Abstract: Background　At present Kawasaki disease （KD） has become the leading cause of acquired heart disease in children.Although intravenous immunoglobulin （IVIG） combined with aspirin were commonly adopted as the standard therapy protocol in KD which can effectively decrease the incidence of coronary artery lesions （CALs），some KD patients are resistant to IVIG therapy and they are at higher risk of developing CALs than patients who response to IVIG therapy.Therefore，it is very important to predict IVIG-resistance indicators in patients with KD.Objective　To analyze the risk factors of IVIG resistance in KD，develop a new predictive model to identify IVIG resistance，and evaluate the efficacy of the new predictive model compared with other four existing predictive models（Egami，Kobayashi，Fu and Lin）.Methods　A retrospective analysis was made on the clinical data of 173 KD children diagnosed，treated and followed up in Jiangxi Provincial Children's Hospital from January 2016 to June 2018.The patients were divided into IVIG resistant group and IVIG response group.Multivariate Logistic regression was used to analyze the independent risk factors of IVIG resistance.A new predictive model was established by assignment.Four other predictive models were used to evaluate the subjects.The predictive efficacy of the new model was compared with the other four predictive models through the receiver operating characteristic （ROC） curve.Results　Of the 173 cases with KD，31 cases（17.9%）were IVIG resistance and 142（82.1%）were IVIG response.Compared with the IVIG response group，IVIG resistance group had significantly higher percentage of male and higher white blood cell count （WBC），neutrophil count （N），percentage of neutrophils ，C-reactive protein （CRP） and ferritin （SF） levels and lower serum sodium level（P<0.05）.Multivariable Logistic regression analysis showed that male〔OR=5.151，95%CI（1.583，16.756）〕，CRP〔OR=1.166，95%CI（1.070，1.272）〕，sodium level〔OR=0.837，95%CI（0.719，0.914）〕 and SF〔OR=1.192，95%CI（1.079，1.318）〕 were independent risk factors of IVIG resistance in KD（P<0.05）.The cut-off values of each indicator were calculated by ROC curve.The results showed that the cut-off values of CRP，blood Na and SF were 94 mg/L，135 mmol/L and 148 μg/L，respectively.Taking 4 point as the cut off value，and the sensitivity and specificity were 79.9% and 53.3%.The sensitivity and specificity of the Egami model for the data were 38.7% and 73.8%，respectively.The sensitivity and specificity of the Kobayashi model were 35.5% and 88.7%，respectively.The sensitivity and specificity of the Fu model were 48.4% and 78.7%，respectively.The sensitivity and specificity of the Lin model are 77.4% and 55.3%，respectively.Conclusion　Male，sodium，CRP and SF are independent risk factors of IVIG resistance in KD.The new predictive model have a higher sensitivity and prediction efficacy than models of Egami，Kobayashi，Fu and Lin.

Key words: Mucocutaneous lymph node syndrome, Intravenous immunoglobulin resistance, Risk factors, Predictive models, ROC curve

摘要： 背景　目前，川崎病（KD）已成为儿童获得性心脏病的主要病因。丙种球蛋白（IVIG）联合阿司匹林作为KD的标准治疗方案，可使冠状动脉损害发生率降低，但仍有部分患儿对IVIG不敏感，而此类患儿的冠状动脉损害发生率更高。因此，研究KD发生IVIG耐药的预测指标十分重要。目的　探讨KD发生IVIG耐药的危险因素并建立其预测模型，比较新模型和其他4种IVIG耐药预测模型（Egami、Kobayashi、Fu和Lin）的预测效能。方法　回顾性分析2016年1月—2018年6月于江西省儿童医院确诊并经规范治疗及随诊的173例初诊KD患儿的临床资料，根据患儿是否发生IVIG耐药，分为IVIG耐药组和IVIG敏感组。采用多因素Logistic回归分析IVIG耐药的独立危险因素，通过赋值建立新的预测模型，并应用其他4种耐药预测模型对入组患者进行评估，通过受试者工作特征（ROC）曲线比较新模型与其他4种预测模型的预测效能。结果　173例KD患儿中，IVIG耐药组31例（17.9%），IVIG敏感组142例（82.1%）。IVIG耐药组男性患儿比例高于IVIG敏感组，白细胞计数（WBC）、中性粒细胞计数（N）、中性粒细胞分数、C反应蛋白（CRP）、铁蛋白（SF）水平高于IVIG敏感组，血Na水平低于IVIG敏感组（P<0.05）。多因素Logistic回归分析结果显示，男性〔OR=5.151，95%CI（1.583，16.756）〕、CRP〔OR=1.166，95%CI（1.070，1.272）〕、血Na〔OR=0.837，95%CI（0.719，0.974）〕和SF〔OR=1.192，95%CI（1.079，1.318）〕是IVIG耐药的独立危险因素（P<0.05）。通过ROC曲线计算各指标的截断值，结果显示，CRP、血Na和SF的截断值分别为94 mg/L、135 mmol/L和148 μg/L。通过对CRP≥94 mg/L和血Na≤135 mmol/L赋值1分，男性和SF≥148 μg/L赋值2分建立新的预测模型，以总分4分为截断值，其灵敏度和特异度分别为79.9%和53.3%。而Egami模型对本组资料的灵敏度和特异度分别为38.7%和73.8%，Kobayashi模型的灵敏度和特异度分别为35.5%和88.7%，Fu模型的灵敏度和特异度分别为48.4%和78.7%，Lin模型的灵敏度和特异度分别为77.4%和55.3%。结论　男性、血Na、CRP和SF是KD发生IVIG耐药的独立危险因素。联合以上4个指标建立的新模型具有较高灵敏度，较Egami、Kobayashi、Fu和Lin 4种IVIG耐药预测模型具有更好的预测效能。

关键词: 黏膜皮肤淋巴结综合征, 丙种球蛋白耐药, 危险因素, 预测模型, ROC曲线