Abstract
Background
Early identification of intravenous immunoglobulin‐resistant Kawasaki disease is important for reducing the risk of coronary artery lesions. We aimed to develop, externally validate, and deploy an interpretable machine‐learning model for early prediction of intravenous immunoglobulin‐resistant Kawasaki disease.
Methods
The derivation cohort included 3023 patients with Kawasaki disease admitted to Children's Hospital of Soochow University from January 2020 to December 2024 and was used for model training and internal validation. External validation included 1632 patients from 3 independent hospitals during the same period. Thirty‐three clinical variables available within 24 hours of admission were used to develop 12 machine‐learning models. Model discrimination was assessed using the area under the curve, and Shapley Additive Explanations were used for model interpretation.
Results
Among the 12 algorithms, the Extra Trees model showed the best discriminative performance. After feature reduction, an interpretable Extra Trees model incorporating 8 variables was selected. The model achieved area under the curve of 0.865 in internal validation, 0.890 in the Anhui cohort (n=654), 0.805 in the Xuzhou cohort (n=574), and 0.853 in the Suqian cohort (n=404). The final model was implemented as a web‐based application for individualized risk estimation.
Conclusions
We developed and externally validated an interpretable machine‐learning model for early prediction of intravenous immunoglobulin‐resistant Kawasaki disease. Shapley Additive Explanations‐based interpretation and web deployment may facilitate individualized risk assessment and clinical decision‐making.