川崎病连续静脉注射免疫球蛋白治疗耐药性的预测建模：一项全国性研究-动脉网

Predictive modeling of consecutive intravenous immunoglobulin treatment resistance in Kawasaki disease: A nationwide study

Nature 等信源发布 2025-01-06 15:25



可切换为仅中文







Abstract

摘要

Kawasaki disease (KD) is a leading cause of acquired heart disease in children, often resulting in coronary artery complications such as dilation, aneurysms, and stenosis. While intravenous immunoglobulin (IVIG) is effective in reducing immunologic inflammation, 10–15% of patients do not respond to initial therapy, and some show resistance even after two consecutive treatments.

川崎病（KD）是儿童获得性心脏病的主要原因，通常会导致冠状动脉扩张，动脉瘤和狭窄等并发症。虽然静脉注射免疫球蛋白（IVIG）可有效减轻免疫性炎症，但10-15%的患者对初始治疗无反应，有些患者甚至在连续两次治疗后仍表现出耐药性。

Predicting which patients will not respond to these two IVIG treatments is crucial for guiding treatment strategies and improving outcomes. This study aimed to forecast resistance to two consecutive IVIG treatments using advanced machine learning models based on clinical and laboratory data. Data from the 9th National Kawasaki Disease Patient Survey by the Korean Kawasaki Disease Society encompassing 15,378 patients (mean age 33.0 ± 24.8 months; sex ratio 1.4:1) were used.

预测哪些患者对这两种IVIG治疗没有反应对于指导治疗策略和改善预后至关重要。。韩国川崎病学会第九次全国川崎病患者调查的数据包括15378名患者（平均年龄33.0±24.8个月）；性别比1.4:1）。

Clinical and laboratory findings included white blood cell count, absolute neutrophil count (ANC), platelet count, erythrocyte sedimentation rate, serum protein, aspartate aminotransferase, alanine aminotransferase, total bilirubin, N-terminal pro-brain natriuretic peptide, and presence of pyuria. Machine learning models, including Logistic Regression (LR), Multi-Layer Perceptron (MLP), Random Forest (RF), CATBoost, Explainable Boosting Machine (EBM), and Gradient Boosting Machine (GBM), were applied to predict treatment resistance.

临床和实验室检查结果包括白细胞计数，绝对中性粒细胞计数（ANC），血小板计数，红细胞沉降率，血清蛋白，天冬氨酸转氨酶，丙氨酸转氨酶，总胆红素，N末端脑钠肽前体和脓尿的存在。应用机器学习模型，包括逻辑回归（LR），多层感知器（MLP），随机森林（RF），CATBoost，可解释提升机（EBM）和梯度提升机（GBM）来预测治疗耐药性。

The machine learning models achieved Area Under the Receiver Operating Characteristic Curve (AUROC) values between 0.664 and 0.791, with the GBM model exhibiting the highest AUROC of 0.791. Analysis of feature importance revealed that ANC, serum protein, platelet count, and C-reactive protein (CRP) levels were the most significant predictors of treatment resistance.

机器学习模型在接收器工作特性曲线（AUROC）值下的面积在0.664和0.791之间，GBM模型的AUROC最高为0.791。特征重要性分析显示，ANC，血清蛋白，血小板计数和C反应蛋白（CRP）水平是治疗抵抗的最重要预测因子。

The cutoff values for these predi.

这些预测的截止值。

Introduction

简介

Kawasaki disease (KD), an acute febrile illness predominantly affecting young children, is characterized by systemic vasculitis that can lead to coronary artery complications such as dilation, aneurysms, and stenosis

川崎病（KD）是一种主要影响幼儿的急性发热性疾病，其特征是全身性血管炎，可导致冠状动脉扩张，动脉瘤和狭窄等并发症

. Despite extensive research since its identification by Dr. Tomisaku Kawasaki in 1967, the etiology of KD remains unknown

尽管自1967年川崎Tomisaku博士鉴定以来进行了广泛的研究，但KD的病因仍然未知

. The standard treatment for KD involves intravenous immunoglobulin (IVIG) combined with aspirin, which significantly reduces immunologic inflammation

KD的标准治疗方法是静脉注射免疫球蛋白（IVIG）联合阿司匹林，可显着降低免疫炎症

. However, 10–15% of patients do not respond to initial IVIG therapy, classifying them as high-risk for coronary complications

然而，10-15%的患者对初始IVIG治疗没有反应，将其归类为冠状动脉并发症的高风险

. A second dose of IVIG is commonly used as the most frequent second-line therapy for those resistant to initial treatment

对于那些对初始治疗有抵抗力的患者，第二剂IVIG通常被用作最常见的二线治疗

. Predicting which patients will not respond to two consecutive IVIG treatments is crucial for guiding treatment strategies and improving outcomes. This approach helps identify patients who might benefit more from alternative second-line therapies earlier in their treatment course.

预测哪些患者对连续两次IVIG治疗无反应对于指导治疗策略和改善预后至关重要。这种方法有助于确定在治疗过程早期可能从替代二线治疗中获益更多的患者。

Traditional scoring systems, such as the Kobayashi, Egami, and Sano scores, have been widely used to predict resistance to initial IVIG therapy. These methods incorporate clinical and laboratory parameters like age, fever duration, and serum biomarkers to assess risk. However, their performance has been inconsistent across different populations due to genetic and environmental variability, limiting their clinical applicability outside Japan.

。这些方法结合了临床和实验室参数，如年龄，发热持续时间和血清生物标志物来评估风险。然而，由于遗传和环境的变异性，它们在不同人群中的表现不一致，限制了它们在日本以外的临床适用性。

. Furthermore, these scoring systems are designed to predict resistance to a single IVIG dose and may not adequately capture the complexity of predicting resistance to two consecutive treatments. This highlights the need for more sophisticated predictive tools capable of addressing these limitations..

此外，这些评分系统旨在预测对单个IVIG剂量的耐药性，并且可能无法充分捕捉预测对连续两次治疗的耐药性的复杂性。这突出表明需要能够解决这些限制的更复杂的预测工具。。

Advanced machine learning models offer a promising alternative, leveraging large datasets and complex algorithms to identify patterns and interactions among clinical and laboratory variables that traditional approaches might overlook.

先进的机器学习模型提供了一种有前途的替代方案，利用大型数据集和复杂算法来识别传统方法可能忽略的临床和实验室变量之间的模式和相互作用。

Accurate prediction can help tailor early intervention strategies and implement more effective alternative treatments for high-risk patients

准确的预测可以帮助制定早期干预策略，并为高危患者实施更有效的替代治疗

. Previous efforts to develop predictive models for IVIG resistance after a single dose have yielded inconsistent results, further emphasizing the need for innovative approaches.

以前在单次剂量后开发IVIG耐药性预测模型的努力产生了不一致的结果，进一步强调了创新方法的必要性。

This study aims to address this gap by utilizing advanced machine learning models to predict resistance to two consecutive IVIG treatments, thereby enhancing personalized treatment strategies for KD patients

这项研究旨在通过利用先进的机器学习模型来预测对两种连续IVIG治疗的耐药性，从而解决这一差距，从而增强KD患者的个性化治疗策略

Methods

方法

Ethical approval and data availability

道德认可和数据可用性

This study was approved by the Institutional Review Board of [Insert Institution Name] (Approval number: Seoul National University Hospital is No. 1612-071-813). All methods were carried out in accordance with relevant guidelines and regulations. Informed consent was obtained from all subjects and/or their legal guardians.

这项研究得到了[插入机构名称]的机构审查委员会的批准（批准号：首尔国立大学医院，编号1612-071-813）。所有方法均按照相关指南和规定进行。所有受试者和/或其法定监护人均已获得知情同意。

The data used in this study were sourced from the 9th National Kawasaki Disease Patient Survey conducted by the Korean Kawasaki Disease Society. The dataset used and analyzed during the current study is available from the corresponding author upon reasonable request..

本研究中使用的数据来自韩国川崎病学会进行的第九次全国川崎病患者调查。本研究中使用和分析的数据集可根据合理要求从通讯作者处获得。。

Study patients

研究患者

We analyzed the medical records of 15,378 children from the ninth triennial nationwide questionnaire survey (2015–2017) conducted by the Korean Society of Kawasaki Disease. This survey involved 98 hospitals with residency programs and 108 community-based children’s hospitals without residency programs.

我们分析了韩国川崎病学会进行的第九次三年期全国问卷调查（2015-2017）中15378名儿童的病历。这项调查涉及98家有住院医师计划的医院和108家没有住院医师计划的社区儿童医院。

This is the most recent national survey conducted on KD in Korea.

这是最近在韩国对KD进行的全国性调查。

Diagnosis and treatment of Kawasaki disease

川崎病的诊断与治疗

KD was diagnosed and treated according to the American Heart Association guidelines, which include fever with at least 4 of the following 5 features: polymorphous exanthema, changes in the extremities, changes in the lips and mouth, or non-purulent cervical lymphadenopathy

KD是根据美国心脏协会的指南诊断和治疗的，其中包括发烧，至少有以下5种特征中的4种：多形性皮疹、四肢改变、嘴唇和嘴巴改变或非化脓性颈部淋巴结病

. Initial treatment involved IVIG (2 g/kg/day) infused over 10–12 h with 30–50 mg/kg aspirin. Laboratory tests were performed on admission and 48 h after IVIG treatment. A body temperature below 37.5 °C was considered afebrile

初始治疗包括IVIG（2 g/kg/天）在10-12小时内注入30-50 mg/kg阿司匹林。入院时和IVIG治疗后48小时进行实验室检查。体温低于37.5°C被认为是不发热的

IVIG resistance

IVIG抵抗

Primary IVIG resistance was defined as a fever (body temperature of 38.0 °C or higher) persisting for more than 36 h after the first IVIG course

原发性IVIG耐药性定义为在第一次IVIG疗程后持续超过36小时的发烧（体温为38.0°C或更高）

. Secondary IVIG resistance was defined as a fever following the second IVIG infusion that required further treatment as judged by the clinician. Most patient resistant to the second IVIG dose received pulse methylprednisolone therapy (15–30 mg/kg) for 3 days

继发性IVIG耐药性被定义为第二次IVIG输注后发烧，需要临床医生进一步治疗。大多数对第二剂IVIG耐药的患者接受了脉冲甲基强的松龙治疗（15-30 mg/kg），持续3天

or infliximab infusion

或英夫利昔单抗输注

Coronary artery dilatation (CAD)

冠状动脉扩张（CAD）

CAD was defined using two criteria depending on the hospital’s preference: (1) an internal diameter of 3 mm (< 5 years) or 4 mm (> 5 years) based on Japanese Ministry of Health guidelines

根据医院的偏好，使用两个标准定义CAD：（1）根据日本厚生省指南，内径为3毫米（<5年）或4毫米（>5年）

or (2) American Heart Association guidelines using a BSA-adjusted z-score system where a z-score of ≥ + 2.5 SD indicated abnormality

或（2）美国心脏协会指南，使用BSA调整的z评分系统，其中z评分≥2.5 SD表示异常

. A giant aneurysm was defined as a diameter greater than 8 mm or a z-score greater than 10

巨大动脉瘤的定义是直径大于8毫米或z得分大于10

Model development and validation

模型开发和验证

For data preparation, out of 14,805 cases, 1,663 cases with the outcome of resistance to the second-line treatment recorded were initially selected. After excluding cases with outliers or missing values, 968 cases were analyzed. Among these, 721 were labeled as not resistant and 247 as resistant to the second dose of IVIG.

。排除异常值或缺失值的病例后，分析了968例病例。其中，721人被标记为不耐药，247人被标记为对第二剂IVIG耐药。

Statistical analysis was conducted on both continuous and categorical data. For continuous variables, the mean and standard deviation were calculated for each outcome, and p-values were determined. For categorical variables, the frequency and proportion of the input variables when the outcome was resistance were calculated along with p-values..

。对于连续变量，计算每个结果的平均值和标准差，并确定p值。对于分类变量，计算结果为阻力时输入变量的频率和比例以及p值。。

Using Python 3.9, various models were developed: Logistic Regression (LR), Multi-Layer Perceptron (MLP), Random Forest (RF), CATBoost, Explainable Boosting Machine (EBM), and Gradient Boosting Machine (GBM). Data preprocessing included mean imputation for continuous variables, default value substitution for categorical variables, and capping outliers at three standard deviations from the mean.

使用Python 3.9，开发了各种模型：逻辑回归（LR），多层感知器（MLP），随机森林（RF），CATBoost，可解释提升机（EBM）和梯度提升机（GBM）。数据预处理包括连续变量的平均插补，分类变量的默认值替代，以及将异常值限制在平均值的三个标准偏差。

The dataset was split into training and evaluation sets in an 80:20 ratios, maintaining the balance of negative and positive cases. Both subsets were normalized using the Standard Scaler method, and class imbalance was addressed by applying the Synthetic Minority Over-Sampling Technique (SMOTE). SMOTE is a method used to balance class distributions by generating synthetic samples for the minority class.

数据集以80:20的比例分为训练集和评估集，保持阴性和阳性病例的平衡。使用标准定标器方法对这两个子集进行了归一化，并通过应用合成少数过采样技术（SMOTE）解决了类别不平衡问题。SMOTE是一种通过为少数类生成合成样本来平衡类分布的方法。

It works by creating new instances that lie along the line segments between existing minority class samples, which helps improve model performance when predicting rare outcomes..

它通过创建位于现有少数群体样本之间的线段上的新实例来工作，这有助于在预测罕见结果时提高模型性能。。

Explainable Boosting Machine (EBM), one of the models used, is an interpretable machine learning algorithm based on generalized additive models. Unlike traditional black-box machine learning models, EBM allows for clear visualization of feature contributions, making it particularly useful in clinical research where interpretability is critical..

可解释提升机（EBM）是一种基于广义加性模型的可解释机器学习算法，是所使用的模型之一。与传统的黑箱机器学习模型不同，EBM允许清晰地可视化特征贡献，使其在可解释性至关重要的临床研究中特别有用。。

Model validation metrics included Area Under the Receiver Operating Characteristic Curve (AUROC), sensitivity, and specificity. Optimal threshold determination ensured balanced evaluation.

。最佳阈值确定确保了平衡评估。

The selection of machine learning models was guided by both the clinical characteristics of Kawasaki Disease (KD) and the need for robust predictive performance. Gradient Boosting Machine (GBM) and CATBoost were prioritized due to their proven effectiveness in handling heterogeneous tabular datasets and their ability to model complex, non-linear interactions.

机器学习模型的选择受到川崎病（KD）临床特征和强大预测性能需求的指导。梯度提升机（GBM）和CATBoost因其在处理异构表格数据集方面的有效性以及对复杂非线性交互建模的能力而被优先考虑。

These models also provide interpretable outputs, such as feature importance scores, which are crucial for identifying key predictors of treatment resistance. Logistic Regression was included as a baseline model to compare performance against more complex models, while Random Forest and Explainable Boosting Machine (EBM) were selected for their complementary strengths in classification tasks and interpretability..

这些模型还提供了可解释的输出，例如特征重要性评分，这对于识别治疗抵抗的关键预测因子至关重要。包括逻辑回归作为基线模型，以比较更复杂模型的性能，而选择随机森林和可解释提升机（EBM）是因为它们在分类任务和可解释性方面的互补优势。。

Results

结果

Several key clinical and laboratory parameters showed significant differences between the two groups: Group 1 (patients who responded to either the initial or the second dose of IVIG) and Group 2 (patient resistant to two consecutive doses of IVIG). Group 2 patients were older (36.8 ± 22.9 months vs.

几个关键的临床和实验室参数显示两组之间存在显着差异：第1组（对初始或第二剂IVIG有反应的患者）和第2组（对连续两剂IVIG有抗性的患者）。第2组患者年龄较大（36.8±22.9个月vs。

32.8 ± 23.0 months, .

32.8±23.0个月。

= 0.0006) and had higher weights (14.9 ± 5.1 kg vs. 13.9 ± 4.9 kg,

==0.0006），体重更高（14.9±5.1公斤比13.9±4.9公斤，

< 0.001) and heights (94.74 ± 17.20 cm vs. 91.63 ± 16.58 cm,

0.001）和身高（94.74±17.20厘米对91.63±16.58厘米，

= 0.0002). Laboratory parameters showed that Group 2 had higher mean absolute neutrophil count (71.52 ± 14.65 vs. 62.79 ± 16.10,

实验室参数显示，第2组的平均绝对中性粒细胞计数较高（71.52±14.65比62.79±16.10，

< 0.001), C-reactive protein levels (9.83 ± 5.98 mg/L vs. 7.21 ± 5.34 mg/L,

P<0.001）、C反应蛋白水平（9.83±5.98±L vs.7.21±5.34 mg/L，

< 0.001), and erythrocyte sedimentation rate (60.80 ± 27.18 mm/hr vs. 56.27 ± 26.41 mm/hr,

，和红细胞沉降率（60.80±27.18 mm/hr vs.56.27±26.41 mm/hr，

= 0.0007). Protein and albumin levels were lower in Group 2, with protein levels at 6.68 ± 0.81 g/dL vs. 6.59 ± 0.60 g/dL (

第2组的蛋白质和白蛋白水平较低，蛋白质水平为6.68±0.81 g/dL，而6.59±0.60 g/dL(

= 0.0063) and albumin at 3.72 ± 0.79 g/dL vs. 3.92 ± 0.63 g/dL (

==0.0063）和白蛋白分别为3.72±0.79 g/dL和3.92 g/dL（0.63 g/dL）

< 0.001). Liver function tests indicated higher aspartate aminotransferase and alanine aminotransferase levels in Group 2 (112.89 ± 137.24 vs. 77.37 ± 107.55,

肝功能检查显示第2组天冬氨酸转氨酶和丙氨酸转氨酶水平较高（112.89±137.24 vs.77.37±107.55，

< 0.001; 117.03 ± 129.75 vs. 84.32 ± 118.00,

< 0.001;117.03±129.75对84.32±118.00，

< 0.001). Total bilirubin and sodium levels also differed significantly, with Group 2 showing higher total bilirubin (0.95 ± 0.85 mg/dL vs. 0.57 ± 0.58 mg/dL,

总胆红素和钠水平也有显着差异，第2组总胆红素较高（0.95±0.85 mg/dL vs.0.57±0.58 mg/dL，

< 0.001) and lower sodium (135.52 ± 2.50 mmol/L vs. 136.53 ± 2.59 mmol/L,

<0.001）和低钠（135.52±2.50 mmol/L vs.136.53±2.59 mmol/L，

< 0.001). Other parameters, including sex, fever duration before initial treatment, conjunctivitis, red lips, palm erythema, desquamation, rash, Bacille Calmette-Guerin scar reactivation, cervical lymphadenopathy, white blood cell count, hemoglobin level, and platelet count, did not show statistically significant differences between the two groups (Table .

<0.001）。其他参数，包括性别，初始治疗前发热持续时间，结膜炎，红唇，手掌红斑，脱屑，皮疹，卡介苗疤痕再激活，颈部淋巴结肿大，白细胞计数，血红蛋白水平和血小板计数，两组之间无统计学差异（表。

Fig. 1

图1

ROC curves for the machine learning models indicate varying levels of predictive accuracy for IVIG resistance. ROC: Receiver Operating Characteristic, AUROC: Area Under the Receiver Operating Characteristic Curve, SVM: Support Vector Machine, KNN: K-Nearest Neighbors, MLP: Multi-Layer Perceptron, GBM: Gradient Boosting Machine..

机器学习模型的ROC曲线表明IVIG抗性的预测准确性水平不同。ROC：接收器工作特性，AUROC：接收器工作特性曲线下的面积，SVM：支持向量机，KNN：K-最近邻，MLP：多层感知器，GBM：梯度提升机。。

Full size image

全尺寸图像

Fig. 2

图2

Feature importance analysis highlights the key predictors of resistance to IVIG treatment. ANC = Absolute Neutrophil Count, PLT = Platelet Count, CRP = C-reactive Protein, Hb = Hemoglobin, bil = Bilirubin, ALT = Alanine Aminotransferase, AST = Aspartate Aminotransferase, ESR = Erythrocyte Sedimentation Rate, WBC = White Blood Cell Count, fvbefore = Fever duration before initial treatment, conj = Conjunctival injection, redlip = Red lips, palm = Palmar erythema, desqua = Digit desquamation, BCG = Bacille Calmette-Guerin erythema, cervical = Cervical lymphadenopathy, IVIG= intravenous immunoglobulin..

特征重要性分析突出了IVIG治疗耐药性的关键预测因子。ANC=绝对中性粒细胞计数，PLT=血小板计数，CRP=C反应蛋白，Hb=血红蛋白，bil=胆红素，ALT=丙氨酸氨基转移酶，AST=天冬氨酸氨基转移酶，ESR=红细胞沉降率，WBC=白细胞计数，fvbefore=初始治疗前发热持续时间，conj=结膜注射，红唇=红唇，手掌=手掌红斑，脱皮=手指脱皮，BCG=卡介苗红斑，宫颈=颈淋巴结病，IVIG=静脉注射免疫球蛋白。。

Full size image

全尺寸图像

Table 1 Clinical and Laboratory parameters in patients responsive and non-responsive to IVIG.

表1对IVIG有反应和无反应的患者的临床和实验室参数。

Full size table

全尺寸表

Table 2 Performance Metrics of Machine Learning Models.

表2机器学习模型的性能指标。

Full size table

全尺寸表

Fig. 3

图3

Correlation heatmap of clinical and laboratory parameters associated with IVIG resistance.

与IVIG耐药性相关的临床和实验室参数的相关性热图。

Full size image

全尺寸图像

Predictive model performance

预测模型性能

The performance of the machine learning models in predicting resistance to two consecutive IVIG treatments was evaluated using the AUROC, sensitivity, and specificity. The Gradient Boosting Machine (GBM) model demonstrated the highest AUROC of 0.791, indicating the best predictive performance among the evaluated models.

使用AUROC，敏感性和特异性评估机器学习模型在预测对两次连续IVIG治疗的抗性方面的性能。。

The Random Forest (RF) and Explainable Boosting Machine (EBM) models also showed strong performance with AUROCs of 0.765 and 0.780, respectively (Table .

随机森林（RF）和可解释提升机（EBM）模型也显示出强大的性能，AUROC分别为0.765和0.780（表）。

). The ROC curve illustrated the true positive rate (sensitivity) against the false positive rate (1 - specificity) for each model. The GBM model consistently demonstrated the highest AUROC of 0.791. The EBM model also showed strong performance with an AUROC of 0.780. Other models, including Logistic Regression (LR), Multi-Layer Perceptron (MLP), RF, and CATBoost, exhibited AUROC values of 0.692, 0.664, 0.765, and 0.763, respectively, demonstrating varied performance in predicting resistance to two consecutive IVIG treatments (Fig. .

)。ROC曲线显示了每个模型的真阳性率（敏感性）与假阳性率（1特异性）。GBM模型始终显示出最高的AUROC为0.791。EBM模型也表现出强劲的表现，AUROC为0.780。其他模型，包括逻辑回归（LR），多层感知器（MLP），RF和CATBoost，分别显示AUROC值为0.692、0.664、0.765和0.763，在预测对两种连续IVIG治疗的耐药性方面表现出不同的性能（图。

). Based on these results, the GBM model was selected as the preferred model for predicting treatment resistance due to its superior performance in terms of AUROC, sensitivity, and specificity, providing the most reliable predictions for guiding treatment strategies in Kawasaki disease.

)。基于这些结果，GBM模型被选为预测治疗耐药性的首选模型，因为它在AUROC，敏感性和特异性方面具有优异的性能，为指导川崎病的治疗策略提供了最可靠的预测。

To enhance the statistical robustness of these findings, 95% confidence intervals (CIs) were calculated for AUROC and other performance metrics. For the GBM model, the AUROC was 0.791 (95% CI: 0.765–0.817), and for the EBM model, it was 0.780 (95% CI: 0.754–0.806). The Random Forest (RF) model achieved an AUROC of 0.765 (95% CI: 0.738–0.792).

为了增强这些发现的统计稳健性，计算了AUROC和其他性能指标的95%置信区间（CI）。对于GBM模型，AUROC为0.791（95%CI:0.765-0.817），对于EBM模型，AUROC为0.780（95%CI:0.754-0.806）。随机森林（RF）模型的AUROC为0.765（95%CI:0.738-0.792）。

These confidence intervals were derived using bootstrap resampling with 1,000 iterations. Sensitivity and specificity metrics were also supplemented with their respective confidence intervals (Table .

这些置信区间是使用1000次迭代的自举重采样得出的。敏感性和特异性指标也补充了各自的置信区间（表）。

Feature importance analysis revealed that neutrophil count, serum protein, platelet count, and C-reactive protein were the most significant factors in predicting resistance to two consecutive IVIG treatments (Fig.

特征重要性分析显示，中性粒细胞计数、血清蛋白、血小板计数和C反应蛋白是预测连续两次IVIG治疗耐药性的最重要因素（图）。

). Neutrophil count showed the highest importance score. Serum protein, platelet count, and C-reactive protein levels were also crucial, with their respective cutoff values at 7.0 g/dL, 519,000/mm

)。。血清蛋白，血小板计数和C反应蛋白水平也至关重要，其各自的临界值为7.0 g/dL，519000/mm

, and 10.4 mg/dL.

，和10.4 mg/dL。

The correlation between various clinical and laboratory parameters and resistance to IVIG treatment was visualized through a heatmap, which highlighted moderate to strong positive correlations for neutrophil count and C-reactive protein levels with treatment resistance, while serum protein and albumin levels exhibited negative correlations (Fig. .

各种临床和实验室参数与IVIG治疗耐药性之间的相关性通过热图可视化，热图突出了中性粒细胞计数和C反应蛋白水平与治疗耐药性的中度至强正相关，而血清蛋白和白蛋白水平呈负相关（图。

To evaluate the performance of the GBM model, a confusion matrix was created to summarize the classification results, including true positives, true negatives, false positives, and false negatives (Fig.

为了评估GBM模型的性能，创建了一个混淆矩阵来总结分类结果，包括真阳性，真阴性，假阳性和假阴性（图）。

). The model successfully classified most patients, with minimal misclassification errors.

)。该模型成功地对大多数患者进行了分类，错误分类错误最小。

Further model comparisons were conducted using precision-recall curves, which demonstrated that the GBM model outperformed other machine learning models, including LR, MLP, RF, and CATBoost, in terms of precision and recall (Fig.

使用精确召回曲线进行了进一步的模型比较，这表明GBM模型在精度和召回率方面优于其他机器学习模型，包括LR，MLP，RF和CATBoost（图）。

Fig. 4

图4

Confusion matrix evaluating the performance of the GBM model.

混淆矩阵评估GBM模型的性能。

Full size image

全尺寸图像

Fig. 5

图5

Precision-recall curve comparing the performance of different models.

精确召回曲线比较了不同模型的性能。

Full size image

全尺寸图像

Discussion

讨论

Kawasaki disease (KD) is a leading cause of acquired heart disease in children and is characterized by systemic vasculitis, which can lead to coronary artery lesions (CALs)

川崎病（KD）是儿童获得性心脏病的主要原因，其特征是全身性血管炎，可导致冠状动脉病变（CAL）

. Intravenous immunoglobulin (IVIG) combined with aspirin remains the standard treatment, significantly reducing the incidence of CALs

静脉注射免疫球蛋白（IVIG）联合阿司匹林仍然是标准治疗方法，可显着降低CAL的发生率

. However, approximately 10–20% of patients are resistant to initial IVIG treatment, necessitating alternative therapeutic strategies to mitigate the risk of severe complications such as coronary artery aneurysms and myocardial infarction

然而，大约10-20%的患者对初始IVIG治疗有抵抗力，因此需要替代治疗策略来减轻冠状动脉瘤和心肌梗死等严重并发症的风险

For patient resistant to initial IVIG treatment, the most common second-line This approach remains the standard due to its proven efficacy and safety

对于对初始IVIG治疗有抵抗力的患者，最常见的二线方法仍然是标准方法，因为它已被证明具有有效性和安全性

. Alternative treatments, such as intravenous methylprednisolone (IVMP) and infliximab (IFX), have shown promise. IVMP is effective in fever resolution, while IFX, a tumor necrosis factor (TNF) inhibitor, has demonstrated faster inflammation resolution and fewer adverse effects compared to corticosteroids, making it an increasingly preferred option in clinical settings.

替代疗法，如静脉注射甲基强的松龙（IVMP）和英夫利昔单抗（IFX），已显示出前景。IVMP在发热消退方面是有效的，而肿瘤坏死因子（TNF）抑制剂IFX与皮质类固醇相比表现出更快的炎症消退和更少的不良反应，使其成为临床环境中越来越优选的选择。

. Studies indicate that IFX not only reduces the duration of fever but also lowers the need for additional therapy and shortens hospital stays

研究表明，IFX不仅可以减少发烧的持续时间，还可以减少额外治疗的需要，缩短住院时间

. The reliance on a second IVIG dose highlights the critical need for accurate prediction models to identify patients who may not respond to this treatment, thus allowing for timely implementation of more effective alternative therapies.

对第二次IVIG剂量的依赖突出了对准确预测模型的迫切需求，以识别可能对这种治疗无反应的患者，从而可以及时实施更有效的替代疗法。

Despite extensive research on predicting initial IVIG resistance, there is a scarcity of studies specifically addressing resistance to two consecutive IVIG treatments. Existing research primarily focuses on models predicting resistance to the first IVIG treatment

尽管对预测初始IVIG耐药性进行了广泛的研究，但缺乏专门针对连续两次IVIG治疗耐药性的研究。现有的研究主要集中在预测对第一次IVIG治疗的耐药性的模型上

. Traditional scoring systems, such as the Kobayashi, Egami, and Sano scores, have been widely used but show limited efficacy in non-Japanese populations due to genetic and environmental differences

.传统的评分系统，例如小林，Egami和Sano评分，已被广泛使用，但由于遗传和环境差异，在非日本人群中的疗效有限

. Recent studies incorporated machine learning approaches to enhance prediction accuracy for initial IVIG resistance, utilizing clinical and laboratory parameters such as total bilirubin, procalcitonin, alanine aminotransferase, and platelet count

最近的研究结合了机器学习方法，利用临床和实验室参数，如总胆红素、降钙素原、丙氨酸氨基转移酶和血小板计数，提高了IVIG初始耐药性的预测准确性

. Numerous efforts have been made to develop predictive models for IVIG resistance in KD, but the results have been inconsistent, limiting their clinical applicability. These models were not widely adopted in clinical practice due to their varying accuracy across different populations and settings

。已经做出了许多努力来开发KD中IVIG耐药性的预测模型，但结果不一致，限制了它们的临床适用性。这些模型在临床实践中没有被广泛采用，因为它们在不同人群和环境中的准确性不同

High ANC levels, as revealed in our feature importance analysis, may reflect an amplified inflammatory response, characterized by increased production of pro-inflammatory cytokines (e.g., interleukin-6 and tumor necrosis factor-alpha) and reactive oxygen species

正如我们的特征重要性分析所揭示的那样，高ANC水平可能反映了一种放大的炎症反应，其特征是促炎细胞因子（例如白细胞介素-6和肿瘤坏死因子-α）和活性氧的产生增加

,27]

. This hyperinflammatory state can exacerbate vascular damage and endothelial dysfunction, reducing the therapeutic efficacy of IVIG

这种高炎症状态会加剧血管损伤和内皮功能障碍，降低IVIG的治疗效果

. Similarly, low serum protein and albumin levels are associated with capillary leakage and systemic inflammation, both of which impair IVIG distribution and function

同样，低血清蛋白和白蛋白水平与毛细血管渗漏和全身炎症有关，这两者都会损害IVIG的分布和功能

. Elevated CRP levels, a hallmark of acute inflammation, indicate persistent immune activation and endothelial injury, which are strongly associated with IVIG resistance

CRP水平升高是急性炎症的标志，表明持续的免疫激活和内皮损伤，这与IVIG耐药性密切相关

. These mechanisms suggest that inflammatory and hematologic imbalances play a central role in mediating resistance to two consecutive IVIG treatments.

这些机制表明，炎症和血液学失衡在介导对两种连续IVIG治疗的耐药性中起着核心作用。

Emerging evidence also suggests that genetic factors may contribute to IVIG resistance. Polymorphisms in genes encoding Fc receptors (e.g., FCGR2A and FCGR3B) have been implicated in altered IVIG binding and clearance, potentially impacting treatment efficacy

新出现的证据还表明，遗传因素可能导致IVIG耐药性。Fc受体编码基因（如FCGR2A和FCGR3B）的多态性与IVIG结合和清除的改变有关，可能影响治疗效果

8,10)

. Additionally, dysregulated activation of immune cells, such as monocytes and macrophages, could amplify the production of pro-inflammatory cytokines, further exacerbating the hyperinflammatory state

此外，免疫细胞（如单核细胞和巨噬细胞）的激活失调可能会放大促炎细胞因子的产生，进一步加剧高炎症状态

Our study contributes to this gap by developing and validating machine learning models specifically for predicting resistance to predicting treatment resistance. The GBM model in our study demonstrated the highest AUROC of 0.791, showcasing its robust predictive performance. Key predictors identified in our models included white blood cells, fever duration, and serum albumin levels, aligning with findings from other studies that highlighted the relevance of these factors in predicting IVIG resistance.

我们的研究通过开发和验证专门用于预测耐药性的机器学习模型来预测治疗耐药性，从而弥补了这一差距。我们研究中的GBM模型显示出最高的AUROC为0.791，显示出其强大的预测性能。在我们的模型中确定的关键预测因子包括白细胞，发热持续时间和血清白蛋白水平，与其他研究的结果一致，这些研究强调了这些因素在预测IVIG耐药性方面的相关性。

The ability to predict which patients will not respond to treatment resistance can significantly enhance clinical decision-making. Early identification allows for the timely implementation of alternative therapeutic strategies such as corticosteroids or IFX, potentially improving patient outcomes and reducing the incidence of coronary artery complications.

预测哪些患者对治疗抵抗没有反应的能力可以显着增强临床决策。早期识别可以及时实施替代治疗策略，如皮质类固醇或IFX，有可能改善患者预后并降低冠状动脉并发症的发生率。

. Given the variability in response rates and the associated risks, personalized treatment plans based on predictive models can optimize patient care.

。

This study is one of the first to focus specifically on predicting resistance to two consecutive IVIG treatments, addressing a significant gap in the existing literature. While previous research has primarily centered on resistance to the initial IVIG dose, our approach offers new insights into managing patients who may require multiple rounds of therapy.

这项研究是第一个专门关注预测对两种连续IVIG治疗的耐药性的研究之一，解决了现有文献中的重大差距。虽然之前的研究主要集中在对初始IVIG剂量的抵抗力上，但我们的方法为管理可能需要多轮治疗的患者提供了新的见解。

We hope this contributes to a better understanding and more effective management of KD..

我们希望这有助于更好地理解和更有效地管理KD。。

One of the strengths of this study is the use of ensemble models like GBM and Random Forest, which combine the predictive power of multiple decision trees. This approach likely contributed to the superior performance of these models compared to simpler models like Logistic Regression and Multi-Layer Perceptron (MLP).

这项研究的优势之一是使用GBM和随机森林等集成模型，它们结合了多个决策树的预测能力。与逻辑回归和多层感知器（MLP）等简单模型相比，这种方法可能有助于这些模型的优异性能。

Ensemble models have the advantage of capturing complex interactions between variables, which are critical for accurately predicting treatment resistance in a heterogeneous disease like KD..

集合模型具有捕获变量之间复杂相互作用的优势，这对于准确预测KD等异质性疾病的治疗耐药性至关重要。。

Feature importance analysis revealed that ANC, serum protein, platelet count, and CRP were the most significant factors in predicting resistance to two consecutive IVIG treatments. These results highlight the critical role of inflammatory markers and hematologic parameters in assessing treatment resistance, which could inform early intervention strategies for high-risk patients..

特征重要性分析显示，ANC，血清蛋白，血小板计数和CRP是预测连续两次IVIG治疗耐药性的最重要因素。这些结果突出了炎症标志物和血液学参数在评估治疗耐药性中的关键作用，这可以为高危患者的早期干预策略提供信息。。

However, this study has some limitations in its dataset and methodology. Potential biases introduced during the data collection process, such as differences in baseline characteristics across participating institutions, could affect the generalizability of the findings. Additionally, the retrospective nature of the dataset may limit the ability to control for all confounding factors.

然而，这项研究在数据集和方法上有一些局限性。在数据收集过程中引入的潜在偏差，例如参与机构之间基线特征的差异，可能会影响调查结果的普遍性。此外，数据集的回顾性可能会限制控制所有混杂因素的能力。

These limitations should be addressed in future studies by incorporating more diverse datasets and prospective study designs..

这些局限性应该在未来的研究中通过纳入更多不同的数据集和前瞻性研究设计来解决。。

In comparison to previous studies, which predominantly focused on resistance to the first IVIG treatment, this study provides a novel approach by predicting resistance to two consecutive treatments. Traditional scoring systems, such as the Kobayashi and Egami scores, have shown limited applicability across different populations.

与之前主要关注对第一次IVIG治疗的耐药性的研究相比，本研究通过预测对两次连续治疗的耐药性提供了一种新方法。传统的评分系统，如小林和Egami评分，在不同人群中的适用性有限。

By leveraging machine learning, our model demonstrated improved predictive accuracy with an AUROC of 0.791 and identified key predictors such as ANC, serum protein, and CRP levels that could guide more personalized treatment strategies. Given the scarcity of research focusing on resistance to two consecutive IVIG treatments, this study provides valuable insights into the management of KD.

通过利用机器学习，我们的模型证明了改进的预测准确性，AUROC为0.791，并确定了关键预测因子，如ANC，血清蛋白和CRP水平，可以指导更个性化的治疗策略。鉴于缺乏针对连续两次IVIG治疗耐药性的研究，本研究为KD的管理提供了有价值的见解。

Future research should focus on validating these predictive models in diverse populations to ensure their generalizability.

未来的研究应侧重于在不同人群中验证这些预测模型，以确保其普遍性。

. Integrating genetic and immunological biomarkers could further enhance prediction accuracy. Additionally, developing standardized guidelines for managing IVIG-resistant KD based on robust clinical evidence will be crucial for improving patient outcomes

整合遗传和免疫生物标志物可以进一步提高预测准确性。此外，根据强有力的临床证据制定管理IVIG耐药KD的标准化指南对于改善患者预后至关重要

Data availability

数据可用性

The dataset used and analyzed in this study is available from the corresponding author upon reasonable request.

本研究中使用和分析的数据集可根据合理要求从通讯作者处获得。

References

参考文献

Nakamura, Y. et al. Epidemiologic features of Kawasaki disease in Japan: results from the nationwide survey in 2005–2006.

Nakamura，Y.等人。日本川崎病的流行病学特征：2005-2006年全国调查的结果。

J. Epidemiol.

J、流行病。

, 167–172 (2008).

Article

文章

PubMed

PubMed Central

公共医学中心

MATH

数学

Google Scholar

谷歌学者

Kuo, H. C. et al. Serum albumin level predicts initial intravenous immunoglobulin treatment failure in Kawasaki disease.

Kuo，H.C.等人。血清白蛋白水平预测川崎病初始静脉注射免疫球蛋白治疗失败。

Acta Paediatr.

儿科记录。

, 1578–1583 (2010).

Article

文章

PubMed

MATH

数学

Google Scholar

谷歌学者

Kato, H. et al. Kawasaki disease: Effect of treatment on coronary artery involvement.

Kato，H.等人，《川崎病：治疗对冠状动脉受累的影响》。

Pediatrics

儿科学

, 175–179 (1979).

Article

文章

CAS

中科院

PubMed

MATH

数学

Google Scholar

谷歌学者

Newburger, J. W. et al. A single intravenous infusion of gamma globulin as compared with four infusions in the treatment of acute Kawasaki syndrome.

Newburger，J.W.等人。单次静脉输注丙种球蛋白与四次输注治疗急性川崎综合征的比较。

N Engl. J. Med.

N英语。医学杂志。

324

, 1633–1639 (1991).

Article

文章

CAS

中科院

PubMed

MATH

数学

Google Scholar

谷歌学者

Miura, M., Ishii, M., Uemura, S. & Kato, H. Comparison of second-line therapies in IVIg-refractory Kawasaki disease.

Miura，M.，Ishii，M.，Uemura，S。＆Kato，H。IVIg难治性川崎病二线治疗的比较。

Pediatr. Rheumatol.

儿科。风湿病。

, 12–20 (2018).

Google Scholar

谷歌学者

Dallaire, F. & Burns, J. C. IVIG-resistant Kawasaki disease: An overview of recent studies.

Dallaire，F。＆Burns，J.C。IVIG抗性川崎病：最近研究的概述。

Clin. Rheumatol.

临床。风湿病。

, 2231–2240 (2019).

MATH

数学

Google Scholar

谷歌学者

Harada, K. Intravenous gamma-globulin treatment in Kawasaki disease.

Harada，K。静脉注射丙种球蛋白治疗川崎病。

Acta Paediatr. Jpn

《儿科学报》。Jpn

, 715–719 (1987).

MATH

数学

Google Scholar

谷歌学者

Egami, K. et al. Prediction of resistance to intravenous immunoglobulin treatment in patients with Kawasaki disease.

Egami，K。等人。川崎病患者静脉注射免疫球蛋白治疗耐药性的预测。

J. Pediatr.

J.儿科。

143

, 493–499 (2003).

MATH

数学

Google Scholar

谷歌学者

Kobayashi, T. et al. Prediction of intravenous immunoglobulin unresponsiveness in patients with Kawasaki disease.

Kobayashi，T。等人。川崎病患者静脉注射免疫球蛋白无反应的预测。

Circulation

流通

113

, 2606–2612 (2006).

Article

文章

PubMed

MATH

数学

Google Scholar

谷歌学者

Rowley, A. H. & Shulman, S. T. Pathogenesis and management of Kawasaki disease.

Rowley，A.H。＆Shulman，S.T。川崎病的发病机制和管理。

Expert Rev. Anti Infect. Ther.

专家版抗感染。他们。

, 197–203 (2010).

Article

文章

PubMed

PubMed Central

公共医学中心

MATH

数学

Google Scholar

谷歌学者

McCrindle, B. W. et al. Diagnosis, treatment, and long-term management of Kawasaki disease: A scientific statement for health professionals from the American Heart Association.

McCrindle，B.W.等人，《川崎病的诊断、治疗和长期管理：美国心脏协会卫生专业人员的科学声明》。

Circulation

流通

135

, e927–e999 (2017).

，e927–e999（2017）。

Article

文章

PubMed

MATH

数学

Google Scholar

谷歌学者

Burns, J. C. et al. Kawasaki disease: A brief history.

Burns，J.C.等人，《川崎病：简史》。

Pediatr. Int.

儿科。内景。

, 13–15 (2018).

MATH

数学

Google Scholar

谷歌学者

Furusho, K. et al. High-dose intravenous gammaglobulin for Kawasaki disease.

。

Lancet

柳叶刀

, 1055–1058 (1984).

Article

文章

CAS

中科院

PubMed

MATH

数学

Google Scholar

谷歌学者

Kanai, T. et al. New risk scoring system for predicting intravenous immunoglobulin resistance in patients with Kawasaki disease.

Kanai，T。等人。用于预测川崎病患者静脉注射免疫球蛋白耐药性的新风险评分系统。

Pediatr. Infect. Dis. J.

儿科。感染。说吧。J。

, 20–25 (2019).

MATH

数学

Google Scholar

谷歌学者

Dionne, A. & Dahdah, N. Kawasaki disease: Summary of the American Heart Association guidelines.

Dionne，A。＆Dahdah，N。川崎病：美国心脏协会指南摘要。

Can. Fam Physician

可以。Fam医师

, 25–31 (2018).

MATH

数学

Google Scholar

谷歌学者

Research Committee on Kawasaki Disease. Report of Subcommittee on standardization of diagnostic criteria and reporting of coronary artery lesions in Kawasaki Disease.

川崎病研究委员会。川崎病诊断标准标准标准化和冠状动脉病变报告小组委员会的报告。

Ministry Health Welfare: Tokyo Japan

卫生福利部：日本东京

(1984).

Newburger, J. W. et al. Diagnosis, treatment, and long-term management of Kawasaki disease: A statement for health professionals from the Committee on Rheumatic Fever, Endocarditis, and Kawasaki Disease, Council on Cardiovascular Disease in the Young, American Heart Association.

Newburger，J.W.等人，《川崎病的诊断、治疗和长期管理：美国心脏协会年轻心血管疾病委员会风湿热、心内膜炎和川崎病委员会卫生专业人员声明》。

Circulation

流通

110

, 2747–2771 (2004).

Article

文章

PubMed

MATH

数学

Google Scholar

谷歌学者

Anderson, M. S., Todd, J. K. & Glode, M. P. Delayed diagnosis of Kawasaki syndrome: An analysis of the problem.

Anderson，M.S.，Todd，J.K。＆Glode，M.P。川崎综合征的延迟诊断：问题分析。

Pediatrics

儿科学

115

, e428–e433 (2005).

，e428–e433（2005年）。

Article

文章

PubMed

MATH

数学

Google Scholar

谷歌学者

Lee, S. et al. A novel risk score to predict intravenous immunoglobulin resistance in Kawasaki disease: A multicenter study.

Lee，S.等人。预测川崎病静脉注射免疫球蛋白耐药性的新风险评分：一项多中心研究。

Int. J. Cardiol.

国际心脏病杂志。

236

, 158–162 (2017).

ADS

MATH

数学

Google Scholar

谷歌学者

Durongpisitkul, K., Gurleen, A., Park, J. M. & Martin, C. F. The prevention of coronary artery aneurysm in Kawasaki disease: A meta-analysis on the efficacy of aspirin and immunoglobulin treatment.

Durongpisitkul，K.，Gurleen，A.，Park，J.M。和Martin，C.F。川崎病冠状动脉瘤的预防：阿司匹林和免疫球蛋白治疗效果的荟萃分析。

Pediatrics

儿科学

, 1057–1061 (1995).

Article

文章

CAS

中科院

PubMed

Google Scholar

谷歌学者

Wallace, C. A. et al. Initial intravenous gammaglobulin treatment failure in Kawasaki disease.

Wallace，C.A.等人，《川崎病初始静脉丙种球蛋白治疗失败》。

Pediatrics

儿科学

105

, e78 (2000).

，e78（2000）。

Article

文章

CAS

中科院

PubMed

MATH

数学

Google Scholar

谷歌学者

Kato, H. et al. Long-term consequences of Kawasaki disease. A 10- to 21-year follow-up study of 594 patients.

Kato，H.等人，《川崎病的长期后果》。对594名患者进行了10至21年的随访研究。

Circulation

流通

, 1379–1385 (1996).

Article

文章

CAS

中科院

PubMed

MATH

数学

Google Scholar

谷歌学者

Matsubara, K. & Fukaya, T. The role of interleukin-1 in Kawasaki disease.

Matsubara，K。＆Fukaya，T。白细胞介素-1在川崎病中的作用。

Rheumatology

风湿病学

, 10–14 (1996).

MATH

数学

Google Scholar

谷歌学者

Suzuki, A. et al. Coronary arterial lesions of Kawasaki disease: cardiac catheterization findings of 1,100 cases.

铃木，A。等。川崎病冠状动脉病变：1100例心导管检查结果。

Pediatr. Cardiol.

儿科。心脏病。

, 3–9 (1986).

Article

文章

CAS

中科院

PubMed

MATH

数学

Google Scholar

谷歌学者

Burns, J. C. et al. Clinical and epidemiologic characteristics of patients referred for possible Kawasaki disease. United States and Canada, 1976–2004.

Burns，J.C.等人。可能患有川崎病的患者的临床和流行病学特征。美国和加拿大，1976年至2004年。

Pediatr. Infect. Dis. J.

儿科。感染。说吧。J。

, 260–263 (2005).

MATH

数学

Google Scholar

谷歌学者

Download references

下载参考资料

Author information

作者信息

Authors and Affiliations

作者和隶属关系

Chungbuk National University Hospital, Cheongju, Korea

韩国清州忠北国立大学医院

Eun Jung Cheon & Seung Park

尹正千

Seoul National University Children’s Hospital, Seoul, Korea

首尔国立大学儿童医院，韩国首尔

Gi Beom Kim

金

Department of Pediatrics, Chungbuk National University Hospital, 776, 1 Sunhwan-ro, Seowon-gu, Cheongju-si, Chungcheongbuk-do, Republic of Korea

中北国立大学医院儿科，776，1 Sunhwan ro，Seowon gu，Cheongju si，中北道，大韩民国

Eun Jung Cheon

尹正千

Authors

作者

Eun Jung Cheon

尹正千

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Gi Beom Kim

金

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Seung Park

胜公园

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Contributions

捐款

Conceptualization: E.J.C, G.B.K.; Methodology: E.J.C, G.B.K, S.P.; Formal analysis: E.J.C, G.B.K, S.P.; Investigation: E.J.C, G.B.K; Writing—originaldraft preparation: E.J.C, G.B.K, S.P.; Writing—review and editing: E.J.C, G.B.K, S.P.M.O.R.; Supervision: E.J.C. All authors read and approved the final manuscript.Korean Society of Kawasaki disease members collected the data..

概念化：E.J.C，G.B.K。；方法学：E.J.C，G.B.K，S.P。；形式分析：E.J.C，G.B.K，S.P。；调查：E.J.C，G.B.K；写作原稿准备：E.J.C，G.B.K，S.P。；写作评论和编辑：E.J.C，G.B.K，S.P.M.O.R。；监督：E.J.C.所有作者都阅读并批准了最终手稿。韩国川崎病学会成员收集了数据。。

Corresponding author

通讯作者

Correspondence to

通信对象

Eun Jung Cheon

尹正千

Ethics declarations

道德宣言

Competing interests

相互竞争的利益

The authors declare no competing interests.

作者声明没有利益冲突。

Additional information

其他信息

Publisher’s note

出版商注释

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Springer Nature在已发布的地图和机构隶属关系中的管辖权主张方面保持中立。

Rights and permissions

权限和权限

Open Access

开放存取

This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material.

本文是根据知识共享署名非商业性NoDerivatives 4.0国际许可证授权的，该许可证允许以任何媒介或格式进行任何非商业性使用，共享，分发和复制，只要您对原始作者和来源给予适当的信任，提供知识共享许可证的链接，并指出您是否修改了许可材料。

You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.

To view a copy of this licence, visit .

要查看此许可证的副本，请访问。

http://creativecommons.org/licenses/by-nc-nd/4.0/

Reprints and permissions

重印和许可

About this article

关于本文

Cite this article

引用本文

Cheon, E.J., Kim, G.B. & Park, S. Predictive modeling of consecutive intravenous immunoglobulin treatment resistance in Kawasaki disease: A nationwide study.

Cheon，E.J.，Kim，G.B。＆Park，S。川崎病连续静脉注射免疫球蛋白治疗耐药性的预测模型：一项全国性研究。

Sci Rep

Sci代表

, 903 (2025). https://doi.org/10.1038/s41598-025-85394-4

, 903 (2025).https://doi.org/10.1038/s41598-025-85394-4

Download citation

下载引文

Received

已接收

12 August 2024

2024年8月12日

Accepted