Multicenter Validation of Machine Learning Models for Predicting Myopia Progression Across School‐Based Screening and Optical Intervention Settings in Chinese Children and Adolescents

Accurate prediction of myopia progression is critical for personalized clinical decision‐making. This study aimed to develop machine‐learning models leveraging ocular biometric parameters to predict myopia status, rapid progression, and spherical equivalent (SE) using multicenter data from 40 445 individuals across 10 hospitals. Data from one hospital were used for training and internal validation, while the remaining data were used for external validation. Models (i.e., XGBoost, LightGBM, Gradient Boosting, Random Forest, Extra Trees, Decision Tree, Logistic Regression, K‐Nearest Neighbors, Naive Bayes, and FT‐Transformer) were evaluated using 12 clinical variables as input features, with hyperparameters automatically optimized via Bayesian optimization. In external validation, the best‐performing models in the school‐based screening group were Logistic Regression for myopia status (AUROC = 0.860) and rapid progression (AUROC = 0.620), and Gradient Boosting for SE prediction ( R 2 = 0.596). In the optical interventions group, XGBoost, Extra Trees, and FT‐Transformer performed best for myopia status (AUROC = 0.969), rapid progression (AUROC = 0.663), and SE prediction ( R 2 = 0.830), respectively. Our study demonstrates accurate prediction of myopia trajectories across school‐based screening and optical intervention settings, enabling clinically precise risk stratification and personalized management to inform timely interventions and help mitigate the escalating global myopia burden.