INTRODUCTION:
It is unknown how often a diagnosis of Kawasaki disease (KD) is missed as the illness is naturally self-limited and clinical features are shared by many common childhood illnesses. The COVID-19 pandemic further complicated diagnosis as 40% of patients with Multisystem Inflammatory Syndrome in Children (MIS-C) met the diagnostic criteria for KD. Diagnosis is also challenging when KD occurs in association with another infection or when children do not exhibit all clinical symptoms simultaneously. An accurate diagnostic test is needed for early detection and treatment to prevent acute and long-term cardiac complications. We aimed to identify host-protein biomarkers to discriminate KD from other common pediatric diseases as the basis of a diagnostic test.
METHODS:
We performed proteomic analysis using the aptamer-based SomaScan® platform to quantify levels of 7000 proteins in a cohort of children diagnosed with KD (n=38), MIS-C (n=79), definite bacterial (DB,n=40) and definite viral infections (DV,n=43). Significantly differentially abundant (SDA) proteins (p<0.05) between KD and each comparator group were identified. Feature selection was performed to establish sparse diagnostic signatures that could optimally discriminate KD from other diseases. Findings were validated using antibody-based assays in an independent cohort. Accuracy of the models was assessed with the area under the receiver-operating characteristic curve (AUC). Associations between protein concentrations and clinical variables were tested.
RESULTS:
736 proteins were SDA between KD when compared to all other unwell comparator groups combined (MIS-C+DB+DV), with 388 proteins increasing and 348 decreasing in KD. Two sparse 4-protein diagnostic signatures were identified. Using all SDA proteins, the best-performing 4-protein signature had an AUC of 98.1% (95%CI,95.1%-100%). To enable easier detection on platforms more suited for clinical use, we identified an additional signature based on SDA proteins that were medium to highly abundant and increasing in KD. This translational signature had an AUC of 94.3% (95%CI,89.6%-99%). Antibody-based validation showed an overall AUC of 87.3% (95%CI,79.2%-94.8%).
DISCUSSION:
Our study establishes that as few as 4 proteins can distinguish KD from other pediatric infectious and inflammatory syndromes. The sparse number of proteins comprising both signatures and antibody validation makes translation into a point-of-care test feasible.