Subgroup analysis of patients with Kawasaki Disease based on blood tests and cytokine profiles.

BACKGROUND

We aimed to identify distinct disease patterns and improve the understanding of treatment responses in Kawasaki disease (KD). Patients were classified into subgroups using pre-intravenous immunoglobulin (IVIG) blood tests and cytokine profiles.

METHODS

We retrospectively analyzed 212 KD patients admitted between November 2008 and April 2015. Principal component analysis and hierarchical clustering used 20 variables including age, blood tests, and nine cytokines.

RESULTS

Five subgroups were identified: Cluster 1 (Toddler Favorable Outcome), Cluster 2 (Severe Vasculitis), Cluster 3 (Moderate Vasculitis), Cluster 4 (Infant Favorable Outcome), and Cluster 5 (Infant Cytokine Storm). Cluster 4 demonstrated the highest IVIG response rate (87%), whereas Clusters 2 and 5 showed significantly lower IVIG response rates (20% and 21%, respectively). The acute-phase peak CAL occurrence was highest in Cluster 5 (53%; odds ratio 10.00, 95% CI: 3.14-31.88), compared to 45% in Cluster 2, 10% in Cluster 1, and 8.9% in Cluster 4.

CONCLUSION

Two of the five subgroups showed increased risk of IVIG resistance and CAL development. A logistic regression model incorporating age and monocyte count achieved an AUC of 0.897 for identifying the highest-risk subgroup, suggesting its potential utility as a practical screening tool in routine clinical practice.

IMPACT: Five distinct Kawasaki disease subgroups were identified using pre-intravenous immunoglobulin (IVIG) blood tests and cytokine profiles. Two high-risk subgroups were identified, both associated with IVIG resistance and coronary artery lesion (CAL) development: toddlers with highly elevated inflammatory markers and infants with cytokine storm despite modest elevation of standard inflammatory markers. In the infant cytokine storm subgroup, the acute-phase CAL occurrence was 53% (odds ratio 10.00), the highest among all subgroups. Conventional risk stratification tools, including the Kobayashi score, may fail to identify these high-risk infants. A prediction model incorporating age and monocyte count achieved AUC of 0.897, supporting its utility.