丙球无反应型川崎病危险因素分析及风险评分模型构建

Risk Factors of Intravenous Immunoglobulin-Resistant Kawasaki Disease and Establishment of Risk Scoring Model

目的分析丙球无反应型川崎病危险因素,建立Logistic回归预测模型,评估川崎病患儿丙球无反应的可能性,提高可疑丙球无反应型川崎病的临床预测水平,为可疑丙球无反应型川崎病患者及早确定有效治疗方案提供充分依据。方法回顾性分析2014年1月~2016年9月在武汉儿童医院就诊并符合该研究标准的川崎病患儿637例临床资料,其中丙球有反应594例,丙球无反应43例。按照入院先后顺序,将2014年1月~2015年12月收治的493例作为建模组,2016年1~9月的144例作为验证组。分析建模组病例月龄、白细胞数、血小板计数、血红蛋白、降钙素原、B型脑钠肽等指标对诊断川崎病丙球无反应的价值,利用SPSS软件建立Logistics回归模型。利用验证组病例数据建立Logistic回归模型预测概率的受试者工作特征曲线(ROC),分析其灵敏度、特异度,确定临界值。利用两组数据构建简单评分模型,分析其ROC曲线下面积、灵敏度和特异度。结果丙球无反应型川崎病Logistic回归模型以临床结局对丙种球蛋白是否有反应为因变量,将月龄、血常规、尿常规、C反应蛋白、血沉、生化指标、免疫功能、病原学指标等作为考察变量。最终进入Logistic回归预测模型的有血红蛋白、前白蛋白、B型脑钠肽、谷草转氨酶和免疫球蛋白G(均P<0.05)。训练样本和验证样本生成的ROC曲线下面积分别为0.795和0.870,其灵敏度、特异度分别为73.5%、76.5%和100.0%、70.4%。建立简单评分系统,取截断值,降钙素原≥1.35ng/mL、谷草转氨酶≥42.0U/L、白蛋白≤34.0g/L、B型脑钠肽≥835pg/mL,分值分别为1.5、1.5、1.0、1.0,若分数≥2.5,为丙球无反应型川崎病高风险组,其灵敏度、特异度分别为72.1%、74.6%。结论联合降钙素原、白蛋白、B型脑钠肽和谷草转氨酶4个指标,利用简单评分模型有助于预测患儿丙球无反应的风险,是一种准确率较高的辅助诊断工具。

Objective To explore the immunoglobulin resistance in children with Kawasaki disease(KD)by establishing a Logistic regression model and analyzing the risk factors of intravenous immunoglobulin(IVIG)-resistant KD,in order to accurately diagnose this condition and provide effective treatment strategies.Methods Clinical data of 637 KD patients who were treated from January 2014 to September 2016 in Wuhan Children’s hospital were retrospectively analyzed.There were 594 patients responsive to IVIG and 43 not.Among them,493 cases admitted between January 2014 and December 2015 felt into the modeling group and 144 between January 2016 and September 2016 into the verification group.Age,white blood cell count,platelet count,hemoglobin(Hb)level,procalcitonin(PCT)and NT-proB-type natriuretic peptide(NT-proBNP),etc.,were analyzed in the modeling group to examine their role in the diagnosis of IVIG-resistant KD.Logistic regression model was established with data from the verification group to obtain the receiver operating characteristic(ROC)curve,analyze the sensitivity and specificity of the ROC curve and identify the cut-off value.Additionally,a simple scoring model was constructed on the basis of the data from both groups in order to analyze the area under the ROC and the sensitivity and specificity of the ROC curve.Results The Logistic regression model of IVIG-resistant KD was established with responsiveness to IVIG as a dependent variable and age,blood routine,urinary routine,CRP,ESR,biochemical indicators,immune function,and pathogenic indicators and so on as study variables.Hb,pre-albumin,NT-proBNP,AST and IgG eventually entered the Logistic regression model(P<0.05),which generated an area under the curve of 0.795 and 0.870 for the modeling group and verification group,respectively.The sensitivity and specificity of the ROC curve were 73.5%,76.5%and 100.0%,70.4%in the modeling group and verification group,respectively.Four variables were used to generate a simple scoring model,including NT-proBNP,AST,albumin and PCT.The scores of PCT≥1.35 ng/mL,AST≥42.0 U/L,albumin≤34.0 g/L,and BNP≥835 pg/mL were 1.5,1.5,1.0,and 1.0,respectively.Patients were considered at high risk of IVIG-resistant KD if the score was equal to or more than 2.5.The sensitivity and specificity of the ROC curve were 72.1%and 74.6%in the simple scoring model.Conclusion Use of a simple scoring model combined with PCT,albumin,NT-proBNP and AST helps to predict the risk of IVIG-resistant KD.It is a highly accurate auxiliary diagnosis tool.