Machine learning for screening and predicting the risk of developing uveitis in juvenile idiopathic arthritis.

OBJECTIVE

Establishing a predictive model using clinical indicators for the early identification of JIA-U.

METHOD: A cross-sectional study was conducted with 255 patients admitted at Beijing Children's Hospital between 2018 and 2023. The model was fitted using stepwise logistic regression as well as least absolute shrinkage and selection operator (LASSO) regression. Calibration and decision curve analysis were used for validation.

RESULTS

The final predictive model included four clinical variables (patient's gender, age at onset, arthritis subtype, and ANA status). A nomogram for risk prediction was developed, which demonstrated good discrimination in both the training cohort (AUC = 0.8417; 95% CI = 0.775-0.9085) and the testing cohort (AUC = 0.782; 95% CI = 0.6752-0.8884). Calibration curves showed that, through bootstrap resampling, the nomogram performed well in predicting the occurrence of uveitis in JIA.

CONCLUSION

This study established a predictive model using routine clinical indicators to assess the risk of uveitis in JIA patients.