Machine Learning for Predicting Recurrent Course in Uveitis Using Baseline Clinical Characteristics.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Informatics

Background:

• Uveitis is a complex intraocular inflammatory condition.
• Predicting uveitis recurrence is crucial for effective patient management and risk stratification.
• Current methods for predicting recurrence have limitations.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting the risk of recurrent uveitis.
• To utilize baseline clinical characteristics for risk stratification.
• To inform clinical decision-making in uveitis management.

Main Methods:

• Retrospective analysis of 966 patients with uveitis from the Ocular Autoimmune Systemic Inflammatory Infectious Study registry.
• Training of three ML classifiers (Random Forest, eXtreme Gradient Boosting, RBF-SVC) on baseline data.
• Feature selection via bivariate analysis and model optimization using grid search with cross-validation.

Main Results:

• The Random Forest model achieved the highest accuracy (0.77) with high specificity (0.93) but modest sensitivity (0.44).
• eXtreme Gradient Boosting and RBF-SVC showed comparable accuracies.
• Key predictors identified include vitreous haze, retrolental cells, and noninfectious etiology.

Conclusions:

• ML models, especially Random Forest, show promise in identifying patients at low risk for uveitis recurrence.
• High specificity suggests reliable identification of low-risk individuals.
• Limited sensitivity highlights the ongoing challenge in predicting rare events in a heterogeneous patient population.