Machine learning-based predictive model for hungry bone syndrome following parathyroidectomy in secondary...

Area of Science:

• Endocrinology
• Medical Informatics
• Machine Learning

Background:

• Secondary hyperparathyroidism (SHPT) often requires parathyroidectomy.
• Hungry bone syndrome (HBS) is a potential complication following parathyroidectomy.
• Accurate prediction of HBS risk is crucial for patient management.

Purpose of the Study:

• To develop an interpretable machine learning model for predicting HBS risk in SHPT patients after parathyroidectomy.
• Identify key predictors of HBS post-parathyroidectomy.
• Provide a tool for estimating HBS risk.

Main Methods:

• Retrospective analysis of 181 SHPT patients undergoing parathyroidectomy.
• Utilized logistic regression and Boruta algorithm to select five key predictors from 46 variables.
• Trained and evaluated seven machine learning models, including XGBoost, using ROC curves, calibration curves, and DCA.
• Quantified model interpretability using SHapley Additive exPlanations (SHAP).

Main Results:

• The XGBoost model achieved an AUC of 0.878 and an F1 score of 0.871 in the validation cohort.
• Identified key predictors: preoperative parathyroid hormone (Pre-PTH), %PTH decay, alkaline phosphatase, serum calcium, and age.
• Developed a web application for HBS risk estimation.

Conclusions:

• An interpretable machine learning model effectively predicts HBS risk in SHPT patients post-parathyroidectomy.
• The model aids in guiding postoperative surveillance strategies.
• Early identification of high-risk patients can improve clinical outcomes.