Logistic回归模型在早期鼻咽癌多指标联合辅助诊断实验的应用  

Application of logistic regression in combination with multiple diagnostic tests for auxiliary diagnosis of nasopharyngeal carcinoma

作　　者：姜世强[1] 柳青[2,3] 

机构地区：[1]中山大学公共卫生学院医学统计与流行病学系,广东广州510080 [2]华南肿瘤学国家重点实验室,广东广州510060 [3]中山大学肿瘤防治中心流行病学室,广东广州500060

出　　处：《癌症》2009年第2期213-216,共4页Chinese Journal of Cancer

基　　金：国家"十一五"科技支撑计划(No:2006BAI02A11)~~

摘　　要：背景与目的:目前用于鼻咽癌辅助诊断的指标较多,但大部分早期诊断的效果较差。如何联合多个指标进行辅助诊断是近年研究的重点。本研究旨在评价鼻咽癌常用诊断指标VCA/IgA、EA/IgA、EBV DNA、EBNA1/IgA、EBNA1/IgG和ZTA/IgG的诊断价值,并采用logistic回归筛选优化的联合诊断组合。方法:对81例未经治疗的病理确诊鼻咽癌患者和89例健康体检者分别采用实时荧光定量PCR法测定EB病毒DNA,计算病毒拷贝数,采用间接免疫酶法测定EBV、VCA/IgA和EA/IgA,采用酶联吸附试验(ELISA法)检测EBNA1/IgG、EBNA1/IgA以及ZTA/IgG。通过ROC曲线评价各指标单独和联合诊断鼻咽癌的效果。用logistic回归综合两指标信息,根据ROC曲线确定诊断界值,以增加鼻咽癌诊断的准确性。结果:logistic回归方法的诊断敏感度和特异度比常规的并联实验均有提高。筛选的优化组合方案为EBV DNA+EBNA1/IgA和VCA/IgA+EBNA1/IgA,两组合的并联试验敏感度和特异度分别为0.96、0.82和1.00、0.84,而logistic回归预测概率为诊断指标,ROC曲线确定界值时,两组合的灵敏度和特异度分别为1.00、0.87和0.98、0.88。结论:多指标联合建立logistic回归模型,利用预测概率进行辅助诊断能够达到同时提高灵敏度和特异度的效果,可供临床参考。Background and Objective. Although there are many markers for the clinical diagnosis of nasopharyngeal carcinoma （NPC）, the efficacy of most of the markers for the early diagnosis is poor. This study was to evaluate the diagnostic value of VCA/IgA, EA/IgA, EBV DNA, EBNA1/IgA, EBNA1/IgG and ZTA/IgG for NPC, as well as to screen out an optimized combination to increase diagnostic accuracy of NPC. Methods. Eight-one newly pathologically diagnosed NPC patients prior to treatment and 89 health cases from the routine physical checks were entered into the study. Epstein Barr Virus （EBV） DNA was detected by quantitative real-time PCR. VCA/IgA and EA/IgA were assessed by immunofluorescence assays （IFA）. The receiver operating characteristic （ROC） curve and the area under the curve were used to evaluate the diagnostic value of a single test or combined tests for NPC, thus to decide the cut-off value. The logistic regression model was used to combine the results from multiple tests to increase the accuracy. Results. Comparing to the routine parallel sequential test, combination tests achieved a higher specificity and sensitivity. Two optimal combinations were EBV DNA ＋ EBNA1/IgA and VCA/IgA ＋ EBNA1/IgA, whose sensitivity and specificity reached 0.96 and 0.82, 1.00 and 0.84, respectively. When the logistic model was used and the cut-off value was decided by ROC, the sensitivity and specificity of the two combination group became 1.00 and 0.87, 0.98 and 0.88, respectively. Conclusions; Combining multiple test results by a logistic model and using the probability prediction to decide the cut-off value may help increase the diagnostic sensitivity and specificity of NPC.

关 键 词：鼻咽肿瘤 EB病毒 诊断指标 LOGISTIC回归 联合诊断 敏感度 特异度 

分 类 号：R739.63[医药卫生—肿瘤]