A retrospective study was conducted at the Third Affiliated Hospital of Soochow University between January 2022 and May 2023. The Institutional Review Board of Changzhou First People’s Hospital approved this study. Our study collected data from 387 patients admitted to our hospital’s thyroid surgery department, who were confirmed to have PTC through pathology and who underwent LND. All patients with suspicious lymph nodes were randomly split into the training dataset and the validation dataset (2: 1). Subsequently, based on the pathology of LND, those in the training dataset (208 patients) were divided into the LLNM group and the non-LLNM group. These patients underwent preoperative ultrasound (US) examination of the thyroid and suspicious lateral cervical lymph nodes, as well as fine-needle aspiration biopsy (LN-FNAB) for suspected lymph nodes. This retrospective study was approved by the Institutional Review Board of Changzhou First People’s Hospital, and the informed consent was waived.

All study patients had undergone clinical examination, preoperative ultrasound, and the measurement of thyroglobulin in FNAB. The inclusion criteria were as follows: (1) patients with pathology-confirmed PTC who had underwent LND; (2) suspicious LLNs with preoperative LN-FNAB, including cytopathology and thyroglobulin; (3) complete clinical data; and (4) patients with a postoperative follow-up period of at least 6 months. The exclusion criteria were as follows: (1) patients who did not undergo LND; (2) LLNs without preoperative LN-FNABs; (3) subtypes other than classic PTC; (4) incomplete clinical data; (5) missing follow-up data; and (6) patients who had received treatment for head and neck cancer previously, or with a history of neck radiation history or a family history of thyroid carcinoma (TC). Finally, 208 patients with suspected LLNs with LND in 2022 were enrolled in this study as the training group, and 89 consecutive suspicious LLNs in 2023 were enrolled in this study as the validation group. In the training group, according to the pathological results of the LND, patients were divided into the LLNM group (n = 79) and the non-LLNM group (n = 129). Figure 1 shows the flow chart of the patients enrolled in our study.

The clinical and ultrasound parameters included the patients’ sex and age, the aspect ratio of the thyroid nodule (A/T), chronic lymphocytic thyroiditis (CLT), thyroid nodule site (upper pole, middle-upper third of the thyroid gland, middle pole, middle-lower third of the thyroid gland, and lower pole). If there were multiple nodules in the thyroid gland, we selected the largest or most typical nodule.

The clinical parameters included the cytopathology of the suspicious LLNs in FNABs, Tg level, and Tg/sTg. The suspicious LLNs may have one of the following ultrasound (US) characteristics, such as an A/T > 1, a globular hyperechoic mass, loss of the fatty hilum, microcalcification, and necrosis. The positive aspect of a suspicious LLN cytopathology was inherent to the detection of thyroid cells among the lymphocytes. The negative aspect of the suspicious LLN cytopathology was inherent to the fact that no thyroid cells appeared among the lymphocytes and that no suspicious LLNs were found even after 6 months of follow-up. The normal level of serum Tg ranged from 3.5 ng/mL to 77 ng/mL. Before our study, considering that the cutoff values of Tg had not been standardized, the surgeon performed LND on the basis of standard serum Tg levels. The experienced radiologists performed all the preoperative US and FNABs under ultrasound guidance. The decision of whether or not to perform LND depends on the LN-FNAB results, Tg range, doctor’s experience, and individual patient conditions. (1) When the cytopathology revealed definite malignant cells in the suspicious LLNs and the Tg level exceeded 77 ng/mL, the surgeon performed LND; (2) when indefinite malignant cells were found in the cytopathology or the Tg level ranged from 3.5 ng/mL to 77 ng/mL, LND was carefully considered; (3) if both the cytopathology or the Tg less than 3.5ng/mL, surgeons would carry out preventive lateral neck LND based on their own experience and individual circumstances; and (4) if both the cytopathology and the Tg were negative, and there were no surgical indications based on the experience of the physician and the individual condition of the patient, then LND was avoided, but it was suggested that patients were followed up at least every 3 months. Considering the fact that some patients have a high level of serum Tg, which could have been an indication of nodular goiter or subacute thyroiditis, some studies (9) suggested incorporating Tg/sTg into the study.

As suspicious lateral cervical lymph nodes were discovered in patients with thyroid nodules, a detailed preoperative examination including FNAB and clinical characteristics were carried out. We determined the status of suspicious LLNs based on the results of cytopathology and the level of Tg before surgery. The LND was performed for patients with a positive FNAB. The resected thyroid and LLN specimens were processed for pathological examination (including evaluating for the existence of chronic lymphocytic thyroiditis, maximum size of the thyroid nodule, and the neck level of the metastatic nodes). The pathological results pertaining to the thyroid nodules and LLNs were used as the gold standard.

Two or more experienced pathologists, who were blinded to the clinical characteristics and ultrasonographic findings, reviewed all the surgical specimens. The gold standard of this study was the pathological diagnosis. Every patient who needed to undergo LND was marked on the skin under preoperative ultrasound guidance and injected with tissue glue around the suspected lymph nodes. During the surgery, the surgeon would first remove the marked lymph nodes and send them to the pathology department separately. Subsequently, the lymph nodes around the suspected lymph nodes that are visible or palpable to the naked eye were also removed. Each cervical lymph node was fixed in 20% buffered formalin, embedded in paraffin, sectioned, and stained with hematoxylin and eosin. The pathologists investigated the lymph nodes with suspected cancer through immunohistochemical staining. All the pathological specimens were reviewed and cross-checked by two or more experienced pathologists microscopically.

The statistical analysis was performed using IBM SPSS Statistics version 26.0 software (IBM Corporation, Armonk, NY, USA), MedCalc (version 15.6; MedCalc software, Mariakerke, Belgium), and R software version 3.5.3 (The R Foundation for Statistical Computing, Vienna, Austria). Normally distributed variables were expressed as means ± standard deviations (SDs), and skew-distributed variables were reported as numbers and percentages. An independent t-test, a Pearson’s chi-squared test, or a Fisher’s exact test was used for the univariable analysis. The variables with statistical differences (p < 0.05) in the univariate analysis were included in the multivariate binary logistic regression analysis to determine the independent risk factors affecting lateral cervical lymph node metastasis. These independent predictors were used to construct the nomogram in R software for predicting the risk of LLNM. We used the receiver operating characteristic (ROC) curve in MedCalc to test our established prediction model’s discriminative power and consensus. The performance of the nomogram was further evaluated using the calibration chart, which plotted the predicted probability of the nomogram against the observed probability. In the nomogram, each variable was estimated into scores, and the scores of each variable were then added for each patient to determine the probability of LLNM based on their total score. The internal calibration curve, external calibration curve, decision curve analysis (DCA), and ROC were used to evaluate the discrimination and calibration ability of the training set and the validation set. The statistical significance was set at a p-value < 0.05.

Table 1 shows the comparison of the clinical baseline data of the training set and the validation set. A total of 208 patients in 2022 entered the training set (70%), and 89 patients from January to May 2023 entered the validation set (30%). There were no significant differences between the training set and the validation set (p > 0.05 for all comparisons).

In our study, 387 patients with 387 suspicious LLNs underwent LLN FNAB between January 2022 and December 2022. Finally, 208 patients who underwent LND were enrolled in this study. The postoperative pathology revealed the presence of LLNM in 79 patients with suspicious LLNs ( Figure 1 ).

As summarized in Table 2 , of the 208 PTC patients with 208 suspicious LLNs, 67 men (32.2%) and 141 women (67.7%) underwent LND in our study, with an average age of 43.3 years ± 11.9 years (range from 18 years to 78 years), a maximum thyroid nodule diameter of 1.25 cm ± 0.9 cm (range from 0.22 cm to 6.0 cm), and an A/T of 0.8 ± 0.1 (range from 0.37 to 1.2). The suspicious LLNs in the right neck were detected in 104 (50%) patients, and tumors in the left neck were detected in 104 (50%) patients ( Table 1 ).

LLNM had a significant association with the maximum thyroid nodule diameter, tumor site, neck level of the metastatic nodes, cytopathology, Tg level, and Tg/sTg in the univariate analysis (all p < 0.05). A binary logistic regression equation of lateral cervical lymph node metastasis was constructed based on the variables used in the univariate analysis. The multivariate analysis showed that the maximum thyroid nodules diameter [odds ratio (OR) 2.323, 95% CI 1.383 to 3.904; p = 0.001], Tg level (OR 1.007, 95% CI 1.005 to 1.009; p = 0.000), Tg/sTg (OR 1.005, 95% CI 1.001 to 1.008; p = 0.009), and cytopathology (OR 9.738, 95% CI 3.678 to 25.783; p = 0.000) remained independent predictors for LLNM in PTC patients with suspicious LLNs ( Tables 2 – 4 ).

The risk factors, including maximum thyroid nodule diameter, Tg level, Tg/sTg, and cytopathology, identified through multivariate analysis were included in the nomogram for predicting LLNM. The nomogram showed that the maximum thyroid nodule diameter and cytopathology had the greatest effects on the score ( Figure 2 ). Thereafter, this prediction model was used to develop the ROC analysis. The area under the curve (AUC) of the training model was 0.937 (95% CI 0.895 to 0.966), the sensitivity was 0.88, and the specificity was 0.89 ( Figure 3 ). The AUC of the validation model was 0.957 (95% CI 0.892 to 0.989), the sensitivity was 0.96, and the specificity was 0.92 ( Figure 3B ). In addition, we developed the internal and external validation of the LLNM model. In the nomogram, the solid line represents the true LLNM, whereas the dotted line represents the virtual incidence rate based on the prediction model. The higher the matching degree between the dashed and solid lines, the higher the accuracy of the prediction model. Our prediction model showed that the risk of LLNM was consistent with the actual results. Furthermore, we conducted the external calibration for the established model. The AUC of the training cohort and validation cohort were in good agreement ( Figures 4A, B ). The decision curve analysis (DCA) was also frequently used to evaluate the practical value of these models ( Figure 5 ). The abscissa of this graph represents the threshold probability. The vertical axis is the net benefit (NB) after subtracting the pros and cons. The solid black line (none line) and the solid grey line (all line) represent two extreme situations, respectively. The first extreme situation is represented by the solid black line. At this point, it was assumed that the patient did not have LLNM and had not undergone LND, and the NB was zero. The second extreme situation is represented by the solid grey line. At this point, it was assumed that all patients had LLNM and had undergone LND. The NB is a backslash line. The solid blue line is the result of our nomogram. The solid blue line performed significantly better than the solid grey line at most of the threshold points.