Decision Tree Modeling to Predict Myopia Progression in Children Treated with Atropine: Toward Precision...

Area of Science:

• Ophthalmology
• Machine Learning
• Pediatric Health

Background:

• Myopia is a significant global health issue, particularly in East Asian children.
• Topical atropine is used for pediatric myopia control, but treatment responses vary significantly.
• Predicting individual responses to atropine is crucial for effective myopia management.

Purpose of the Study:

• To develop an interpretable machine learning model for predicting individualized responses to topical atropine in pediatric myopia control.
• To identify key risk factors influencing myopia progression despite treatment.
• To support personalized clinical decisions in myopia management.

Main Methods:

• Retrospective observational study of 1545 pediatric eyes treated with topical atropine over 3 years.
• Classification and Regression Tree (CART) model used to predict changes in spherical equivalent (SE).
• Comparison of CART model performance with Least Absolute Shrinkage and Selection Operator (Lasso) regression.

Main Results:

• Both CART and Lasso models showed comparable predictive performance.
• Baseline spherical equivalent (SE) was identified as the primary determinant of myopia progression.
• Children with baseline SE more negative than -3.125 D showed greater progression, especially with longer treatment and higher atropine dosage.

Conclusions:

• Baseline SE is the critical factor influencing myopia progression.
• Explainable machine learning, through CART decision rules, can enhance precision myopia management.
• The study supports individualized treatment strategies for pediatric myopia control.