系统性红斑狼疮自身抗体谱数据的解读与疾病模型预测

Interpretation and predictive modeling of antibody spectrum data for system lupus erythematosus

目的通过数据挖掘建立系统性红斑狼疮(SLE)的最佳预测模型,以及自身抗体谱检测报告标准化解读数据库,以提高诊断效率。方法收集绵阳市中心医院2015年1月至2016年9月首次送检自身抗体谱检测且诊断明确的受试者8 904例,其中SLE患者668例,其他自身免疫性疾病(AID)1 279例,非AID 6 957例。通过受试者工作特征曲线(ROC)分析,从年龄、性别及16种自身抗体中筛选出对SLE具有预测价值的指标后,分别采用决策树、Logistic回归和人工神经网络(ANN)建立SLE预测模型,并选择最优模型,建立其不同预测结果下的验后概率、误诊率等性能指标的扩展数据库。结果经ROC分析,年龄、性别以及抗核抗体(ANA)、抗干燥综合征A抗体(SSA)、抗核糖核蛋白抗体(nRNP/Sm)、抗干燥综合征Ro-52抗体(Ro-52)、抗组蛋白抗体(Histone)、抗核小体抗体(Nuclesome)、抗核糖体抗P蛋白抗体(Rib·P)、抗双链DNA抗体(dsDNA)、抗Sm抗体(Sm)、抗干燥综合征B抗体(SSB)和抗线粒体M2抗体(AMA-M2)均存在一定的诊断价值(P<0.01)。3种预测模型中,Logistic模型优于其他两种模型以及任何单项自身抗体检测,差异有统计学意义(P<0.05)。结论自身抗体谱联合性别、年龄因素建立Logistic模型,并根据相应地区的患病率建立预测性能指标数据库,可有效提高疾病的诊断效率。

Objective To establish the optimum predictive model of systemic lupus erythematosus(SLE)and database for the standardized interpretation of the autoantibodies test report by data mining,so as to improve the diagnostic efficiency.Methods Autoantibodies test results of the 8 904 subjects who were detected at first time and had a definite diagnosis were collected from January 2015 to September 2016,including 668 cases of SLE patients,1 279 cases of other autoimmune disease(AID)patients and 6 957 cases of non-AID patients.ROC curve analysis was used to screen valuable indexes for diagnosis of SLE from age,sex and 16 kinds of autoantibodies,followed by application of decision tree,Logistic regression and artificial neural network(ANN)to establish predictive models of SLE respectively.The optimal model was selected,and the extended database of performance indicators such as posterior probability,misdiagnosis rate and missed diagnosis rate was established for interpretation of the antibodies test report.Results According to the analysis of ROC curve,age,gender,ANA,SSA,nRNP/Sm,Ro-52,Histone,Nuclesome,Rib·P,ds-DNA,Sm,SSB and AMAM2 showed application value in different degrees(P<0.01).The Logistic model was better than the other two models,as well as any single antibody test,and the difference was statistically significant(P<0.05).Conclusion The Logistic model established by combining the antibodies with the sex and age factors and the prediction performance index database established according to the local prevalence,could effectively improve the diagnostic efficiency of the disease.