Development and validation of an interpretable machine learning model for predicting central lymph node metastasis in papillary thyroid cancer.

Objective: To develop and validate a machine learning-based prediction model for central lymph node metastasis (CLNM) in papillary thyroid carcinoma (PTC) patients using routine blood test-derived inflammatory markers and clinical features. Methods: This retrospective study included 1,697 PTC patients. The cohort was randomly divided into training (70%) and validation (30%) sets. Clinical variables and inflammatory markers derived from routine blood tests including neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, lymphocyte-to-monocyte ratio (LMR), and systemic immune-inflammation index (SII) were collected. LASSO regression was applied for feature selection, followed by development of eight machine learning algorithms. The optimal model was selected based on discrimination, calibration, and clinical utility. SHAP analysis was performed to enhance interpretability. Results: CLNM occurred in 516 patients (30.4%). LASSO regression identified 29 predictive features. The Stacking Ensemble model achieved superior performance with AUC of 0.988 in training and 0.923 in validation sets, significantly outperforming traditional logistic regression (AUC: 0.721). SHAP analysis revealed maximum tumor size as the most important predictor, followed by LMR, age, VEGF, and SII. Decision curve analysis demonstrated substantial clinical benefit across threshold probabilities. Conclusion: The machine learning-based model incorporating routine blood test-derived inflammatory markers and clinical features demonstrates excellent performance for CLNM prediction in PTC patients, providing a valuable tool for surgical decision-making and precision medicine approaches.