In this large-scale multicenter retrospective study, we constructed a nomogram and scoring system that screened the clinical risk factors for LNM in patients with PTC. The results showed that both models could significantly improve the preoperative prediction of LNM based on US. Moreover, both models performed well on different baseline datasets, indicating that they can be applied to assist clinicians in improving the diagnostic accuracy of LNM in patients with PTC in clinical practice.
It has been reported that LNM occurs in 30%-60% of PTC patients, and up to 30% with a diameter less than 0.3 cm show microscopic LNM[2]. LNM is significantly associated with local recurrence[7]. The predictive performance of most universal preoperative examinations for LNM, including US, is relatively low owing to the anatomy of the cervical lymph nodes and the physician’s experience. In this study, the sensitivity, specificity, and mean AUROC of US in each dataset were approximately 59%, 72%, and 0.6428, respectively, indicating poor performance in detecting LNM. Therefore, there is an urgent need to develop noninvasive preoperative prediction models with better performance, convenience, and operability to predict LNM. As shown in this study, both the nomogram and scoring system constructed using noninvasive clinical characteristics can significantly improve the performance of the preoperative prediction of LNM based on US alone.
As results show, ultimately six independent risk factors, including age, sex, US diagnosis of LNM, tumor diameter, tumor location, and preoperative TPOAb level, were found to be significantly associated with LNM. In summary, younger and male patients with PTC and larger bilateral tumors are relatively more likely to develop LNM, which is consistent with previous studies[7, 8]. As an important risk factor for LNM, the cutoff value for age remains controversial, but it is generally believed that younger patients are at a higher risk of LNM. The cutoff value for age in our study was 40 years, which is consistent with trends reported in other studies[9, 10]. Although PTC is less common in men than in women, men are more likely to develop a more aggressive phenotype and have a higher incidence of LNM according to clinical observations and many studies[11–13]. This gender disparity of tumor behavior may be associated with sex hormones and distinct immune microenvironments[14, 15]. Both models in our study were consistent with those used in previous studies. Therefore, more rigorous preoperative examinations and surgical strategies may be needed in younger male patients. The cutoff tumor diameter in this study was 1 cm, and larger tumor size is widely considered to be associated with LNM and higher mortality, which could serve for risk stratification[10, 16]. Previous studies have reported controversial results regarding the bilaterality of tumors as a risk factor for LNM. Several retrospective studies and meta-analyses, as well as our study, have demonstrated that patients with bilateral PTC are more likely to develop LNM[17–19], whereas others have shown that bilaterality is not significantly associated with LNM and local recurrence compared with multifocality[20, 21].
Our study showed that a low TPOAb level was an independent risk factor for LNM. TPOAb is a serological marker of Hashimoto’s thyroiditis. Previous studies have shown that patients with PTC concomitant with Hashimoto’s thyroiditis have a lower incidence of LNM and better prognosis[22]. However, there have been no consistent conclusions regarding the effect of TPOAb levels on LNM. Low TPOAb level is found to be one of the significant risk factors of central lymph node metastasis[16, 23]. While some studies have shown the opposite result, high TPOAb level (> 100 IU/mL) is an independent risk factor for LNM in PTC[24, 25]. Further studies on the effects of TPOAb on LNM and its underlying biological mechanisms are needed.
Both the nomogram and scoring systems can help clinicians improve the diagnostic accuracy of LNM and make more reasonable surgical decisions and postoperative follow-up strategies. In brief, the cut-off value of the nomogram is 0.63 (∼58 scores), that is, patients with risk values > 0.63 or scores > 58 are considered to be at high risk of LNM. For the scoring system, the cut-off value is 3.5, that is, patients with scores ≤ 3 are considered to be low-risk, and ≥ 4 to be high risk of LNM. It is recommended that PTC patients with a score ≥ 4 should undergo more rigorous examination and surgery. The major advantages of these two models are that they are noninvasive, constructed using indicators detectable preoperatively, with strong operability, using data from a large-scale multicenter study, favorable performance in multiple centers with inconsistent baseline levels, strong clinical practicability, and universality.
However, this study has several limitations. First, neither model specifically predicted LNM in the central and lateral cervical compartments owing to a lack of information on central or lateral LNM in the original datasets. Second, multiple datasets were obtained from hospitals in Southern China, which may represent specific geographical and demographic characteristics. Finally, the study may have been biased during data collection and screening owing to its retrospective nature. Multicenter prospective cohort studies are needed to further verify their predictive performance.
In conclusion, a nomogram and scoring system were constructed based on noninvasive preoperative clinical characteristics of patients with PTC and validated using distinct datasets. Both models significantly improved the preoperative prediction of LNM based on US alone. The performance of the two models was robust in datasets with inconsistent baselines, indicating that they could be applied in medical centers with varying levels of resources and heterogenic PTC populations.