The nomogram is a highly accurate, distinguishable, simple, and convenient tool for clinical applications19 and can change patients’ treatment patterns20. Previous studies of predictive models for PTMC have ignored biochemical indicators, and most are based on postoperative histopathological findings and genetic screening. More importantly, they did not perform external validation, which affects the accuracy of the model and limits its clinical application. Our research was based on demographic characteristics, US findings, serum indicators of thyroid function, and serological indicators of autoimmune thyroid systems. We then built a model that was simple, low-cost, repeatable, radiation-free, and non-intrusive. Further, this model could be easily used in real time. Our study found young age, larger tumors, high FT3 and FT4 levels, low TPOAb levels, male sex, multifocality, calcification, an aspect ratio ≥ 1, and indistinct lymphatic hilum to be associated with increased LNM in PTMC. Then, by stepwise backward selection, we adopted AIC as the termination criterion to generate a nomogram. The nomogram had good accuracy and calibration. Therefore, the risk of CLNM occurring in PTMC can be quantified using our nomogram so that clinicians and patients can make more profitable decisions before surgery. Further, the nomogram provides a reference for evaluating treatment efficacy in patients who have undergone surgery.
Previous reports have identified sex and age as independent predictors of CLNM6,9,11,12,21,22. For young men, higher basal metabolic rates may stimulate the growth and metastasis of tumor cells. However, after reviewing 2,930 cases, Lee et al. suggested that sex is not a prognosticator of PTMC but is an independent predictor of PTC prognosis23. Our conclusions are not in agreement with their findings but confirm the finding that sex is an independent predictor of CLNM in PTMC.
US imaging is a powerful and reliable method for diagnosing PTMC and determining the CLNM risk, several US imaging findings were significantly associated with CLNM and incorporated into our nomogram. Reports generally agree that tumor size (maximum tumor diameter) based on US imaging is an essential predictor of CLNM20,24−26. Tumor size had an OR of 3.57 in this study, which indicates that for every 1 cm increase in tumor size, patients are 3.57 times more likely to have CLNM. Calcification refers to white opaque specks on the thyroid that are scattered or partially concentrated on a US image. Calcification of thyroid tumors in imaging is mainly used to determine whether the tumor is benign or malignant. Microcalcification is a critical and unique US characteristic of PTCs. It is pathologically characterised by Psammoma bodies (PBs), and this finding on FNA cytology is a significant basis for the diagnosis of PTC27. However, its mechanism of formation is unclear. It may be related to rapid tumor growth, active metabolism, or coagulation necrosis in some tissues. Previous studies have found a correlation between US microcalcification and PBs in tumors, and microcalcification is associated with LNM28. In our model, the OR of calcification was 1.7, indicating that patients with calcified lesions are 1.7 times more susceptible to develop CLNM than those without calcified lesions under the same conditions. This has not only been reported in previous thyroid cancers6,11 but also in breast cancer29,30.
We confirmed the association between CLNM and multifocality that was described in a previous meta-analysis6,9,11,31,32. However, it is not clear whether multifocality represents multiple newly developed tumors or multiple lesions originating from a single thyroid tumor33. There is evidence that multifocal lesions represent different tumors, as they often have different RET/PTC gene rearrangements and independent cloning sources. However, some studies have shown that more than 80% of multifocal PTMC tumors may be monoclonal based on genome-wide allele genotypes34. This suggests that PTMCs develop from a single clone and then develop through intrathyroidal metastasis, similar to the theory that there is a rich network of micro-lymphatic tubes in the thyroid gland35. Tumors are spread through these lymphatic networks inside the thyroid gland and the lymph nodes in the central region.
There have been few reports on the relationship between LNM and the aspect ratio27. Some studies have suggested that PTMCs grow faster in length than in width; therefore, an aspect ratio of 1 or greater can indicate the malignant nature of the tumor. Lymph hilus is a more intuitive way to determine CLNM risk than these other US characteristics. However, due to the limitations of US susceptibility to interference and the reliance on the practitioners’ ability, we had only 80 cases in the two sets where the lymphatic hilus structure disappeared, accounting for 7.7% of the total cases.
Our model also incorporates FT3, FT4, and TPOAb. Thyroid peroxidase is the main antigen component of thyroid microsomes, and its function is related to thyroxine synthesis. TPOAb and TgAb can cause chronic damage to thyroid cells and may increase the risk of PTC36–38. Unlike TgAb, which is associated with multifocal PTC and may therefore cause LNM, higher TPOAb may reduce LNM risk39. Our findings agree with these roles, although the underlying mechanism is not very clear. In contrast to FT3, higher FT4 levels are negatively correlated with PTC prevalence40. FT4 is the active component of thyroid hormone in the blood, and it can more accurately reflect the state of thyroid function. PTMC is a well-differentiated thyroid cancer, so higher FT4 levels may be related to the active state of cancer cells.
Our study has several limitations. This study only examined patients at diagnosis; therefore, the risk of CLNM in patients with recurrent disease cannot be predicted. In addition, for US characteristics such as tumor size and aspect ratio, the data measured by different technicians will deviate and decrease the accuracy of the nomogram. Second, TPOAb and FT4 levels fluctuate, and blood samples obtained at different times will be biased and interfere with the nomogram results. Third, although our model has some application prospects, positive predictive values were 28% and 37% in the training and validation corhorts, which are relatively low. Therefore, it is need to develop more detailed, accurate, uniform US evaluation standards, perform studies with larger, multi-centre sets to further improve the model evaluation capabilities, and perform forward-looking research on CLND to strengthen the model's reliability.