Subgroups of children with Kawasaki disease: a data-driven cluster analysis

Abstract

Background: Although Kawasaki disease is commonly regarded as a single disease entity, variability in clinical manifestations and disease outcome has been recognised. We aimed to use a data-driven approach to identify clinical subgroups.

Methods: We analysed clinical data from patients with Kawasaki disease diagnosed at Rady Children's Hospital (San Diego, CA, USA) between Jan 1, 2002, and June 30, 2022. Patients were grouped by hierarchical clustering on principal components with k-means parcellation based on 14 variables, including age at onset, ten laboratory test results, day of illness at the first intravenous immunoglobulin infusion, and normalised echocardiographic measures of coronary artery diameters at diagnosis. We also analysed the seasonality and Kawasaki disease incidence from 2002 to 2019 by subgroup. To explore the biological underpinnings of identified subgroups, we did differential abundance analysis on proteomic data of 6481 proteins from 32 patients with Kawasaki disease and 24 healthy children, using linear regression models that controlled for age and sex.

Findings: Among 1016 patients with complete data in the final analysis, four subgroups were identified with distinct clinical features: (1) hepatobiliary involvement with elevated alanine transaminase, gamma-glutamyl transferase, and total bilirubin levels, lowest coronary artery aneurysm but highest intravenous immunoglobulin resistance rates (n=157); (2) highest band neutrophil count and Kawasaki disease shock rate (n=231); (3) cervical lymphadenopathy with high markers of inflammation (erythrocyte sedimentation rate, C-reactive protein, white blood cell, and platelet counts) and lowest age-adjusted haemoglobin Z scores (n=315); and (4) young age at onset with highest coronary artery aneurysm but lowest intravenous immunoglobulin resistance rates (n=313). The subgroups had distinct seasonal and incidence trajectories. In addition, the subgroups shared 211 differential abundance proteins while many proteins were unique to a subgroup.

Interpretation: Our data-driven analysis provides insight into the heterogeneity of Kawasaki disease, and supports the existence of distinct subgroups with important implications for clinical management and research design and interpretation.

Funding: US National Institutes of Health and the Irving and Francine Suknow Foundation.

Figure 1.. Hierarchical clustering on principal components

(A) Three-dimensional map of hierarchical clustering on principal components. (B) The optimal number of clusters determined by the calculated inertia. (C) Dendrogram of hierarchical map and identified clusters.

Figure 2.. Distribution of the Kawasaki disease subgroups

(A) Monthly cumulative numbers (bars) and proportions (lines) of new cases in each phenotypic subgroup in 2002–19. Error bars denote 95% CIs. (B) Yearly numbers (bars) and proportions (lines) of new cases in each phenotypic subgroup in 2002–21. Highlighted are the years affected by the COVID-19 pandemic.

Figure 3.. Proteomic analysis

(A) Venn diagram showing the numbers of Kawasaki disease-associated differentially abundant proteins (false discovery rate <0·05) that were shared among the subgroups or unique to a specific subgroup of patients with Kawasaki disease, comparing Kawasaki disease and the control group of 24 healthy children. (B) Selected subgroup-related differential abundance proteins from comparison among the four subgroups. The normalised abundance in the control group (yellow) was also plotted as reference. (C) Tissue enrichment analysis using gene symbols of the 40 subgroup-related differential abundance proteins based on the Human Gene Atlas. Labelled p values were from post hoc Dunn’s test for pairwise comparisons. p values were adjusted for multiple comparisons, with the Benjamini-Hochberg method. DLDH=dihydrolipoamide dehydrogenase. HTRA2=serine protease HTRA2. IL-18 Rα=IL-18 receptor 1. suPAR=soluble urokinase plasminogen activator receptor. SIGIRR=single immunoglobulin IL-1-related receptor.

Figure 4.. Proposed Kawasaki disease model

Interactions between different environmental triggers and genetic susceptibility lead to activation of a common disease pathway that varies among patients and results in discrete clinical subgroups.