An interpretable machine learning model for predicting central lymph node metastasis in cN0 T1–T2 papillary thyroid carcinoma: a retrospective study

Introduction This study aimed to develop and validate an interpretable machine learning model for preoperative prediction of central lymph node metastasis (CLNM) in patients with clinically node-negative (cN0) T1–T2 papillary thyroid carcinoma (PTC). Methods A retrospective analysis was conducted on 710 patients (971 lesions), integrating pathological, ultrasound features, thyroid function, and systemic inflammatory indicators. A hierarchical feature selection strategy combining L2 and LASSO regularization was employed to optimize multimodal predictors and reduce overfitting. An explainable gradient boosting decision tree (GBDT) model was constructed and evaluated using calibration curves, decision curve analysis, and SHAP interpretability frameworks. Results The model identified five independent predictors of CLNM: bilateral laterality, tumor size >1.0 cm, age ≤55 years, systemic immune-inflammation index (SII) >449.85, and platelet-to-lymphocyte ratio (PLR) ≤134.88; free triiodothyronine (FT3) was also included as an adjunct variable to enhance performance. The model achieved an AUC of 0.830 (95% CI: 0.773–0.887) in the test set, with robust performance confirmed after correcting for sample overlap (AUC 0.812) and external validation on an independent temporal cohort (n=50, AUC 0.800). The model showed clinical utility across a wide decision threshold range (0–85%). Discussion This multimodal, interpretable prediction tool provides a non-invasive and clinically transparent aid for individualized surgical decision-making in cN0 T1–T2 PTC, bridging endocrine and inflammatory biomarkers with machine learning to advance precision thyroid oncology.