The research, conducted by a team based at Shanghai Children’s Hospital and Shanghai Children’s Medical Centre affiliated with Shanghai Jiao Tong University School of Medicine, was published in BMC Pediatrics. The investigators set out to solve two intertwined problems that have hampered earlier attempts to use molecular markers for Kawasaki disease. The first is the absence of a standardized reference gene for normalizing microRNA measurements, a technical gap that makes results difficult to compare across laboratories and patient cohorts. The second is the lack of a rigorously validated clinical model that combines molecular signals with the everyday measurements clinicians already collect at the bedside.

MicroRNAs are short, non-coding RNA molecules, roughly eighteen to twenty-five nucleotides in length, that fine-tune gene expression after transcription. Platelets, the small cell fragments best known for their role in blood clotting, carry a rich cargo of these regulatory molecules. Because platelets are deeply involved in the vascular inflammation that defines Kawasaki disease, their microRNA profile offers a molecular window into the disease process itself. Previous work had implicated platelet microRNAs in the pathogenesis of the illness, but without a reliable normalization strategy, the field lacked the reproducibility needed to translate those signals into a usable diagnostic tool.

The study unfolded in two carefully sequenced phases. In the first, the researchers evaluated three candidate reference microRNAs in forty children, applying three complementary statistical algorithms—geNorm, NormFinder, and BestKeeper—to determine which candidate held steadiest across samples. The winner was miR-126-3p, a microRNA well known for its roles in vascular biology and endothelial integrity. Establishing miR-126-3p as the most stable reference gene matters because quantitative reverse transcription polymerase chain reaction, the workhorse technique used to measure microRNA abundance, produces meaningful numbers only when samples are normalized against a reference whose expression does not shift with disease state. Without that anchor, apparent differences between patients and controls can be artifacts of measurement rather than biology.

With the reference gene secured, the team moved to the second phase, enrolling 120 children: sixty with Kawasaki disease and sixty febrile controls, that is, children with fever but without the disease. The febrile control group is critical to the study’s design, because the central diagnostic dilemma in real clinical practice is distinguishing Kawasaki disease from the many other childhood infections and inflammatory conditions that also present with fever. The researchers quantified twelve platelet microRNAs using qRT-PCR, normalizing each measurement against miR-126-3p. Seven of the twelve microRNAs were significantly upregulated in the children with Kawasaki disease, a pattern consistent with the idea that platelets actively participate in the inflammatory cascade rather than serving as passive bystanders.

Identifying differentially expressed microRNAs, however, is only half the task. The harder question is whether a combination of these molecular signals and ordinary clinical variables can reliably sort sick children into the right diagnostic category. To answer it, the team turned to machine learning. They built five different predictive models and used least absolute shrinkage and selection operator, or LASSO, regression to perform feature selection—a statistical technique that shrinks the influence of less informative variables toward zero, effectively winnowing the field down to the features that genuinely carry diagnostic weight. This step guards against overfitting, the common failure mode in which a model memorizes the quirks of its training data rather than learning patterns that generalize to new patients.

The standout performer was a support vector machine model, an algorithm that finds the optimal boundary separating two classes of data points in a high-dimensional space. When the selected microRNAs were combined with routine clinical indicators, the optimized model achieved an area under the receiver operating characteristic curve of 0.944, with a 95 percent confidence interval of 0.748 to 0.969. In practical terms, the area under the curve measures how well a model distinguishes diseased from non-diseased patients across all possible diagnostic thresholds, with 1.0 representing perfect discrimination. The model reached a sensitivity of 0.889, meaning it correctly identified nearly nine of every ten children with Kawasaki disease, and a specificity of 0.833, correctly clearing about five of every six children who did not have the disease.

Just as important as those headline numbers is where the model held up. The researchers assessed performance using tenfold cross-validation, a technique in which the data are repeatedly split so that every patient serves at some point as an unseen test case, and they examined calibration and decision curve analysis to confirm that the model’s predicted probabilities were trustworthy and clinically useful. Crucially, performance remained robust in two of the hardest subgroups: children evaluated within the first five days of fever, when clinical signs are still emerging, and children with incomplete Kawasaki disease, who lack the full constellation of classic symptoms. Incomplete cases are precisely the ones most likely to be missed and most likely to suffer coronary artery complications as a result of delayed treatment.

The clinical implications are considerable. Intravenous immunoglobulin, the standard treatment for Kawasaki disease, dramatically reduces coronary damage when given within the first ten days of illness, and the benefit is greatest the earlier therapy begins. A diagnostic tool that works in the earliest days of fever and in incomplete presentations could therefore shorten the dangerous interval between symptom onset and treatment. The study’s authors caution that their model was internally validated, meaning it was tested within the same cohort used to build it, and external validation in independent, geographically distinct populations will be needed before the approach enters routine practice. Still, the combination of a rigorously selected reference gene, a febrile control group, and disciplined machine learning methodology gives the finding a solidity that many biomarker studies lack.

Beyond the immediate diagnostic promise, the work reinforces a broader shift in pediatric medicine toward molecularly informed decision-making. Platelets are easy to obtain from a routine blood draw, and qRT-PCR is an established technology available in many hospital laboratories, which means the barrier to translating this research into a clinical assay is lower than for many emerging diagnostics. If future external validation confirms these results, a platelet microRNA panel could become a practical adjunct to clinical judgment, helping physicians act quickly for the children who need it most and avoid unnecessary interventions for those who do not. For a disease whose greatest danger lies in being recognized too late, that would represent a meaningful advance.

Subject of Research: Platelet microRNA-based clinical model for early diagnosis of Kawasaki disease in children

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

Article References: Zhou, Y., Chen, L., Chen, L., Xiao, T., Fu, L., Wu, J., Li, G., Liu, J., Zhang, C., Song, S., & Huang, M. (2026). Platelet miRNAs for early diagnosis of Kawasaki disease: development and internal validation of a clinical model. BMC Pediatrics. https://doi.org/10.1186/s12887-026-07673-x

Image Credits: AI Generated

DOI: 10.1186/s12887-026-07673-x

Keywords: Kawasaki disease, platelet miRNAs, miR-126-3p, diagnostic model, machine learning, support vector machine, LASSO, qRT-PCR, early diagnosis, coronary artery complications, pediatrics, BMC Pediatrics