New age-specific prediction models improve incomplete Kawasaki disease diagnosis

Incomplete Kawasaki disease (IKD) remains hard to recognize because it often lacks the classic signs of complete Kawasaki disease (CKD), yet delayed diagnosis can lead to coronary artery damage. A new data-mining study has developed age-specific prediction models that combine routine laboratory markers to estimate the probability of IKD in children with fever. The models were built for three age bands—0–24 months, 24–60 months, and older than 60 months—and validated in separate datasets. The work identifies uric acid (UA) as a new independent risk factor and shows that age-tailored models can improve early diagnostic support, especially in infants and toddlers.

Kawasaki disease (KD) is an acute, self-limited febrile illness that mainly affects children under five years old and can cause acquired heart disease. Incomplete Kawasaki disease (IKD) lacks the typical features used to diagnose complete KD, so it is easily confused with common febrile illnesses such as respiratory infections, adenovirus infection, lymphadenitis, and sepsis. Children with IKD may have fewer clinical signs but a higher risk of coronary artery abnormalities, making early recognition critical. Most published prediction models target KD overall rather than IKD, and one model may not fit children at different developmental stages. Given these challenges, in-depth research is needed to develop age-specific, routinely available tools for early IKD diagnosis.

Researchers at the Children's Hospital of Chongqing Medical University and collaborating units in Chongqing, China, published (DOI: 10.1002/pdi3.2516) the study online on 23 December 2024 in Pediatric Discovery. They analyzed routine clinical and laboratory data from children with IKD and other febrile diseases (RFD), used data-mining and logistic-regression methods, and built separate prediction models for three age groups. The models were tested in a held-out dataset and then validated with newly admitted cases, with performance assessed by receiver operating characteristic curves, calibration, and decision-curve analysis.

The study included 809 children with IKD and 2,427 children with RFD for model development, plus 607 IKD and 1,821 RFD cases for validation. In the 0–24-month group, eight independent factors entered the model: C-reactive protein (CRP), lactate dehydrogenase (LDH), UA, total protein (TP), albumin (ALB), absolute value of red blood cell distribution (RDA), platelet count (PLT), and hemoglobin (HGB). This model reached an area under the curve (AUC) of 0.862 in the test set and 0.878 in new data. The 24–60-month model used LDH, UA, ALB, PLT, and mean corpuscular hemoglobin concentration (MCHC), with AUCs of 0.829 and 0.818. The older group used LDH, UA, MCHC, and PLT, with AUCs of 0.694 and 0.804. PLT was a consistent risk factor (odds ratio [OR] > 1), whereas LDH and UA were protective relative to RFD (OR < 1); UA emerged as a new independent marker. In the youngest group, CRP and TP were also risk factors, while ALB, RDA, and HGB were protective. The age-specific models outperformed several published KD models in AUC and sensitivity/specificity comparisons, and the youngest model was especially strong. Nomograms were created to help clinicians apply the equations.

The authors said the study was designed to move beyond a one-size-fits-all diagnosis. They said splitting children by age helped capture developmental differences in laboratory markers and improved performance in the youngest group, where IKD is especially hard to recognize. They described UA as a promising new marker and said the models are intended as decision support, not a replacement for clinical judgment. They said the tool could help clinicians ask the right questions earlier and use routine data more systematically. They also said multicenter prospective validation is the next step.

For clinical teams, the models and nomograms could turn routine blood tests into an early warning aid when a febrile child lacks classic KD signs. Because the inputs are commonly ordered laboratory markers, the approach may fit into electronic health records without adding extra tests. The strongest performance in the 0–24-month group is important because IKD is hardest to spot in this age band and coronary complications can be serious. The findings may help prioritize echocardiography and follow-up, but the authors caution that the models need prospective, multicenter validation before broad clinical use. The study also highlights UA as a candidate marker for future research.