Kawasaki disease (KD) is a benign but self-limited systemic vasculitis, primarily affecting children. The clinical presentation is very variable in children with recurrent fever, rash, and diffuse congestion of conjunctiva involving lips and bilateral bulbar conjunctivas (1, 2). If not diagnosed and properly treated at an early stage, serious complications include the development of coronary artery lesions (CALs), which is one of the most important cardiovascular outcomes of the disease (3). Recent studies have stated that cardiac complications due to KD now rank among the top causes of acquired heart disease in children (4, 5). Thus, the early and precise evaluation of the risk of a CAL in the clinical process is very important for reducing the complications related to cardiovascular events and improving prognosis and also enables children to receive more active and timely treatment.

An extended range of laboratory indicators was used in our study, so it was useful for defining the best predictive factors related to the risk of the development of a CAL. Considering its advantages, least absolute shrinkage and selection operator (LASSO) regression showed better performance when compared with other techniques based on conventional statistics. Thus, as compared with such classic variants, LASSO regression generates simpler models, can avoid overfitting issues, and can better solve the multicollinearity issues that arise between the independent variable features, leading to the generation of high-quality, precise models, and enhancing reliability (6). A nomogram, a new, increasingly used tool in the realm of risk prediction, is able to combine heterogeneous elements of clinical symptoms, demographic information, and laboratory data into a succinct and intuitive graphical format. It graphically illustrates the contribution of each variable to the predicted outcome and thus allows clinicians to avoid complex regression analysis. Using scores corresponding to individual variables, physicians can easily predict risk probabilities (7, 8). Unlike most of the traditional prediction models, which are usually overcomplicated or difficult to interpret, a nomogram simplifies the process and increases the practical utility of risk assessments. This makes it especially relevant for early disease risk prediction and the implementation of precision medicine strategies (9). Our study focuses on aiding physicians in predicting the likelihood of CAL development during the baseline period of KD, specifically at initial diagnosis and before the initiation of drug therapy. By analyzing variations in laboratory indicators with high predictive value, we aim to facilitate early interventions that improve clinical outcomes. In addition, our findings provide a theoretical framework for future monitoring and management strategies for Kawasaki disease.

IVIG therapy is the primary treatment for KD. It strengthens the immune response in patients with KD, suppresses the progression of systemic vasculitis, alleviates or halts vasculitic reactions, reduces the occurrence of CALs, and prevents the formation of coronary artery thrombosis (11–13). However, studies have shown that approximately 15%–20% of patients exhibit resistance to initial IVIG treatment, significantly increasing their risk of developing CALs (14–16). Despite this, predicting CAL risk among IVIG-sensitive patients remains crucial. In certain KD cases, whether in the acute stage or diagnosed later, the inflammatory response may have already caused damage to the coronary arteries, even in patients who respond well to IVIG.

Although the exact etiology of KD remains unclear, it is thought to involve abnormal immune system activation in genetically predisposed individuals following microbial infection or environmental exposure. This abnormal activation ultimately leads to acute systemic vasculitis syndrome (17, 18). Our study identified five key predictors, namely, sodium, hemoglobin, platelet count, D-dimer, and cystatin C, that collectively form a novel indicator system for predicting CALs in children with KD. Previous research suggests potential mechanisms for these high-risk factors. Patients with severe hyponatremia often experience intensified inflammation, characterized by prolonged fever and elevated levels of various inflammatory biomarkers. Thus, severe hyponatremia in KD patients may reflect greater disease severity, potentially accelerating the development of a CAL (19, 20). Although not fully explained, the mechanism underlying the decrease in hemoglobin levels among patients with CALs has been claimed to be associated with the increased production of certain cytokines, including interleukin 6 (IL-6) and IL-1β (21). Some studies have reported that changes in red blood cell (RBC) structure and function might independently or in combination affect blood circulation and potentially lead to vascular occlusion (22). Furthermore, early RBC aging and subsequent removal from the circulation may contribute to an increased risk of CALs (23). Thrombocytosis, especially in the acute phase, is a common finding in patients with KD and may contribute to cardiovascular complications in the disease. Recent studies indicate that thrombocytosis in KD results from enhanced platelet stimulation and disturbed apoptosis (24). The sharp increase in PLT as an independent risk factor for CALs is in agreement with previous reports (25). Thus, serial monitoring of PLT in KD patients is an important predictor for the development of a CAL. D-dimer levels reflecting endothelial injury are also an important index. High levels of D-dimer support the relationship between endothelial injury and the risk factors for CAL (26). CysC was a novel predictive marker in this work. While traditionally considered an early biomarker of renal impairment, there is an increasing body of evidence pointing to its strong association with cardiovascular diseases. High levels of CysC might be associated with inflammatory status and increased vascular inflammation, and further cause injury to cardiovascular tissues. CysC is an important molecule in the pathogenesis and development of KD, acting as a good predictor of the presence, severity, and prognosis of a CAL (27, 28).

In this study, LASSO-logistic multivariate analysis was performed using a set of 17 indicators, including demographic, clinical, and laboratory characteristics. LASSO regression has some unique advantages in the management of high-dimensional data and variable selection, where redundant factors are eliminated and the focus is on key predictors. This greatly enhances the accuracy and stability of the prediction model. The nomogram is important in the practical management of Kawasaki disease. It combines early clinical manifestations, laboratory parameters, and other relevant data that are used in disease diagnosis into one nomogram. With systematic analysis and model training, it gives a quantified risk score for suspected KD patients. This tool will be very useful in the early phase of the disease when symptoms are not distinct and helps the clinician in making more accurate assessments about the likelihood of KD. It greatly improves diagnostic precision by reducing the chances of misdiagnosis or missed diagnosis (29, 30). In treatment decision-making, the risk probabilities provided by the nomogram enable physicians to design personalized treatment plans. For example, for patients identified as high-risk for a CAL, the early initiation of high-dose intravenous immunoglobulin shock therapy can be prioritized, while patients categorized as low-risk may be managed with milder approaches. Moreover, during treatment, the nomogram can guide a revision of the therapeutic strategy based on dynamic changes, such as the recovery of body temperature or the speed of decline of inflammatory markers. This facilitates decisions on escalation, adjustment of medication, or maintaining the current treatment strategy for optimal therapeutic effect while reducing unnecessary drug adverse reactions (31, 32). In the prognosis assessment and long-term management, the nomogram plays an indispensable and irreplaceable role. It integrates multiple dimensions of patient information, including residual symptoms after treatment, the degree of recovery of the coronary artery as shown by echocardiography, and whether there is an abnormal inflammatory marker, to predict the possibility of future adverse cardiovascular events. This enables the provision of follow-up intervals appropriate for the level of risk. Shorter follow-up intervals can facilitate early detection and timely management of potential complications in high-risk patients. In contrast, the follow-up interval can be extended for low-risk patients, which may optimize medical resources without affecting the health of patients (8, 33). Clinical cases should be systematically divided into a training set and a validation set to guarantee the reliability and clinical utility of the prediction model. Advanced analytical methods, including ROC curve analysis, calibration curve assessment, and decision curve analysis, were applied to comprehensively evaluate the model's performance. The strictness of this validation process confirms that the model is reliable for practical applications and thus forms a firm basis for using the model in routine clinical practice.

Our study focused on the critical issue of CALs in children with KD, combining an in-depth analysis with the development of a practical prediction tool capable of providing accurate guidance for clinical practices. The research moves away from the traditionally unidirectional methodologies to adopt a multidimensional outlook, hence providing novel insights into the prevention and management of CALs in KD children.

Hyponatremia, low hemoglobin levels, elevated platelet counts, increased D-dimer levels, and higher cystatin C levels were identified as independent high-risk factors for CALs. The nomogram developed using these factors exhibits strong predictive accuracy, providing an effective tool for estimating CAL risk in patients with KD. This assists physicians in the early identification of high-risk individuals and supports targeted treatment approaches, contributing to the optimal allocation of medical resources.

This research is a single-center study, which introduces certain limitations. The absence of some clinical data and constraints on sample size have precluded mechanistic investigations into the risk factors for IVIG resistance. In the future, we aim to conduct multi-center studies to refine and expand the prediction model framework. In addition, as this study is confined to a single center, there is a slightly higher representation of patients with typical complications compared to other institutions. This may introduce bias in patient distribution, which we anticipate resolving through participation in multi-center collaborative research initiatives.