双变量模型在诊断试验Meta分析中的应用

Application of Bivariate Analysis of Sensitivity and Specificity in Meta-Analysis

目的阐述灵敏度和特异度双变量Meta分析模型对于诊断试验的评价效果,为选择更好的诊断试验评价方法提供依据。方法通过最近发表的评价2种血吸虫免疫学诊断方法的Meta分析数据来阐述双变量随机效应模型分析方法。结果双变量模型可直接得出待比较的几种诊断方法的logit灵敏度、特异度及诊断优势比(DOR)均值及各自的95%CI,还可得出灵敏度均值、特异度均值以及DOR均值在2种诊断方法间有无统计学差异,同时还能合并估计不同研究间灵敏度和特异度可能存在的关联性。结论双变量模型保留了数据的二维特性,可以将灵敏度和特异度的效应分开估计,这比传统的SROC法得出诊断优势比的净效应更加合理。因此,双变量模型用于诊断试验评价是恰当的、灵活的,可作为经典SROC法的改进与补充。

Objective To illustrate the evaluation effect of bivariate analysis of sensitivity and specificity meta-analysis model in diagnosis test to provide basis for selecting better evaluation method of diagnostic test. Methods Bivariate model was presented by reanalyzing the data from a published meta analysis of two diagnostic techniques in diagnosis of schistosomiasis japonica. Results The bivariate model could directly provide summary estimates of（logit）sensitivity, specificity and DOR with corresponding 95% CI for two diagnostic tests（IHA and ELISA）. Also,it could elicit any significant difference that existed a mong sensitivity, specificity and DOR between the two diagnostic methods, and incorporate any correlation that existed between sensitivity;specificity. Conclusion The bivariate model preserves the two dimensional nature of the original data,and separates effects of sensitivity and specificity, which is more rational than a net effect on diagnostic odds ratio scale as in SROC approach. The bivariate model is appropriate and agile,and can be used as an extension and improvement of the traditional SROC method.