Development and validation of a machine learning model for predicting high-risk distant metastatic recurrence in differentiated thyroid cancer

Background Distant metastatic recurrence significantly impacts the prognosis of patients with differentiated thyroid cancer (DTC). Current risk stratification systems have limited accuracy in predicting high-risk distant metastatic recurrence. Objective This study aimed to develop and validate a machine learning model for predicting high-risk distant metastatic recurrence in DTC patients. Methods We retrospectively analyzed 1,245 DTC patients treated between January 2020 and December 2024. Patients were randomly divided into training ( n = 871) and validation ( n = 374) sets. Forty-two clinical, pathological, molecular, and treatment-related variables were collected. LASSO regression was used for feature selection. Six machine learning algorithms (Random Forest, Support Vector Machine, XGBoost, Logistic Regression, K-Nearest Neighbors, and Decision Tree) were employed to build prediction models. Model performance was evaluated using AUC, accuracy, sensitivity, specificity, and F1-score. Calibration was assessed using calibration curves, and clinical utility was evaluated using decision curve analysis. Results During a median follow-up of 72 months, 126 patients (10.1%) developed distant metastatic recurrence. LASSO regression identified eight predictors: age, tumor size, extrathyroidal extension, lymph node metastasis, BRAF V600E mutation, postoperative stimulated thyroglobulin (sTg) level, radioactive iodine dose, and TNM stage. The XGBoost model demonstrated the best performance, with an AUC of 0.88 (95% CI, 0.83–0.93) in the validation set. Patients were stratified into low-risk (recurrence rate: 1.7%), intermediate-risk (14.4%), and high-risk (64.1%) groups with significantly different distant metastasis-free survival ( p 0.001). The XGBoost model showed good calibration and superior clinical utility compared to the TNM staging system. Conclusion We developed and validated an XGBoost-based machine learning model that accurately predicts high-risk distant metastatic recurrence in DTC patients. This model may help clinicians identify patients who could benefit from more aggressive treatment and intensive follow-up, enabling personalized management strategies.