Development of a Machine Learning-based Model for Methimazole Dosage Adjustment in Youth With Hyperthyroidism

Area of Science:

• Endocrinology
• Pediatric Medicine
• Computational Biology

Background:

• Accurate methimazole (MMI) dosage is critical for pediatric hyperthyroidism, but current titration relies on clinical expertise due to a lack of predictive tools.
• Individualized MMI dosing requires validated methods to optimize treatment outcomes in children.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting optimal MMI dosage in pediatric hyperthyroidism.
• To establish a data-driven approach for MMI dose adjustment in young patients.

Main Methods:

• A retrospective, multicenter study involving ML models (linear regression, decision tree, support vector regression, XGBoost, feed-forward neural networks).
• Models were trained on data from 1,512 visits and validated using two external cohorts (666 and 31 visits).
• Predictors included age, sex, anthropometrics, prior MMI dose, treatment duration, and thyroid function tests. Performance was measured by Mean Absolute Error (MAE).

Main Results:

• The eXtreme Gradient Boosting (XGBoost) model achieved the best performance with MAE of 1.72 mg (internal validation) and 1.08 mg (external validation).
• Shapley Additive Explanations (SHAP) analysis identified previous MMI dose, triiodothyronine, and free thyroxine levels as key predictors.

Conclusions:

• This study presents the first data-driven tool to guide methimazole dosing in pediatric hyperthyroidism.
• The developed ML models can enhance clinical efficiency and support individualized MMI titration for children.