Platelet miRNAs for early diagnosis of Kawasaki disease: development and internal validation of a clinical model

Abstract

Early diagnosis of Kawasaki disease (KD) remains challenging, particularly in incomplete or early-stage cases, increasing the risk of coronary artery complications. Platelet-derived microRNAs (miRNAs) are implicated in KD pathogenesis, but standardized normalization and validated clinical models are lacking. We aimed to identify a robust reference miRNA and develop an early diagnostic model integrating platelet miRNAs with clinical indicators. A multicenter study was conducted in two phases. First, three candidate reference miRNAs were evaluated in 40 children using geNorm, NormFinder, and BestKeeper. Second, 120 children (60 KD, 60 febrile controls) were enrolled. 12 platelet miRNAs were quantified by qRT-PCR and normalized using the selected reference gene. Differentially expressed miRNAs and clinical variables were incorporated into five machine learning models. Feature selection was performed using LASSO, and model performance was assessed by tenfold cross-validation, receiver operating characteristic (ROC) analysis, calibration, and decision curve analysis. miR-126-3p was identified as the most stable reference gene. Seven platelet miRNAs were significantly upregulated in KD. The optimized support vector machine model combining selected miRNAs and clinical indicators achieved an area under the curve (AUC) of 0.944 (95% CI: 0.748–0.969), sensitivity 0.889, and specificity 0.833. Model performance remained robust in early fever (≤ 5 days) and incomplete KD. Platelet miRNAs combined with routine clinical indicators provide a reliable approach for early KD diagnosis. This internally validated model may assist clinicians in identifying KD promptly, particularly in incomplete cases, facilitating timely treatment.