To establish and validate short-term and long-term prediction models for the therapeutic outcomes of orthokeratology (OK) lenses in children, incorporating baseline ocular parameters and dynamic changes in axial length (AL) as predictive variables. This retrospective cohort study included 896 pediatric patients who underwent OK lenses treatment at Tianjin Medical University Eye Hospital from January 2020 to July 2025, with a minimum follow-up period of one year. Baseline demographic and ocular parameters were collected, and AL changes were introduced as dynamic predictors in models starting from year two. Seven machine learning algorithms were used to construct prediction models, including LightGBM, random forest, support vector machine (SVM), and artificial neural network (ANN). Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), F1 score, sensitivity, specificity, calibration, and decision curve analyses. Model interpretability was explored using SHAP and LIME analyses. Among the 896 included children, 566 achieved reasonable AL control, and 330 showed poor control during the first year. Baseline AL, corneal eccentricity parameters, and refractive error were significantly associated with treatment outcomes. The optimal second-year prediction model (LightGBM) achieved an area under the receiver operating characteristic curve (AUC) of 0.96 with an F1 score of 0.86. For long-term prediction, the SVM model demonstrated moderate and stable performance, with F1 scores of 0.73 and 0.75 in the third- and fourth-year models, respectively. Prediction models based on baseline parameters and dynamic AL changes can effectively estimate both short-term and long-term efficacy of OK treatment in children. Dynamic AL change in the first year is a robust predictor of long-term outcomes, offering potential for individualized myopia management strategies.
Wang et al. (Tue,) studied this question.