Abstract / Summary
Abstract Coronary artery lesions (CAL) represent the most severe complication of Kawasaki disease (KD). This study aims to analyze the relationship between serum Interleukin-6 (IL-6) levels and CAL in KD, and develop a predictive model to provide a scientific basis for clinical risk assessment and personalized treatment of KD. Newly diagnosed children with complete KD admitted to Hainan Women and Children’s Medical Center from January 1, 2022 to December 31, 2024 were enrolled. Clinical baseline data and inflammatory indicators of all children were collected, and serum IL-6 levels were detected by microsphere-based flow cytometry. Echocardiography combined with the Canadian Dallaire database Z-score standard was used to diagnose and classify CAL. Restricted cubic spline (RCS) logistic regression, LASSO regression and multivariable logistic regression were applied for nonlinear analysis, predictor selection and model construction, and a nomogram was developed and validated. To further evaluate the temporal generalizability of the prediction model, independent external validation adopted a temporal segmentation dataset, prospectively collecting another complete cohort of KD patients from the same center using identical diagnostic and laboratory testing protocols between January 1, 2025, and May 31, 2026. A total of 160 newly diagnosed children with complete KD were enrolled included in the original cohort and divided into CAL ( n = 32) and non-CAL ( n = 128) groups. The serum IL-6 level (logarithmic transformation) in the CAL group was significantly higher than that in the non-CAL group (4.72 ± 1.01 vs. 3.78 ± 1.10, P < 0.001). Nonlinear analysis revealed an S-shaped association between IL-6 and CAL risk with two inflection points at 35.8 pg/ml and 103.3 pg/ml; within the range of 35.8 ~ 103.3 pg/ml, a two-fold increase in IL-6 was associated with a 9.55-fold increase in CAL risk (95%CI:1.24 ~ 73.41, P = 0.030), and no significant association was observed outside this range. C-reactive protein (CRP) showed a U-shaped association with CAL risk with an inflection point at 32 mg/L. A total of 110 children were included in the temporal segmentation cohort. The final model constructed with IL-6 (nonlinear), CRP (nonlinear) and Hb as predictors had the optimal predictive performance for CAL, with an AUC of 86.5%, a sensitivity of 81.3% and a specificity of 85.9% in original cohort (In the temporal segmentation dataset, these values were 0.652, 0.636, and 0.717, respectively). The nomogram established based on this model had good calibration (Hosmer-Lemeshow P = 0.09, Brier score = 0.109) in original cohort, but not well in the temporal segmentation dataset (Hosmer-Lemeshow P < 0.001, Brier score = 0.127). And DCA confirmed its favorable clinical net benefit and practical application value. DCA analysis results show good clinical net benefit and practical application value in both the original cohort and the temporal segmentation dataset. IL-6 is a critical factor for CAL in children with KD with a distinct nonlinear dose-response relationship, and the prediction model constructed with IL-6, CRP and Hb can assist clinicians in early identification of children at high risk for CAL, providing a scientific basis for individualized clinical management of KD.