Machine Learning Models for Predicting 24-Hour Intraocular Pressure Changes: A Comparative Study

BACKGROUND Predicting 24-hour intraocular pressure (IOP) fluctuations is crucial for enhancing glaucoma management. Traditional methods of measuring 24-hour IOP fluctuations are complex and present certain limitations. The present study leverages machine learning techniques to forecast 24-hour IOP fluctuations based on daytime IOP measurements. MATERIAL AND METHODS A binary method was used to classify 24-hour IOP fluctuations as either >8 mmHg or £8 mmHg. Data were collected from 24-hour IOP monitoring, including 22 different features. Feature selection involved the chi-square test and point-biserial correlation, leading to the establishment of 4 subsets with significance levels of P<1, P<0.1, P<0.05, and P<0.025. Five binary classification machine learning algorithms were used to construct the model. Model performance was assessed by comparing accuracy, specificity, 10-fold cross-validation, precision, sensitivity, F1 score, area under the curve (AUC), and Area Under the Precision-Recall Curve (AUCPR). The model with the highest performance was selected, and feature importance was assessed using Shapley additive explanations. RESULTS In the subset of features where P<0.05, all models performed better than those in the other subsets, with XGBoost standing out the most. The XGBoost algorithm achieved an accuracy of 0.886, a specificity of 0.972, a 10-fold cross-validation of 0.872, a precision of 0.857, a sensitivity of 0.585, and an F1 score of 0.696. Additionally, it obtained an AUC of 0.890 and an AUCPR of 0.794. CONCLUSIONS Our study illustrates the predictive capabilities of machine learning algorithms in forecasting 24-hour IOP fluctuations. The exceptional performance of the XGBoost algorithm in predicting IOP fluctuations underscores its significance for future research and clinical applications.