Comprehensive Summary
This study assessed a classification and regression tree (CART) machine learning model’s analysis of myopic eyes treated with topical atropine. The CART model’s performance was compared to least absolute shrinkage and selection operator regression (Lasso), which is a tool that is already used in many ophthalmology studies. Preexisting data of 1545 myopic eyes was collected and used to construct the model, and analysis was repeated many times in order to conclude the highest quality model. After both the CART and Lasso models were used to predict SE progression in the sample data provided, the results of both models were similar, with Lasso only slightly outperforming CART by demonstrating smaller mean squared error, root mean squared error, and mean absolute error. The CART model concluded that higher baseline myopia (specifically higher than -3.125 D) leads to more severe myopia progression in the future, which is consistent with existing studies. The similarity of CART’s analysis to Lasso suggests its compatibility in clinical settings to be used as a guide to predict SE progression in children with myopia.
Outcomes and Implications
The artificial intelligence CART model developed by Chen et. al. serves as a potentially useful tool in ophthalmology clinics to create highly specific predictions of disease progression in children with myopia treated with topical atropine. The model’s analysis that higher baseline SE leads to more severe myopia that is difficult to treat even with therapy highlights the importance of treating myopia early. The CART model’s consistency with the existing Lasso tool suggests its future utility in more AI-assisted clinical environments.