid carcinoma: A systematic review and meta-analysis for prevalence, risk factors, and location. Int J Surg 2018;50:94-103. [Crossref] [PubMed]
- Tao Y, Wang C, Li L, et al. Clinicopathological features for predicting central and lateral lymph node metastasis in papillary thyroid microcarcinoma: Analysis of 66 cases that underwent central and lateral lymph node dissection. Mol Clin Oncol 2017;6:49-55. [Crossref] [PubMed]
- Hu W, Wang H, Wei R, et al. MRI-based radiomics analysis to predict preoperative lymph node metastasis in papillary thyroid carcinoma. Gland Surg 2020;9:1214-26. [Crossref] [PubMed]
- Moon HJ, Kim EK, Yoon JH, et al. Differences in the diagnostic performances of staging US for thyroid malignancy according to experience. Ultrasound Med Biol 2012;38:568-73. [Crossref] [PubMed]
- Machens A, Hauptmann S, Dralle H. Lymph node dissection in the lateral neck for completion in central node-positive papillary thyroid cancer. Surgery 2009;145:176-81. [Crossref] [PubMed]
- Feng JW, Wu WX, Qi GF, et al. Nomograms based on sonographic and clinicopathological characteristics to predict lateral lymph node metastasis in classic papillary thyroid carcinoma. J Endocrinol Invest 2022;45:2043-57. [Crossref] [PubMed]
- Ito Y, Higashiyama T, Takamura Y, et al. Risk factors for recurrence to the lymph node in papillary thyroid carcinoma patients without preoperatively detectable lateral node metastasis: validity of prophylactic modified radical neck dissection. World J Surg 2007;31:2085-91. [Crossref] [PubMed]
- Tong Y, Zhang J, Wei Y, et al. Ultrasound-based radiomics analysis for preoperative prediction of central and lateral cervical lymph node metastasis in papillary thyroid carcinoma: a multi-institutional study. BMC Med Imaging 2022;22:82. [Crossref] [PubMed]
- Glenn JA, Yen TW, Fareau GG, et al. Institutional experience with lateral neck dissections for thyroid cancer. Surgery 2015;158:972-8; discussion 978-80. [Crossref] [PubMed]
- Feng JW, Yang XH, Wu BQ, et al. Predictive factors for central lymph node and lateral cervical lymph node metastases in papillary thyroid carcinoma. Clin Transl Oncol 2019;21:1482-91. [Crossref] [PubMed]
- Haugen BR, Alexander EK, Bible KC, et al. 2015 American Thyroid Association Management Guidelines for Adult Patients with Thyroid Nodules and Differentiated Thyroid Cancer: The American Thyroid Association Guidelines Task Force on Thyroid Nodules and Differentiated Thyroid Cancer. Thyroid 2016;26:1-133. [Crossref] [PubMed]
- Takami H, Ito Y, Okamoto T, et al. Revisiting the guidelines issued by the Japanese Society of Thyroid Surgeons and Japan Association of Endocrine Surgeons: a gradual move towards consensus between Japanese and western practice in the management of thyroid carcinoma. World J Surg 2014;38:2002-10. [Crossref] [PubMed]
- Zhu J, Chang L, Li D, et al. Nomogram for preoperative estimation risk of lateral cervical lymph node metastasis in papillary thyroid carcinoma: a multicenter study. Cancer Imaging 2023;23:55. [Crossref] [PubMed]
- Zhuo X, Yu J, Chen Z, et al. Dynamic Nomogram for Predicting Lateral Cervical Lymph Node Metastasis in Papillary Thyroid Carcinoma. Otolaryngol Head Neck Surg 2022;166:444-53. [Crossref] [PubMed]
- Wang J, Gao Y, Zong Y, et al. Nomogram Model Based on Iodine Nutrition and Clinical Characteristics of Papillary Thyroid Carcinoma to Predict Lateral Lymph Node Metastasis. Cancer Control 2023;30:10732748231193248. [Crossref] [PubMed]
- Kim SK, Park I, Woo JW, et al. Predictive Factors for Lymph Node Metastasis in Papillary Thyroid Microcarcinoma. Ann Surg Oncol 2016;23:2866-73. [Crossref] [PubMed]
- Zheng H, Lai V, Lu J, et al. Comparing the rate and extent of malignancy in surgically excised thyroid nodules across race and ethnicity. Am J Surg 2022;223:617-23. [Crossref] [PubMed]
- Caliskan O, Unlu MT, Yanar C, et al. Predictive Factors Affecting the Development of Lateral Lymph Node Metastasis in Papillary Thyroid Cancer. Sisli Etfal Hastan Tip Bul 2023;57:312-9. [Crossref] [PubMed]
- Huang H, Xu S, Ni S, et al. A nomogram for predicting lateral lymph node metastasis in cN0 unifocal papillary thyroid microcarcinoma. BMC Cancer 2023;23:718. [Crossref] [PubMed]
- Khan U, Al Afif A, Aldaihani A, et al. Patient and tumor factors contributing to distant metastasis in well-differentiated thyroid cancer: a retrospective cohort study. J Otolaryngol Head Neck Surg 2020;49:78. [Crossref] [PubMed]
