Risk for ocular hypertension progression to early glaucoma: A predictive model and key predictors

Abstract

Background: Ocular hypertension (OHT) is the most significant risk factor for glaucoma. We aimed to develop a model for predicting OHT progression to early glaucoma and to identify key predictors. Methods: Patients with OHT with at least two follow-up visits within 3 years were categorized into non-progressive and progressive groups based on optic nerve morphology and/or functional changes during follow-up. Data were split into training and testing sets (8:2 ratio). Least absolute shrinkage and selection operator regression and logistic regression were used to select predictors. Machine learning models were constructed using the selected predictors as input and evaluated using area under the receiver operating characteristic (AUC) and precision-recall (AP) curve values. The optimal model was further evaluated using 10-fold cross-validation and a validation set. The Shapley additive explanations method was applied to interpret the predictors. Results: Overall, 395 eyes from 395 patients were included (non-progressive: n = 295; progressive: n = 100). The random forest model outperformed all others, achieving AUC values of 0.881 (95 % confidence interval [CI]: 0.835–0.926) in the training set and 0.937 (95 % CI: 0.884–0.991) in the testing set. In the independent validation set (n = 82), the AUC and AP values were 0.865 (95 % CI: 0.782–0.947) and 0.707, respectively. Key predictors were baseline intraocular pressure, rim area, ganglion cell-inner plexiform layer (GCIPL) inferior-temporal thickness difference, and GCIPL superior-temporal thickness difference. Family history and male sex also contributed. Conclusions: The developed model could effectively predict the risk for OHT progression to early glaucoma and may aid devising individualized treatment plans and follow-up protocols.

Keywords

• Ocular hypertension
• Predictive model
• Early glaucoma
• Risk stratification
• OCT