Machine-learning model predicts coronary artery lesions in Kawasaki disease

Photo by Dan Meyers / Unsplash

Frontiers in Medicine Published April 23, 2026 Medically reviewed April 24, 2026 Study authors: Xuemei Li, Zihan Zhou, Jingyi Fan, Lin Zhao, Ruidi Xu, Dong Li, Xu Ma, Lu Sun, Yujian Wu, Zhouping W… DOI ↗ By Dr. Amelia Tan, PhD · Internal Medicine & Chronic Disease

Key Takeaway

Consider that a CatBoost machine-learning model shows promise for predicting coronary artery lesions in Kawasaki disease, but requires prospective validation.

This retrospective cohort study conducted at Shengjing Hospital of China Medical University aimed to develop a machine-learning algorithm (CatBoost) to predict coronary artery lesions (CAL) in patients with Kawasaki disease. The study collected 41 demographic, clinical, and laboratory parameters, though the exact number of patients was not reported.

In the training set, the model achieved an AUC of 0.953, sensitivity of 0.908, specificity of 0.860, and accuracy of 0.883. In internal validation, performance was slightly lower: AUC 0.874, sensitivity 0.721, specificity 0.848, accuracy 0.837. External validation showed an AUC of 0.876, sensitivity 0.894, and specificity 0.954.

Safety and tolerability were not reported, as this was a predictive modeling study without an intervention. Key limitations include the retrospective design, single-center setting, and unreported sample size, which may limit generalizability.

Clinicians should interpret these results cautiously. The model may aid in creating personalized treatment strategies, but prospective validation in diverse populations is needed before clinical implementation.

Study Details

Study typeCohort

EvidenceLevel 3

PublishedApr 2026

View Original Abstract ↓

This study aimed to analyze the risk factors of coronary artery lesions (CAL) in patients with Kawasaki disease (KD) and establish predictive models for CAL in patients with KD. This retrospective cohort study included KD patients admitted to Shengjing Hospital of China Medical University, collecting data on 41 demographic, clinical, and laboratory parameters. LASSO regression identified key predictive variables. The dataset was split into 70% training and 30% validation. Ten models were trained using 10-fold cross-validation, with the training set balanced through ROSE oversampling. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. The CatBoost algorithm achieved the best results: AUC, 0.953; sensitivity, 0.908; specificity, 0.860; and accuracy, 0.883. Internal validation results were as follows: AUC, 0.874; sensitivity, 0.721; specificity, 0.848; accuracy, 0.837. External validation results were as follows: AUC, 0.876.sensitivity, 0.894; specificity, 0.954. We present a machine-learning model that predicts the risk of CAL in patients with KD in China, aiding doctors in creating personalized treatment strategies to improve outcomes.