Recurrence Risk Evaluation in Patients with Papillary Thyroid Carcinoma: Multicenter Machine Learning Evaluation of Lymph Node Variables.

Background: Lymph node (LN)-related risk factors have been updated to predict long-term outcomes in patients with papillary thyroid carcinoma (PTC). However, those factors’ analytic appropriateness and general applicability must be validated. This study aimed to assess LN-related risk factors, and suggest new LN-related risk categories. Methods: This multicenter observational cohort study included 1232 patients with PTC with N1 disease treated with a total thyroidectomy and neck dissection followed by radioactive iodine remnant ablation. Results: The median follow-up duration was 117 months. In the follow-up period, structural recurrence occurred in 225 patients (18.3%). Among LN-related variables, the presence of extranodal extension (p < 0.001), the maximal diameter of metastatic LN foci (p = 0.029), the number of retrieved LNs (p = 0.003), the number of metastatic LNs (p = 0.003), and the metastatic LN ratio (p < 0.001) were independent risk factors for structural recurrence. Since these factors showed a nonlinear association with the hazard ratio of recurrence-free survival (RFS) rates, we calculated their optimal cutoff values using the K-means clustering algorithm, selecting 0.2 cm and 1.1 cm for the maximal diameter of metastatic LN foci, 4 and 13 for the number of metastatic LN, and 0.28 and 0.58 for the metastatic LN ratio. The RFS curves of each subgroup classified by these newly determined cutoff values showed significant differences (p < 0.001). Each LN risk group also showed significantly different RFS rates from the others (p < 0.001). Conclusions: In PTC patients with an N1 classification, our novel LN-related risk estimates may help predict long-term outcomes and design postoperative management and follow-up strategies. After further validation studies based on independent datasets, these risk categories might be considered when redefining risk stratification or staging systems.