Objective This study aimed to develop and validate a predictive model integrating multidimensional clinical indicators and laboratory biomarkers for the early and precise assessment of coronary artery lesion (CAL) risk in Kawasaki disease. Methods A retrospective analysis was conducted on 320 children with Kawasaki disease admitted between January 2020 and June 2024. The sample size was calculated using PASS 2021 software (α = 0.05, 1-β = 80%), with post-hoc power exceeding 85%. Patients were randomly allocated into a training set (n = 224) and a validation set (n = 96) in a 7:3 ratio. In the training set, univariate analysis identified CAL-associated indicators (P < 0.05). Least Absolute Shrinkage and Selection Operator (LASSO) regression was used for variable compression and feature selection. The core variables were incorporated into a multivariate logistic regression to identify independent risk factors. Based on these factors, three machine learning models—Gradient Boosting Machine (GBM), Support Vector Machine (SVM), and Convolutional Neural Network (CNN)—were constructed. Model performance was evaluated using the AUC, calibration curves, and DCA. Interpretability was analyzed using SHapley Additive exPlanations (SHAP) values. Results No significant differences in baseline characteristics were observed between the training and validation sets (P > 0.05). Univariate analysis identified seven significant indicators. LASSO regression refined these to five core variables. Multivariate logistic regression confirmed all five as independent risk factors for CAL (P < 0.05). Among the machine learning models, the GBM model demonstrated superior performance, with AUC of 0.872 (95% CI: 0.798–0.945) in the training set and 0.790 (95% CI: 0.623–0.957) in the validation set. Calibration curves were closest to the ideal line. DCA indicated net clinical benefit across a wide threshold probability range. SHAP analysis provided intuitive prediction interpretability. Conclusion An integrated clinical-biomarker model was developed and validated for the early prediction of CAL in Kawasaki disease. The model demonstrates good discriminative ability, calibration, and clinical utility. and provides a promising tool for the early identification of high-risk children and the optimization of intervention strategies. Clinical trial number Not applicable.

A clinical-biomarker fusion model for risk prediction of coronary artery lesions in Kawasaki disease

期刊论文
出版模式：

开放获取
发表日期：

2026年3月31日

查看全文

Qiongdan Zhao,
Yonghong Miao

摘要

Objective

This study aimed to develop and validate a predictive model integrating multidimensional clinical indicators and laboratory biomarkers for the early and precise assessment of coronary artery lesion (CAL) risk in Kawasaki disease.

Methods

A retrospective analysis was conducted on 320 children with Kawasaki disease admitted between January 2020 and June 2024. The sample size was calculated using PASS 2021 software (α = 0.05, 1-β = 80%), with post-hoc power exceeding 85%. Patients were randomly allocated into a training set (n = 224) and a validation set (n = 96) in a 7:3 ratio. In the training set, univariate analysis identified CAL-associated indicators (P < 0.05). Least Absolute Shrinkage and Selection Operator (LASSO) regression was used for variable compression and feature selection. The core variables were incorporated into a multivariate logistic regression to identify independent risk factors. Based on these factors, three machine learning models—Gradient Boosting Machine (GBM), Support Vector Machine (SVM), and Convolutional Neural Network (CNN)—were constructed. Model performance was evaluated using the AUC, calibration curves, and DCA. Interpretability was analyzed using SHapley Additive exPlanations (SHAP) values.

Results

No significant differences in baseline characteristics were observed between the training and validation sets (P > 0.05). Univariate analysis identified seven significant indicators. LASSO regression refined these to five core variables. Multivariate logistic regression confirmed all five as independent risk factors for CAL (P < 0.05). Among the machine learning models, the GBM model demonstrated superior performance, with AUC of 0.872 (95% CI: 0.798–0.945) in the training set and 0.790 (95% CI: 0.623–0.957) in the validation set. Calibration curves were closest to the ideal line. DCA indicated net clinical benefit across a wide threshold probability range. SHAP analysis provided intuitive prediction interpretability.

Conclusion

An integrated clinical-biomarker model was developed and validated for the early prediction of CAL in Kawasaki disease. The model demonstrates good discriminative ability, calibration, and clinical utility. and provides a promising tool for the early identification of high-risk children and the optimization of intervention strategies.

Clinical trial number

Not applicable.