**3. Discussion**

Follicular thyroid cell-derived tumors are the most common endocrine malignancy and the fifth most common cancer in women [13–15]. Its annual incidence, about 3% of all cancer, has been shown to more than double over the last decades as a result of the improved ability to diagnose malignant transformation in small thyroid nodules [14,15]. The differentiated papillary (PTC) and follicular (FTC) carcinomas represent most of the epithelial thyroid cancers, which may progressively dedifferentiate giving rise to the more aggressive poorly DTC (PDTC), and to the incurable anaplastic thyroid carcinomas (ATC) [16]. Although the prognosis of DTC patients is favorable, with a 5-years-survival rate of nearly 98%, about 20% of them face the morbidity of disease recurrence [1,2,14,15]. The TNM staging system elaborated by the AJCC based on clinicopathological parameters is the most widely used approach to predict thyroid cancer survival, but it is much less reliable in discriminating patients with a higher risk of developing relapses over time [1–3]. The TNM has been significantly revised in 2016 (TNM-8), with respect to its previous version (TNM-7) [3,4,9]. Various reports have documented the ability of the TNM-8 to better predict the disease-specific survival (DSS) in DTC patients, but we could not verify this point because none of the patients examined died due to PTC [17–28]. It has to be mentioned, however, that in one of these studies it was shown that the new TNM improved the prediction of DSS for PTC, but not for FTC patients [23]. In addition, reasonable concerns have been raised for patients in the 45–54 age range, classified in stages III or IV by the TNM-7 but currently considered in stages I or II by the TNM-8, for whom the severity of the disease could be underestimated [24].

In the present study, as expected, the comparison of the two TNM systems in 1113 PTC patients evidenced a considerable patient downstaging by TNM-8, in agreement with a number of earlier reports [5,17–28]. Even though TNM staging was conceived to predict DSS, some studies showed that both TNM-7 and TNM-8 were significantly associated with disease-free survival and that TNM-8 allowed better discrimination of the recurrence risk over time [25,27,29]. In our case study we found analogous results, but the differences between curves created with TNM-7 and TNM-8 stages and T were much less evident in our Kaplan–Meier analysis due to the need for combining some categories to fulfill statistical requirements. However, when the Cox regression analysis was performed with parameters classified according to TNM-7 instead of TNM-8, lower Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values were obtained, indicating that the model based on TNM-7 should be preferred. This statement would seem to be in contrast with the conclusions of other authors about the superior predictive performance of TNM-8 over TNM-7 for DFI. However, it has to be mentioned that these authors focused attention on HRs of individual predictors, while we evaluated the virtue of model fit. Our findings are in line with a previous work showing that in patients above 55 years, pT7 was superior to pT8 in predicting DFI [28].

In the framework of The Cancer Genome Atlas (TCGA) project, a comprehensive multiplatform analysis was carried out to determine the genomic landscape of 496 PTC cases, and a reclassification of PTC into molecular subtypes was proposed to improve clinicopathological grading and management of patients [12]. In this study, the lowest thyroid differentiation score (TDS) was assigned to a tall cell-like tumor cluster, which was associated with more advanced stage and higher risk, while the classical PTC (CPTC) had an intermediate TDS, and the follicular variant (FVPTC) maintained a high TDS [12]. These results were corroborated by a subsequent multicenter retrospective study including 6282 cases of PTC [29]. Differential risk patterns of disease recurrence and patient mortality were determined for the three major PTC variants, with increasing aggressiveness from the FVPTC to the CPTC, up to the tall cell PTC (TCPTC) [30]. The results of our study do not confirm a higher recurrence-free probability for FVPTC compared to CPTC, whereas the DFI probability was significantly reduced for the tall cell and sclerosing PTC variants. The partial discrepancy of our findings with the previous ones is probably because, due to the numerical scarcity of relapses in FVPTC, we could not distinguish between encapsulated/well-demarcated noninvasive forms and infiltrative FVPTC.

When in our regression analysis the TNM stage was replaced by stage-related parameters analyzed as independent variables, lymph node metastasis and age dichotomized in <55 years and ≥55 years emerged as the only significant predictors of DFI. Taking advantage of a PTC cohort of patients derived from the TCGA, we sought to verify whether, besides lymph node metastasis and age, the DFI prediction could be improved by additional histological and molecular parameters not available for our case study. The results obtained demonstrated that only a couple of them (BRAF-RAF score and differentiation score) were significantly associated with recurrences, and none of the parameters selected for the Cox regression impacted significantly on the model. Additionally, in this PTC cohort, lateral lymph node involvement represented the best predictor of DFI, with the N1b category (spreading beyond the central compartment) having the highest HR.
