Decoding clinical heterogeneity and age-related coronary risk in incomplete Kawasaki disease: a multicenter cluster analysis and structural equation modeling study.

Hierarchical clustering identified four clinically distinct pretreatment subphenotypes in incomplete Kawasaki disease (iKD).

• Analysis was performed on 1,619 iKD patients from six centers using 11 pretreatment clinical and laboratory variables.
• Coronary measures were explicitly excluded from cluster derivation to allow unbiased post-hoc comparison of coronary outcomes.
• C1 showed greater systemic inflammation; C2 was characterized by younger age, longer pretreatment fever duration, and marked thrombocytosis; C3 by prominent hepatobiliary involvement; and C4 by a relatively low-inflammatory profile.
• The clustering framework was also validated by application to the entire KD cohort (n=6,264), and iKD patients were unevenly distributed across the independently derived whole-KD clusters.

Cluster C2, characterized by younger age, longer pretreatment fever duration, and marked thrombocytosis, had the highest coronary artery aneurysm (CAA) prevalence and greatest coronary burden.

• CAA prevalence in C2 was 68.9%, the highest among all four clusters.
• C2 was distinguished by younger age, longer pretreatment fever duration, and marked thrombocytosis.
• The greater post-IVIG coronary burden of C2 remained evident after exclusion of patients whose iKD classification depended on coronary findings in sensitivity analyses.
• Focused sensitivity analyses separately evaluated pre-IVIG coronary abnormalities and newly developed or worsening abnormalities after IVIG.

Cluster C3, characterized by prominent hepatobiliary involvement, had the highest IVIG-resistance rate among the four clusters.

• C3's IVIG-resistance rate was 12.7%, the highest of all clusters.
• C3 was distinguished by prominent hepatobiliary involvement as its defining clinical feature.
• Different clusters were associated with different coronary and treatment-response profiles, demonstrating the clinical utility of the subphenotyping approach.

Younger age, longer pretreatment fever duration, fewer clinical manifestations, lower albumin, lower hemoglobin Z-score, and lower neutrophil ratio were independently associated with CAA.

• These six variables were identified through logistic regression analysis as independent associates of coronary artery aneurysm.
• Fewer clinical manifestations (i.e., incomplete presentation) was among the independent predictors, linking iKD's diagnostic challenge to its coronary risk.
• Lower albumin and lower hemoglobin Z-score suggest a role for nutritional and hematologic status in coronary vulnerability.
• Lower neutrophil ratio was also independently associated with CAA, suggesting that the inflammatory profile composition, not just overall inflammation level, matters.

Structural equation modeling (SEM) demonstrated that age has a strong direct association with CAA and only a small indirect association mediated through symptom count and pretreatment fever duration.

• Exploratory structural equation modeling was used to decompose the pathways by which age relates to CAA.
• The direct association of age with CAA was strong, while the indirect pathway operating through symptom count and pretreatment fever duration was small.
• This finding indicates that younger age's contribution to coronary vulnerability is largely independent of the fact that younger children tend to have longer fever durations or fewer symptoms.
• The authors conclude that 'the coronary vulnerability of younger patients appears multifactorial rather than explained solely by longer pretreatment fever duration.'

LOESS analyses were used to further examine non-linear relationships between continuous factors and coronary artery aneurysm risk.

• LOESS (locally estimated scatterplot smoothing) analysis was employed alongside logistic regression and SEM to examine factors associated with CAA.
• This approach allowed visualization of potentially non-linear associations between continuous clinical variables and coronary outcomes.
• The combination of three analytical methods (logistic regression, SEM, LOESS) provided convergent evidence about CAA risk factors.

iKD patients were unevenly distributed across clusters independently derived from the entire KD cohort, suggesting that iKD does not represent a homogeneous subgroup of KD.

• The same clustering framework applied to the entire KD cohort (n=6,264) produced independent clusters.
• iKD patients (n=1,619) were not uniformly spread across these whole-KD clusters, indicating clinical heterogeneity within iKD itself.
• This finding supports the biological and clinical meaningfulness of subphenotyping iKD separately.
• The multicenter design across six centers supports the generalizability of the cluster structure.