Abstract
BACKGROUND: Axial spondyloarthritis (axSpA) is a chronic inflammatory disease primarily affecting the spine and sacroiliac joints, with anterior uveitis (AU) as a common extra-articular manifestation. Predicting AU onset in axSpA patients is challenging, as traditional statistical methods often fail to capture the disease's complexity. METHODS: This study aimed to develop an interpretable machine learning (ML) model to predict AU onset in axSpA patients through a historical cohort analysis of 1508 patients from a tertiary medical center. Clinical data involving 54 variables were preprocessed through imputation, factorization, oversampling, outlier capping, and standardization. Recursive feature elimination identified 12 key predictors. Subsequently, 10 ML algorithms were assessed using performance metrics and visualization techniques. RESULTS: The gradient boosting machine model incorporating 12 key factors showed high accuracy in predicting AU risk. Shapley additive explanations analysis revealed that hip involvement, nonsteroidal anti-inflammatory drug use, and smoking were the most influential predictors. The model's interpretability provided clear insights into the contribution of each feature to AU risk, supporting early diagnosis and personalized treatment. CONCLUSION: The gradient boosting machine model predicts AU risk in axSpA patients, helping identify high-risk cases for early intervention and personalized treatment to prevent complications such as vision loss.