Nomogram for postoperative pathological stratification of ≥0.2 cm clinically significant lymph node metastasis burden in pN1 papillary thyroid carcinoma: development and validation using L2 ridge penalized logistic regression.

Background: The maximum diameter of metastatic lymph node (LNM) foci serves as a core indicator for individualized postoperative management of papillary thyroid carcinoma (PTC). The 2015 ATA guidelines adopt 0.2 cm as a cutoff value to distinguish clinically significant lymph node metastasis (LNM, ≥0.2 cm) from LNM micrometastasis (<0.2 cm), with distinct follow-up regimens and adjuvant radioactive iodine (¹³¹I) therapy indications for the two groups. Existing predictive models only judge the presence or absence of preoperative LNM, lacking dedicated risk stratification tools for pathologically confirmed pN1 patients to quantify nodal metastatic tumor burden. Conventional maximum-likelihood logistic regression is prone to complete data separation and extreme odds ratio (OR) inflation when analyzing low-prevalence pathological markers such as extranodal extension (ENE). To resolve this methodological limitation, we developed a predictive nomogram based on L2 ridge penalized logistic regression using postoperative paraffin-embedded pathological data only. Objective: To screen independent pathological predictors of ≥0.2 cm clinically significant LNM in patients with classical pN1 PTC, construct a visualized postoperative risk stratification nomogram, and improve the statistical robustness of predictive models built on sparse pathological datasets via L2 ridge penalized logistic regression. Methods: This single-center retrospective study enrolled 212 patients with pathologically confirmed classical pN1 PTC who underwent surgical resection from January 2021 to December 2025. Subjects were divided into a micrometastasis group (n = 92, maximum metastatic lesion diameter< 0.2 cm) and a clinically significant LNM group (n = 120, maximum metastatic lesion diameter ≥ 0.2 cm). All 13 clinicopathological indicators were included in univariate L2 ridge penalized logistic regression; variables with P< 0.01 were entered into the multivariate model. The optimal regularization parameter C = 0.05 was selected by combining ridge trace plots and 10-fold cross-validation, and variance inflation factor (VIF) analysis was performed to detect multicollinearity. Model performance was comprehensively evaluated by ROC curves, AUC value, Hosmer-Lemeshow goodness-of-fit test, calibration curves and decision curve analysis (DCA). A total of 1,000 bootstrap resamples were adopted for internal validation, and leave-one-variable-out sensitivity analysis was conducted to evaluate model robustness. Results: LNmet, ENE, thyroid capsular invasion (TCI), intrathyroidal dissemination (ITD), primary tumor size, and age were significant univariate predictors, yet age lost predictive value after adjustment. Adjusted ORs were LNmet (3.02), ENE (2.94), TCI (1.83), ITD (1.75), and tumor size (1.33). The model achieved an AUC of 0.839 (Hosmer-Lemeshow P = 0.155). Bootstrap validation confirmed stable coefficients without extreme OR inflation. DCA showed net clinical benefit across the 0.2-0.8 threshold range; however, these findings do not support extended lymphadenectomy. Conclusion: This L2 ridge nomogram effectively stratifies postoperative ≥0.2 cm LNM risk and guides personalized surveillance and ¹³¹I therapy. It is not intended for preoperative evaluation or surgical planning, and its generalizability requires external multicenter validation.