Abstract
OBJECTIVE: Coronary artery lesions (CAL) represent the most severe complication of Kawasaki disease (KD). Currently, there is no standardized method for predicting CAL in KD, and the predictive effectiveness varies among different KD patients. Therefore, our study aims to establish distinct predictive models for CAL complications based on the characteristics of different clusters. METHODS: We employed principal component clustering analysis to categorize 1,795 KD patients into different clustered subgroups. We summarized the characteristics of each cluster and compared the occurrence of CAL components within each cluster. Additionally, we utilized LASSO analysis to further screen for factors associated with CAL. We then constructed CAL predictive models for each subgroup using the selected factors and conducted preliminary validation and assessment. RESULTS: Through PCA analysis, we identified three clusters in KD. We developed predictive models for each of the three clusters. The AUCs of the three predictive models were 0.789 (95% CI: 0.732-0.845), 0.894 (95% CI: 0.856-0.932), and 0.773 (95% CI: 0.727-0.819), respectively, all demonstrating good predictive performance. CONCLUSION: Our study identified the existence of three clusters among KD patients. We developed KD-related CAL predictive models with good predictive performance for each cluster with distinct characteristics. This provides reference for individualized precision treatment of KD patients and aids in the health management of coronary arteries in KD.