A clinically practical machine learning nomogram for preoperative CLNM prediction in PTMC: tumor-capsule distance and D2–40 evidence

Objective This study aimed to develop and validate a machine learning-driven nomogram for preoperative prediction of central lymph node metastasis (CLNM) in papillary thyroid microcarcinoma (PTMC), with a focus on elucidating the role of tumor–capsule distance (TCD) and its underlying biological mechanism via D2–40 immunohistochemical evidence. Methods A retrospective cohort of 503 PTMC patients was randomly divided into training, validation, and test sets (6:2:2). Feature selection was performed using XGBoost and interpreted via SHAP analysis. A nomogram was constructed based on logistic regression and validated internally and externally (n = 101). D2–40 staining was used to assess lymphatic vessel density (LVD) in relation to TCD. Results Five key predictors were identified: TCD, microcalcifications, age, BMI, and maximum tumor diameter. A TCD 2 mm was strongly associated with CLNM (p 0.0001). The XGBoost model achieved an AUC of 0.900 on the test set, while the nomogram showed AUCs of 0.862, 0.836, and 0.875 on the training, validation, and test sets, respectively, with external validation AUC of 0.870. D2–40 staining confirmed significantly higher LVD in the pericapsular region (p 0.05), supporting the biological plausibility of TCD as a predictor. Conclusion The proposed nomogram demonstrates high predictive accuracy, clinical interpretability, and biological grounding. It serves as a practical tool for individualized risk assessment of CLNM in PTMC, potentially guiding precision treatment decisions.