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

Area of Science:

• Ophthalmology
• Medical Informatics
• Machine Learning

Background:

• Predicting 24-hour intraocular pressure (IOP) fluctuations is vital for effective glaucoma management.
• Traditional 24-hour IOP monitoring methods are complex and have limitations.
• Machine learning offers a novel approach to forecast IOP fluctuations.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting 24-hour IOP fluctuations.
• To identify key features for accurate IOP fluctuation prediction.
• To assess the performance of different machine learning algorithms in this task.

Main Methods:

• A binary classification approach was used to categorize IOP fluctuations (>8 mmHg or ≤8 mmHg).
• Data from 24-hour IOP monitoring, including 22 features, were analyzed.
• Feature selection utilized chi-square tests and point-biserial correlation, with significance levels P<1, P<0.1, P<0.05, and P<0.025.
• Five binary classification algorithms were employed, with performance evaluated using accuracy, specificity, precision, sensitivity, F1 score, AUC, and AUCPR.
• Shapley additive explanations (SHAP) were used for feature importance assessment.

Main Results:

• Models using features with P<0.05 demonstrated superior performance compared to other subsets.
• The XGBoost algorithm achieved the highest performance metrics.
• XGBoost achieved an accuracy of 0.886, specificity of 0.972, precision of 0.857, sensitivity of 0.585, F1 score of 0.696, AUC of 0.890, and AUCPR of 0.794.
• Feature importance analysis was conducted using SHAP values.

Conclusions:

• Machine learning algorithms can effectively predict 24-hour IOP fluctuations.
• The XGBoost algorithm shows significant promise for clinical application in glaucoma management.
• This predictive capability can enhance patient care and treatment strategies for glaucoma.