An integrative predictive model for orthokeratology lens decentration based on diverse metrics.

Purpose: To develop a predictive model for orthokeratology (Ortho-K) lens decentration 1 month after wear. Methods: This study included myopic children who were fitted with Ortho-K lenses at Fujian Provincial Hospital between December 2022 and May 2024. Corneal topography parameters and other relevant metrics were collected pre- and post-treatment. Feature selection was conducted using univariate logistic regression and Lasso regression analysis. A machine learning approach was used to develop multiple predictive models, including Decision Tree, Logistic Regression, Multilayer Perceptron, Random Forest, and Support Vector Machine. Model performance was evaluated using accuracy, sensitivity, specificity, ROC curves, DCA curves, and calibration curves. SHAP values were employed to interpret the models. Results: The Logistic Regression model demonstrated the best predictive performance, with an AUC of 0.82 (95% CI: 0.69-0.95), accuracy of 77.59%, sensitivity of 85%, and specificity of 61.11%. The most significant predictors identified were age, 8 mm sag height difference, 5 mm Kx1, and 7 mm Kx2. SHAP analysis confirmed the importance of these features, particularly the 8 mm sag height difference. Conclusions: The Logistic Regression model successfully predicted the risk of Ortho-K lens decentration using key corneal morphological metrics and age. This model provides valuable support for clinicians in optimizing Ortho-K lens fitting strategies, potentially reducing the risk of adverse outcomes and improving the quality of vision for patients. Further validation in clinical settings is recommended.