Introduction

Kawasaki disease (KD) with the highest incidence in the Northeast Asia especially in Japan, Korea and China, is the most common systemic vasculitis in children [1, 2]. In China, despite insufficient nationwide data, the latest epidemiological investigation showed that the incidence of KD was on the increase since 1998 and varied from 68.8 to 107.3 per 100,000 children under five in Shanghai City from 2013 to 2017 [3]. For children with KD, intravenous immunoglobulin (IVIG) is the first-line therapy; and patients who have repeated or persistent fever after IVIG therapy for 36–48 hours is defined as IVIG resistance [4]. The incidence of IVIG resistance in recent years has been reported as 4.9–20% [5,6,7,8,9]. The ratio of coronary ectasia is significantly higher in KD patients with IVIG resistance, who requiring a second-line treatment including the repeated dose of IVIG, corticosteroids, infliximab or cyclosporine. Long-term oral anticoagulant and antiplatelet drugs, and regular follow-up of coronary lesions are required in severe cases because of coronary complications with giant coronary aneurysm and myocardial infarction [5, 10].

Previous studies have indicated a variety of high-risk factors of IVIG resistance, including demographic features, clinical manifestations and laboratory findings [11,12,13,14,15,16,17,18,19,20,21,22,23]. A number of predictive models for KD patients with IVIG resistance were reported, such as Kobayashi, Egami and Sano scoring system from Japan, San Diego scoring system from American, and Formosa scoring system from Taiwan, China. However, the aforementioned models seemed relatively ineffective when used to predict children in the Mainland of China [24, 25], which might be related to heterogeneity of the populations. In recent years, investigators also reported useful predictive methods for IVIG resistance in China [11, 21, 26,27,28]. However, their sensitivity and/or specificity of the methods should be optimized.

The main pathology of KD is systemic vasculitis. In previous studies, the laboratory data used for the prediction of IVIG resistance mainly consisted of peripheral white blood cell count (WBC), C-reactive protein (CRP), serum liver enzymes, and platelet count. Actually, the indices reflecting the vascular inflammatory response in KD, including peripheral neutrophil count, lymphocyte count, monocyte count, monocyte-to-lymphocyte ratio (MLR), mean platelet volume (MPV) and mean platelet volume-to-lymphocyte ratio (MPVLR), have not been tested in the predictive tool for IVIG resistance. Therefore, the present research was designed to collect and analyze the clinical data in children with KD to identify the predictors of IVIG resistance and to establish valuable predictive models.

Methods

Study population

Patients diagnosed with KD for the first time in Peking University First Hospital from January 2014 to May 2019 were included in this retrospective research. The diagnosis of KD, including complete KD and incomplete KD, for each patient is consistent with the standards set by the American Heart Association [29]. Patients who received IVIG treatment before admission, or had other rheumatic diseases or infectious diseases were excluded. A total of 277 children (Chinese Han nationality) were enrolled, including 180 boys and 97 girls, aged 2–128 months. They were treated with a total dose of 2g/kg of IVIG and oral aspirin (30–50 mg/kg per day) after administration. The children were divided into two groups, including IVIG-responsive and IVIG-resistant groups. The definition of IVIG resistance was persistent or recrudescent fever (≥ 38 °C) for at least 48 hours after the end of the first IVIG infusion [22]. The Regional Ethics Committee of Peking University First Hospital approved this research.

Demographic data and clinical characteristics

Age (months), gender, weight, height, illness days of initial IVIG, skin rash, extremity changes, cervical lymphadenopathy, conjunctival congestion and changes in oral mucosal were recorded. The laboratory indices of peripheral blood test before IVIG-treatment included WBC, neutrophil count, lymphocyte count, monocyte count, hemoglobin, platelet count, MLR, red blood cell distribution width (RDW), platelet count-to-lymphocyte ratio (PLR), MPV, MPVLR and CRP. Serum alanine aminotransferase (ALT), aspartate aminotransferase (AST), albumin, prealbumin, total bilirubin and serum sodium were also recorded.

Echocardiography

Coronary artery lesions were assessed by two-dimensional ultrasound. The frequency of the ultrasound probe was ≥ 5 MHz. By the short-axis view of the parasternal aorta in the supine position, the diameter of the left main coronary artery (LMCA) and the right coronary artery (RCA) were measured. The measurement is from one inner edge to the other inner edge. Z-value was the diameter of LMCA and RCA corrected by the body surface area [30]. The maximum Z value of coronary artery during hospitalization was determined. A Z value ≥ 2.5 was defined as the coronary artery injury [5].

Statistics analysis

Frequency and constituent ratio were used to describe categorical variables, and the difference between groups was compared by χ2 test. Continuous variables are showed as the mean ± standard deviation or median with quartile range, and independent sample T test or Mann–Whitney U test was used to compare the differences between the two groups depending on the data distribution. Variables with the p values less than 0.1 in the univariate analysis, together with the demographic characteristics (age, gender), were evaluated by logistic regression analysis with backward elimination to choose independent risk factors, and the logistic regression model was established.

To create an easy-to-perform scoring model, we performed the following analysis. To get the cut-off value of independent risk indicators for the prediction of IVIG-resistant patients, receiver operating characteristic (ROC) curve analysis was conducted. Two-categorical variables were converted from the continuous variables based on the optimal cut-off values derived by the abovementioned ROC curves. To verify the independent two-categorical variables, a second logistic regression analysis was performed and the odd ratios were estimated. Then, the score point of each variable was given to each patient according to the approximate odds ratio derived above. Total score points were then calculated for each patient.

Finally, ROC curves and the area under the curves (AUCs) were computed to evaluate the accuracy of the predictive model, and the cut-off values were identified by the Youden index derived from the calculated sensitivity and specificity. We also compared AUCs between the logistic regression model and the easy-to-perform scoring model. In the study, statistical analysis was carried out by SPSS V.25.0 and MedCalc V.15.2.2. P < 0.05 indicated significant difference.

Results

In this research, 277 children (180 boys and 97 girls) with KD meeting with the inclusion criteria were analyzed. The IVIG-responsive group consisted of 156 (63.4%) boys and 90 (36.6%) girls, and the IVIG-resistant group 24 boys (77.4%) and 7 girls (22.6%).

Univariate analysis

No significant differences in age, gender, symptoms and signs were observed between the two groups of children (Table 1). In the IVIG-resistant group, the rate of coronary artery lesion was much higher than that of the IVIG-responsive group (61.3%, 40.2%, P < 0.05).

Table 1 Comparison of clinical characteristics between the IVIG-responsive and IVIG-resistant groups
Full size table

Eighteen variables of laboratory examination were involved in the analysis of the research. In comparison to the IVIG-responsive group, serum total bilirubin, and peripheral neutrophil count, MPV, MPVLR and CRP were increased, and serum albumin and prealbumin were decreased in the IVIG-resistant group (Table 1, all P < 0.05).

Multivariate analysis and the logistic regression model

Eleven variables were evaluated by multivariate logistic regression, including age, gender, serum albumin, prealbumin, total bilirubin and sodium and peripheral neutrophil count, lymphocyte count, MPV, MPVLR and CRP. The results showed that age (months), peripheral neutrophil count, lymphocyte count and mean platelet volume, and serum albumin level were independent predictors of IVIG resistance (Table 2). And we obtained a logistic regression model by the following equation:

$$ \begin{aligned} {\text{Lg }}\left( {{\text{risk of}}\,\,{\text{IVIG resistance}}} \right) \, & = \, -\,0.0{28 } \times {\text{age }}\left( {{\text{months}}} \right) + 0.{1}0{8 } \times {\text{peripheral neutrophil count}}({1}0^{{9}} /{\text{L}}) \\ & \quad - 0.{412 } \times {\text{peripheral lymphocyte count}}\,\,{(1}0^{{9}} /{\text{L)}} + 0.{661 } \\ & \quad \times\,{\text{peripheral mean platelet volume }}\left( {{\text{fL}}} \right) \\ & \quad - 0.{162 } \times {\text{serum albumin}}\left( {{\text{g}}/{\text{L}}} \right) \\ \end{aligned} $$
Table 2 Independent factors identified by multiple logistic regression analysis for predicting IVIG resistance
Full size table

Establishment of predictive scoring system

The cut-off value of each independent predictor for IVIG resistance was obtained by ROC analysis. The results showed that the cut-off value to predict the IVIG resistance was as follows: age (months) ≤ 24 months, peripheral neutrophil count ≥ 10 × 109/L, peripheral lymphocyte count ≤ 3 × 109/L, peripheral mean platelet volume ≥ 10.5 fL and serum albumin ≤ 37 g/L (Table 3).

Table 3 Independent factors for predicting IVIG resistance in a scoring model
Full size table

To create the scoring system, all independent predictors were converted to categorical variables, and multivariate analysis was performed. The odds ratio and score point of each variable were as follows: age (months) ≤ 24 months: odds ratio 2.820, 3.0 points; peripheral neutrophil count ≥ 10 × 109/L: odds ratio 2.903, 3.0 points; peripheral lymphocyte count ≤ 3 × 109/L: odds ratio 3.486, 3.5 points; peripheral mean platelet volume ≥ 10.5 fL: odds ratio 3.572, 3.5 points; and serum albumin ≤ 37 g/L: odds ratio 2.414, 2.5 points (Table 3). A total score points of a study subject were then calculated according to this scoring system.

ROC analysis

ROC curve was utilized to determine the accuracy of the model for predicting IVIG-resistant patients. A cut-off value of − 0.46 yielded a sensitivity of 83.9% and a specificity of 74.8%; and the AUC was 0.808 (95% confidence interval 0.741–0.876, P < 0.001) for the logistic regression model. A cut-off value of 6.5 points yielded a sensitivity of 77.4% and a specificity of 61.0% and the AUC was 0.750 (95% confidence interval 0.666–0.834, P < 0.001) for the scoring system. The comparison between the two models’ AUCs showed no statistical difference (P = 0.113) (Fig. 1).

Fig. 1
figure 1

Receiver operating characteristic (ROC) curve of the logistic regression model and the scoring system for predicting IVIG resistance in children diagnosed with KD. In the logistic regression model and the scoring system, cut-off values of − 0.46 and 6.5 points yielded sensitivities of 83.9% and 77.4%, and specificities of 74.8% and 61.0%, respectively. The area under the curve (AUC) was 0.808 in the logistic regression model and 0.750 in the scoring system, respectively

Full size image

Discussion

Our research showed that age (months), peripheral neutrophil count, lymphocyte count and mean platelet volume, and serum albumin before IVIG-treatment were independent predictors for IVIG resistance. We used the five indicators to establish a logistic regression model for predicting IVIG-resistant children, and the cut-off of − 0.46 yielded a sensitivity of 83.9% and a specificity of 74.8%, respectively. A scoring system was set up for the convenient use in clinical practice, and the cut-off of 6.5 points yielded a sensitivity of 77.4% and a specificity of 61.0%, respectively. When the patient with KD is predicted to be at a high risk of IVIG resistance, we would consider using IVIG plus glucocorticoids in treatment.

In previous studies conducted on the Chinese population, younger age, particularly under 6-months and hypoalbuminemia were already found out as risk factors for IVIG-resistant patients [25]. However, peripheral neutrophil and lymphocyte count, and MPV were firstly involved in the predictive scoring system of IVIG resistance in KD.

Studies by Kobayashi et al. and Tan et al. showed that for IVIG-resistant patients with KD, younger age (months) was an independent risk factor [16, 21], and younger age especially in children under six months had a high risk of coronary artery lesion. Salgado et al. showed that children with KD younger than six months had high levels of peripheral WBC and CRP, but low levels of hemoglobin and serum albumin compared with those older than six months [31], which suggested that a severe inflammation occured in infants and young children with KD. When inflammatory reaction occurs, delayed apoptosis of neutrophils and stimulation of stem cells by growing factors lead to neutrophilia and redistribution in the lymphatic system, and increased apoptosis of lymphocyte leads to the lymphocytopenia [32]. In previous studies, the percentage of neutrophils was used to predict IVIG resistance, but the results of different studies were quite different. Kobayashi considered that the percentage of neutrophils ≥ 80% could predict IVIG resistance [16], Yang et al. showed the percentage of neutrophils ≥ 70% predicted IVIG resistance [27], while the Formosa predictive model assumed that neutrophil percentage ≥ 60% was the risk factor for IVIG resistance [18]. While several studies have focused on the association between the high level of peripheral neutrophil-to-lymphocyte ratio (NLR) and IVIG resistance. In Japan, Kawamura et al. found that NLR ≥ 3.83 and PLR ≥ 150 before IVIG-treatment could predict IVIG resistance [33]. In China, Hua et al. reported a model for the prediction of IVIG resistance including NLR [11], and Chen et al. found that NLR ≥ 2.51 was useful for predicting IVIG-resistant children under 12 months [34]. Considering that an elevated NLR ratio implies high neutrophil counts or low lymphocyte counts or both, this study underlines the association between these altered absolute counts and the higher risk of IVIG resistance. To further improve the capability of predicting IVIG resistance in KD, our study used the absolute count of neutrophil and lymphocyte as an entry point to predict IVIG resistance. Only univariate study confirmed that children with IVIG resistance had high peripheral neutrophil count and low lymphocyte count [17, 20, 21]. While we showed that the high level of neutrophil count and low level of lymphocyte count analyzed by multivariate regression could be used as independent predictors of IVIG resistance.

Hypoalbuminemia can predict IVIG resistance independently or in combination with other indicators [14, 27, 35]. Its possible mechanisms were not fully understood. Firstly, the increased vascular permeability during the systemic vasculitis largely leads to hypoalbuminemia. The severity of vascular leakage can be reflected by the level of vascular endothelial growth factor (VEGF) [36]. Serum VEGF concentration was inversely related to the serum albumin concentration in patients with KD [37]. Terai et al. reported that patients with IVIG resistance had higher VEGF levels and lower albumin levels compared with IVIG-responsive patients in the acute stage of the disease [38]. Secondly, previous studies indicated that indicators of liver function such as serum ALT, AST, GGT and bilirubin could be used as independent predictors of IVIG resistance [16, 17, 22]. Hepatic vasculitis resulting in the abnormal liver dysfunction might likely cause the poor ability of albumin synthesis.

MPV is a classic biomarker for inflammation. It is automatically calculated by blood analyzer according to the volume distribution of peripheral blood platelet morphology test. Peripheral MPV reflects platelet viability and function. The increased peripheral MPV has been found in some cardiovascular diseases, respiratory diseases, chronic kidney diseases, rheumatic diseases, diabetes and cancer. The increase in peripheral MPV is related to a growing number of large platelets during the inflammatory process. Large platelet facilitates the production of thromboxane A2 and β-platelet globulin [39,40,41]. But it has been unclear whether it plays a role in the prediction of IVIG-resistant KD. Our study firstly reported that MPV was an independent predictor of IVIG resistance. MPVLR represents mean platelet volume-lymphocyte ratio. Several studies have indicated that patients suffering from coronary artery disease with elevated MPVLR would have a poor prognosis [42,43,44]. In this study, patients from IVIG-resistant group had significantly increased levels of MPV and MPVLR compared with those in the IVIG-responsive group, which might be associated with the facilitated vascular inflammatory response.

In conclusion, however, there are some limitations in our study. This was a single-center research and the sample size of enrolled patients was not large enough. Further multi-center-based studies are needed to validate the predictive efficiency in the future.