A machine learning prediction model and online calculator for postoperative recurrence of secondary hyperparathyroidism: A dual-center development and validation study.

BACKGROUND: Secondary hyperparathyroidism carries a high recurrence risk after parathyroidectomy (PTX), requiring early identification of high-risk patients. Using a two-center cohort, we developed a machine learning prediction model and deployed it as an online calculator. METHODS: We included 391 SHPT patients undergoing PTX at two hospitals. Cohort 1 was split 7:3 into training and internal validation sets; cohort 2 served as external validation. Feature selection used LASSO and Boruta, SMOTE handled imbalance, and six models (random forest, XGBoost, etc.) were built. After cross-validation and grid search, the best model was chosen by AUC and F1, interpreted with SHAP, and deployed online. RESULTS: Six predictors were identified: preoperative phosphorus, bone pain score, surgical method, total parathyroid volume, and iPTH at postoperative months 1 and 3. The random forest model performed best (internal validation AUC 0.890). External validation showed AUC 0.889. Early postoperative iPTH was the most important predictor.Online calculator to get the address: https://lhssniegkalmhjphs9exbz.streamlit.app/. CONCLUSION: We developed and dual-center validated a robust SHPT recurrence prediction model. The online calculator enables convenient individualized risk assessment, optimizing postoperative monitoring.