Risk assessment model for 2–8-week coronary artery Z score abnormalities in children with KD based on clinical and laboratory predictors available at admission: a multicenter retrospective cohort study

To develop and validate a risk assessment model for 2–8-week coronary artery Z score abnormalities in children with Kawasaki disease (KD) based on clinical and laboratory predictors at admission. A multicenter retrospective cohort study design was used, and children with KD from 3 centers were included. At each examination, zMax was defined as the maximum Z score of the left main coronary artery (LMCA), proximal left anterior descending artery (LAD), and proximal right coronary artery (RCA). The primary outcome was zMax ≥ 2.5 in any available prespecified echocardiographic window during weeks 2–8 after fever onset. Candidate predictors included age, sex, fever duration, KD subtype, and routine laboratory variables at admission. Elastic Net penalized logistic regression was used for modeling, and model performance was evaluated by bootstrap internal validation, leave-one-center-out internal-external validation, calibration assessment, Brier score, decision curve analysis, and sensitivity analyses. A total of 1,163 children were included, of whom 146 met the primary outcome of coronary artery Z score abnormality during 2–8-week follow-up, corresponding to an event rate of 12.55%. Compared with the group without coronary artery Z score abnormalities, the group with coronary artery Z score abnormalities was younger, had a longer fever duration and a higher proportion of incomplete KD, and showed more evident inflammatory response, lower albumin levels, and lower sodium levels. The final model retained all 10 prespecified candidate predictors. The AUC of bootstrap internal validation was 0.781 (95% CI: 0.743–0.819), the calibration slope was 0.964 (95% CI: 0.887–1.052), and the Brier score was 0.092 (95% CI: 0.084–0.101). The weighted pooled AUC of the three rounds of leave-one-center-out validation was 0.780 (95% CI: 0.748–0.812), the calibration slope was 0.952 (95% CI: 0.834–1.085), and the Brier score was 0.093 (95% CI: 0.083–0.103). The AUC range in sensitivity analyses was 0.764–0.785, and decision curve analysis suggested that the model had clinical net benefit within the exploratory threshold probability range of 5%–30%. The risk assessment model based on routine clinical and laboratory predictors available at admission showed acceptable discrimination and calibration in bootstrap internal validation and three-center leave-one-center-out internal–external cross-validation for estimating the risk of 2–8-week coronary artery Z score abnormalities in children with KD. It may provide a reference for early risk stratification and echocardiographic follow-up management, but requires independent external and prospective validation. Not applicable.