Advances in diagnostic prediction of coronary artery lesions in Kawasaki disease

Abstract

Kawasaki disease (KD) is a systemic vasculitis of unknown etiology in children with coronary artery lesions (CALs) being its most concerning complication. Untimely diagnosis may lead to long-term cardiac damage and adult-onset cardiovascular disease. Researchers have long sought to identify risk factors for predicting high-risk CALs development in patients with KD. Domestic and international scholars have established scoring systems and predictive models to assess CAL risk based on these factors. This review summarized recent advances in four key areas: (1) Diagnostic prediction scoring systems; (2) Biomarker indicators; (3) Analytical methods; and (4) Predictive models for KD-associated cardiovascular complications, aiming to provide references for CAL prediction in KD.

Key Words: Kawasaki disease; Coronary artery lesion; Risk factors; Diagnostic and predictive indicators; Diagnostic prediction model

Core Tip: Kawasaki disease is a systemic vasculitis of unknown etiology in children aged 6 months to 5 years. It poses a significant risk of coronary artery lesions, necessitating early risk prediction to prevent long-term cardiac complications. This review highlighted advances in diagnostic scoring systems, biomarker indicators, analytical methods, and predictive models for coronary artery lesions in Kawasaki disease, offering insights to improve clinical risk assessment and intervention strategies.

INTRODUCTION

Kawasaki disease (KD), also known as Kawasaki syndrome or mucocutaneous lymph node syndrome, was first reported in 1967 by the Japanese physician Tomisaku Kawasaki. It is an acute febrile systemic vasculitis of unknown etiology[1,2], predominantly affecting children aged 6 months to 5 years. The main characteristic of KD is widespread inflammation of medium and small blood vessels with cardiovascular system involvement, particularly coronary artery damage, being the most severe[3]. Without effective treatment approximately 25% of patients with KD may develop coronary artery lesions (CALs). Early identification and standardized intravenous immunoglobulin (IVIG) therapy can significantly reduce the incidence of coronary artery damage with reported rates dropping to around 5%[3]. Although the etiology and pathogenesis of KD remain incompletely understood, most scholars in recent years believe it to be an autoimmune disease triggered by certain pathogens, leading to abnormal inflammatory responses in the body and resulting in a series of clinical manifestations and pathological changes. This article reviewed the general research status, diagnostic prediction scoring systems, diagnostic prediction indicator factors, diagnostic prediction analytical methods, diagnostic prediction models, and research frontiers in the diagnosis and prediction of KD complicated by CALs with the aim of providing a reference for its clinical prediction and diagnosis.

OVERVIEW OF RESEARCH ON KD COMPLICATED BY CALS

The cardiac sequelae of KD were first identified in 1970 through autopsy reports of children who died suddenly from cardiac causes[4]. Subsequent studies increasingly showed that CALs are a severe complication of KD and have become the leading cause of acquired heart disease in children in developed countries[5]. CALs include coronary artery dilation, coronary artery aneurysms, coronary artery stenosis, occlusion, atherosclerosis, and thrombosis formation within aneurysms (Figure 1). Coronary thrombosis formation or rupture is a major cause of death in affected children[6,7]. Based on clinical symptoms, KD can be classified into complete KD and incomplete KD (IKD). The clinical features of these two types differ with complete KD typically presenting more severe symptoms and having a higher probability of developing CALs and IVIG resistance compared with IKD[7]. In recent years the incidence of IKD has shown an increasing trend in countries such as China and Japan. Due to its incomplete clinical manifestations, IKD is prone to being missed or misdiagnosed. Clinically, it is emphasized that IKD should be treated as KD.

Figure 1 Coronary artery aneurysm and corresponding echocardiography of a coronary aneurysm in a patient with Kawasaki disease (adapted from Zhang et al[7], 2023). Citation: Zhang M, Cui Q, Zhu DQ, Zhang YQ, Zhong YM, Shen J. [Follow-up Analysis of Coronary Angiography in 21 Children with Kawasaki Disease Complicated by Coronary Artery Lesions]. Shanghai Jiao Tong Daxue Xuebao (Yixueban) 2023; 43: 1535-1541. Published by Shanghai Jiao Tong University (https://xuebao.shsmu.edu.cn/CN/column/column8.shtml).

IVIG was first used for KD treatment in 1982. The 2017 American Heart Association (AHA) guidelines for the diagnosis and treatment of KD clearly recommend that when KD is diagnosed, high-dose (2 g/kg) IVIG should be administered as a single infusion as early as possible (within 10 days)[8]. Currently, diagnosing IKD requires a comprehensive assessment combining laboratory test indicators and echocardiography results. However, echocardiography often yields no positive findings in the early stages of KD; high detection rates typically occur only 2-3 weeks after disease onset, offering limited help for early diagnosis. Furthermore, results can be influenced by the ultrasound equipment and operator proficiency[9]. Some blood biochemical indicators significant for diagnosis, such as high C-reactive protein (CRP) and low albumin, also exhibit variations based on race and region. Therefore, timely and accurate diagnosis of KD is crucial[10]. The pathological basis of KD is inflammatory changes in small and medium arteries, characterized by a cascade “waterfall” effect of inflammatory cytokines closely related to coronary damage[11]. It is hypothesized that inflammatory indicators hold certain clinical significance for KD diagnosis and CAL prediction.

RESEARCH ON DIAGNOSTIC PREDICTION SCORING SYSTEMS FOR KD COMPLICATED BY CALS

Currently, there are several risk prediction scoring systems for CAL in children with KD internationally. Each has distinct characteristics in variable selection, target population, and predictive performance. Different scoring systems assess varying numbers and types of risk factors. Major scoring methods include the Kobayashi score, Egami score, Sano score, Framingham risk score, the 1984 Japanese Ministry of Health and Welfare standard, Z-score method, Dallaire formula, Zheng Lin method, McCrindle method, Lopez method, Kurotobi method, Olivieri formula, and the Age-Creatinine-Ejection Fraction (ACEF) score, among others. The sensitivity and specificity of these evaluation systems vary significantly across populations and regions. (1) The Framingham risk score originated from the Framingham Heart Study initiated in 1948. This long-term, large-scale cohort study tracked the health of residents in Framingham, Massachusetts, United States. It serves as a predictive tool to assess an individual’s risk of experiencing a major cardiovascular event (initially coronary heart disease, later expanded to general cardiovascular disease) within a specific future period (usually 10 years)[12]; (2) The 1984 Japanese Ministry of Health and Welfare standard was established by the Japanese Ministry of Health and Welfare Kawasaki Disease Research Group in 1984 as a prediction standard for CALs and remains an important historical basis for evaluating KD cardiovascular complications; (3) The Egami score was developed by Japanese scholars as a quantitative tool to predict the risk of non-response to IVIG therapy in children with KD. In 2000, Japanese scholars first used a scoring method to predict IVIG responsiveness in children with KD; (4) The Kobayashi score was developed by Japanese scholars in 2006 as a quantitative tool to predict the risk of CALs and non-response to IVIG therapy in patients with KD; and (5) The Sano score was proposed by Japanese scholars in 2007 to predict the risk of IVIG non-response in children with KD (IVIG non-response being a significant risk factor for CAL development)[13].

In 2004 the AHA endorsed the use of Z-scores derived from coronary artery internal diameter measurements to assess CALs. The threshold Z-value defining abnormality varies regionally. The AHA considers an internal diameter Z-score ≥ 2.5 as abnormal, while China’s 2012 “Clinical Recommendations for Coronary Artery Lesions in Kawasaki Disease” proposed a Z-score ≥ 2.0 as the threshold for abnormality. There are several Z-scores including: (1) Dallaire formula (Canada): Z = [ln(Measured Value) - β1 - β2 × ln(BSA)]/MSE (Mean Squared Error) where BSA stands for body surface area; (2) Olivieri formula (United States): Z-value = {ln[Coronary Measured Value (cm)] - β1 - β2 × ln(BSA m²)}/√MSE(Mean squared error); (3) Zheng Lin method: Developed by Lin et al[14] at Beijing Children’s Hospital. this method focuses on Z-score assessment of coronary artery internal diameter specifically for Chinese children. Its core involves establishing a body surface area (BSA) correction model for Chinese children’s coronary artery internal diameter to enhance the diagnostic accuracy of CALs in KD; (4) McCrindle method[15]: Specifically refers to the CAL prediction and assessment system led by Professor Brian W. McCrindle in the 2017 AHA KD management guidelines. Its core principle is replacing traditional age-based cutoffs with BSA-corrected Z-scores for precise quantification of coronary artery abnormalities; (5) Lopez method: A Z-score calculation method for assessing CALs in KD based on echocardiography. It corrects the coronary artery internal diameter for BSA, aiming to improve the precision of lesion identification; and (6) Kurotobi method: A type of Z-score method that measures coronary artery internal diameter via echocardiography, standardizes it based on BSA, and calculates the Z-score (standard deviation units). Its diagnostic standard for CALs is typically a Z-score > 2.0 or > 2.5, indicating coronary abnormality. The applicability of different Z-score assessment methods varies across regions. For instance, the Lopez, Kobayashi, Dallaire, and Zheng Lin methods demonstrate differing applicability for Chinese children with KD with the Zheng Lin method being the most suitable (Table 1)[16].

Table 1 Differences in Z-score assessment methods.

Assessment methods	Detection rate	Sensitive regions	Applicability in Chinese children
Lopez method	18.60%	Left main coronary artery, proximal right coronary artery	Moderate
Kobayashi method	38.20%	All segments (high sensitivity)	Good
Dallaire method	28.60%	Mid-distal right coronary artery (reduced missed diagnosis)	Moderate
Zheng Lin method	17.30%	Localization optimized (high specificity)	Excellent

ACEF score

The ACEF score is a simplified cardiovascular surgery risk assessment tool primarily used to predict mortality risk after cardiac surgery (especially coronary artery bypass grafting). Its core advantage lies in requiring only three basic indicators (age, serum creatinine, left ventricular ejection fraction), making it operationally convenient with significant predictive power. Both the ACEF score and Apelin-13 levels have predictive value for CAL occurrence in children with KD, and their combined predictive value is superior to either alone[17].

RESEARCH ON INFLUENCING FACTORS FOR KD COMPLICATED BY CALS

Factors influencing KD complicated by CALs are diverse and can be categorized as follows: Demographic factors; imaging markers; and independent risk factors.

Demographic factors

Ethnicity, Region, Age, and Gender: While studies specifically examining the impact of ethnicity and region on KD complicated by CALs are scarce, global research papers clearly indicate that the sensitivity of the same scoring method or prediction model varies significantly across different ethnicities and regions. Research by Gong et al[18] found that factors such as gender and ethnicity are associated with KD complicated by CALs. Age research is mainly divided into three groups: Young age (< 1 year); middle age (1-5 years); and older age (> 5 years). Research on the influence of gender on KD complicated by CALs shows that male children have a significantly higher risk of CALs than females (odds ratio = 1.5-2.0) and is particularly prominent in East Asian populations.

Imaging markers

Both echocardiography and myocardial enzyme markers are sensitive indicators of CALs in pediatric KD. Echocardiography can clearly display coronary artery dilation while creatine kinase-MB demonstrates better sensitivity in evaluating different severities of coronary lesions[19]. Coronary artery internal diameter Z-score has a good effect in assessing the severity of coronary lesions in children with KD[20]. Proinflammatory cytokines released from inflamed coronary arteries act on adjacent adipose tissue through paracrine signaling, inducing alterations in the lipid-water phase composition of adipocytes. These changes are detectable via coronary CT angiography (CCTA). The pericoronary fat attenuation index, a validated novel CCTA biomarker, serves as an indicator of coronary inflammation[21]. In patients with KD elevated mean CT attenuation values of pericoronary adipose tissue correlate with the presence of coronary artery aneurysms, reduced myocardial perfusion, and independently predict cardiovascular events[22].

Independent risk factors

Based on previous research, independent risk factors for KD complicated by CALs include the following: Male gender; albumin[23]; elevated circulating free-DNA; elevated neutrophil elastase; neutrophil count[24]; hemoglobin; neutrophils; platelet count; CRP[25]; CRP and WBC levels[26]; procalcitonin, which is currently recognized as a marker for systemic infectious diseases[27]; age; CRP level; duration of fever; timing of IVIG administration; erythrocyte sedimentation rate (ESR)[28]; increased fibrinogen concentration; pericardial effusion[29]; N-terminal pro-B-type natriuretic peptide[30]; serum interleukin (IL)-17 and IL-23 levels in children with IVIG-non-responsive KD[31]; serum angiopoietin-like protein 8, growth differentiation factor-15 levels; ESR[32]; serum 25-hydroxyvitamin D3; prealbumin; and N-terminal pro-B-type natriuretic peptide[27]; platelet miR-26a-5p; total protein and total bilirubin[29]; and pericardial effusion.

RESEARCH ON DIAGNOSTIC PREDICTION ANALYTICAL METHODS FOR KD COMPLICATED BY CALS

Most research on the diagnosis and prediction of KD complicated by CALs involves collecting clinical data from children with KD in a specific region over a certain period for retrospective analysis[27,33,34]. In retrospective analyses subjects can be divided into different groups based on coronary lesions into the CAL group and the non-CAL group[17,24,25,27,35] or based on age into the infant group, toddler group, preschool group, and school-age group[36].

For predictive studies on influencing factors for KD complicated by CALs, univariate analysis[26,28,37] or multivariate logistic regression analysis is commonly used to identify factors related to CAL severity in children with KD, including gender, duration of fever, Mycoplasma pneumoniae, Epstein-Barr virus, cytomegalovirus, Streptococcus pneumoniae, IVIG, serum caveolin-1, soluble low-density lipoprotein receptor 11, ESR, platelet count, WBC, neutrophil percentage, CRP, procalcitonin, and cytokines. Factors with a P value < 0.05 are selected as predictors[17,26,28,37,38]. When studying multiple factors, the random forest algorithm can be used to calculate the importance of predictors identified through univariate analysis for predicting concurrent CALs. Predictors are then selected again based on importance ranking[39]. Some scholars also use the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm to screen for CAL risk factors[29]. R software version 4.2 can be used to perform feature selection on all variables via LASSO, Support Vector Machine (SVM), and XGBoost (XGB) algorithms. Subsequently, three machine learning classification models, LASSO-logistic regression, SVM, and XGB, can be established to build diagnostic prediction models for KD complicated by CALs and evaluate their performance[25].

The receiver operating characteristic (ROC) curve is a statistical tool used to evaluate the performance of binary classification models. It is a core statistical tool in the field of medical imaging technology for assessing diagnostic performance, quantifying the discriminative ability of the model through the area under the curve. It has become a core indicator for evaluating the diagnostic efficacy of radiomics models. ROC curves are constructed to determine the optimal predictive cutoff values, sensitivity, and specificity of risk factors for CALs in children with KD[17,26,29,31]. Some studies also combine calibration curves, Hosmer-Lemeshow goodness-of-fit test, and decision curve analysis to analyze the predictive value of various factors for CALs in children with KD[31].

Pearson correlation analysis is the most common method for analyzing a linear correlation between two variables[31,32,35]. It is used to analyze the correlation between coronary artery internal diameter values and age, height, weight, and BSA[39] and for analyzing tumor necrosis factor-α, IL-1β, and interferon-γ levels[30] in children with KD. The power function model used to establish the Z-score calculation formula achieved a higher CAL detection rate than previous classical Z-score calculation formulas[39].

Echocardiography allows for safe, simple, detailed, and reliable real-time observation and recording of the extent of coronary lesions caused by KD and their outcomes[37]. It is an effective method for monitoring these lesions. By monitoring the coronary artery internal diameter, aneurysm formation rate, and blood flow abnormalities, coronary lesions in children with KD can be detected early. Echocardiographic coronary artery internal diameter Z-scores[32], CT, and echocardiography are all effective for evaluating cardiac function in pediatric patients with KD and coronary lesions. However, CT examination has significant advantages in diagnosing mid and distal lesions. Their combined use in clinical practice can enhance diagnostic value. Echocardiography allows dynamic observation of coronary lesions in KD and plays an important role in early diagnosis and prognosis assessment, but it faces significant difficulties in diagnosing distal coronary lesions[40]. Both dual-source CT and echocardiography have good diagnostic efficacy in the diagnosis of KD complicated by CALs in children and have obvious advantages in assessing cardiac function. However, dual-source CT is superior to echocardiography in diagnosing mid and distal lesions. Their combination can improve clinical diagnostic value.

Catheter-based coronary angiography remains the gold standard for coronary lesion assessment. However, as an invasive procedure it requires hospitalization, entails high costs and procedural complexity, carries risks of complications, and has limited repeatability. These factors render it less acceptable to many families[41,42]. In contrast, CCTA provides accurate visualization of coronary morphological anomalies, including aneurysm size, morphology, quantity, and location. It further detects calcification, thrombosis, stenosis, and distal coronary lesions. Serving as a vital complement to two-dimensional echocardiography, CCTA demonstrates superior diagnostic performance over ultrasound in identifying type IV or higher coronary lesions in KD. Compared with conventional angiography, it offers significant advantages for pediatric populations with fewer procedural risks, establishing itself as a validated modality for precise coronary anomaly assessment[43].

RESEARCH ON DIAGNOSTIC PREDICTION MODELS FOR KD COMPLICATED BY CALS

Diagnostic prediction models for KD complicated by CALs are evolving from single clinical indicators to multimodal integration (genetics + dynamic inflammation + imaging). It is common practice to perform correlation analysis and permutation/combination of highly correlated predictors before constructing multiple multivariate logistic prediction models with other predictors and then build regression equations to verify the goodness of fit of the model[18,38]. Some studies permute/combine highly correlated predictors and then build multiple multivariate logistic prediction models with other predictors to select the best prediction model[41]. Zhang and Zhang[38] used multivariate logistic regression analysis to identify predictive indicators for CAL occurrence in children with KD over 5 years old and constructed a risk prediction model. The performance of the model was evaluated using ROC curves. Finally, based on the Framingham risk scoring method, predictive indicators were stratified and quantified, the contribution value of each indicator to CAL prediction in children with KD over 5 years old was calculated, and a risk prediction scoring model was built.

The Kurotobi method is often used as an auxiliary tool for the diagnosis and prediction of KD complicated by CALs. Coronary artery hemodynamic simulations (such as wall shear stress, Fractional Flow Reserve) are gradually being applied to CAL risk assessment, compensating for the limitations of pure morphological measurements.

Ma et al[12] used the Framingham risk scoring method to assign values to predictive indicators and construct a risk prediction scoring model. Their study on children with KD over 5 years old found that those with longer fever duration before the first IVIG treatment, higher high-sensitivity-CRP levels, or higher serum ferritin levels were more prone to developing CALs. Chen et al[13] established a nomogram prediction model based on anti-neutrophil cytoplasmic antibodies, creating an effective scoring model to predict the risk of CALs in children with KD. This can provide a reference for the early identification of children at high risk and the formulation of personalized treatment and management strategies in clinical practice. Wang et al[39] demonstrated that the Z-score calculation formula based on the power function regression model achieved a higher coronary lesion detection rate. The area under the curve of the ROC curve quantifies the discriminative ability of the model. Zhou et al[44] used multivariate logistic regression analysis to build a risk prediction model and drew ROC curves to evaluate the predictive performance of the model.

To evaluate the predictive performance of models, calibration curves and decision curve analysis can also be used. Five-fold cross-validation can be performed, and the reliability of the model can be further verified using a validation set[18]. Some researchers built diagnostic prediction models for KD complicated by CALs using computer algorithms. Ma et al[45] used the R language to construct and validate a nomogram prediction model that demonstrated good accuracy and predictive capability. Ai[25] utilized machine learning SVM and XGB algorithms to build models. The models were evaluated by comparing discrimination, binary confusion matrices, and clinical decision curves. The XGB model was further explained using Shapley Additive explanations plots. Compared with traditional logistic regression models and the Kobayashi scoring system, the XGB clinical prediction model demonstrated superior performance. Nomograms transform complex prediction formulas into visual graphics through multivariate regression analysis, enabling individualized risk quantification. Cao et al[30] believed that both logistic regression analysis and nomogram model construction are effective methods. The nomogram model demonstrated higher advantages in predictive performance with its precision being particularly significant among various methods. It can positively impact the prediction and clinical diagnosis of KD in children.

RESEARCH FRONTIERS IN THE DIAGNOSIS AND PREDICTION OF KD COMPLICATED BY CALS

The diagnostic accuracy of KD complicated by CALs still requires improvement. Enhancing diagnostic accuracy, especially early in the disease course, is crucial. Future development should focus on discovering specific biomarkers capable of effectively distinguishing KD from other similar diseases (such as scarlet fever, drug reactions, etc.). Diagnostic techniques should be improved by utilizing advanced imaging and molecular biology technologies to increase diagnostic sensitivity and specificity. Future research trends should emphasize strengthening multicenter and international collaborative models. Validation through broader samples will enable resource sharing and experience exchange, enhancing the accuracy and application value of combining scoring systems with laboratory indicators for KD complicated by CALs risk assessment.

CONCLUSION

KD poses a significant risk of CALs, necessitating early risk prediction to prevent long-term cardiac complications. This review highlighted advances in diagnostic scoring systems, biomarker indicators, analytical methods, and predictive models for CAL in KD, offering insights to improve clinical risk assessment and intervention strategies.