The Postoperative Hyperopic Shift Risk Prediction Model for Primary Angle Closure Glaucoma Patients Based on Machine Learning.

PRCIS

This study developed and validated a machine learning-based risk prediction model to estimate the likelihood of postoperative hyperopic shift in patients with primary angle closure glaucoma after IOL implantation, which may help guide individualized surgical decision-making.

OBJECTIVE

This study aims to construct a predictive model for the risk of hyperopic shift (HS) after phacoemulsification combined with intraocular lens (PE+IOL) surgery in patients with primary angle closure glaucoma (PACG), with the goal of providing scientific evidence for personalized treatment and early warning.

MATERIALS AND METHODS

This is a retrospective cohort study that included PACG patients who underwent PE+IOL surgery between June 2019 and June 2024, according to predefined inclusion and exclusion criteria. We collected patients' demographic information, preoperative ocular examination data, and refractive changes 3-6 months postsurgery. The Boruta algorithm was used for feature selection of all clinical variables, and various machine learning models, including logistic regression (LR), random forest, support vector machine (SVM), k-nearest neighbors (KNN), and XGBoost, were developed. Model performance was assessed using receiver operating characteristic (ROC) curves and the area under the curve (AUC). The best-performing model was selected for visualization and interpretability analysis. Data processing and analysis were performed using R version 4.2.3. All statistical tests were 2-sided, with a P -value <0.05 considered statistically significant.

RESULTS

A total of 423 eyes were included, with n=267 in the non-HS group and n=156 in the HS group. Key predictive variables identified by the Boruta algorithm included target refraction, preoperative best-corrected visual acuity (BCVA), axial length (AL), central corneal thickness (CCT), anterior chamber depth (ACD), lens thickness (LT), white-to-white distance (W2W), and pupil diameter (P). Both the SVM and LR models exhibited the best predictive accuracy, with AUCs of 0.704 and 0.696, respectively, demonstrating moderate classification ability.

CONCLUSION

This study successfully developed a risk prediction model for HS after PE+IOL surgery in PACG patients based on various clinical features. The SVM and LR models show promising clinical application in predicting HS risk and can provide personalized postoperative management strategies for glaucoma surgery patients.