多变量随机效应模型在诊断试验Meta分析中的应用与SAS实现  

Multivariate Random Effects Model in Meta-Analysis of Diagnostic Tests and Its SAS Programs

作　　者：刘文华[1] 吴家利[1] 杨扬[1] 宋婷婷[1] 章顺悦[1] 陈远方[1] 尹平[1] 

机构地区：[1]华中科技大学同济医学院公共卫生学院流行病与卫生统计学系,武汉430030

出　　处：《中国循证医学杂志》2012年第2期231-237,共7页Chinese Journal of Evidence-based Medicine

摘　　要：目的介绍多变量随机效应模型(MREM)在多个阈值诊断试验Meta分析中的应用。方法对双变量随机效应模型(BREM)进行扩展和延伸,构建MREM,并通过SAS的PROC NLMIXED过程进行统计实现。结果 MREM可根据贝叶斯估计得到各研究的ROC曲线,且其SROC曲线均匀分布在各研究的ROC曲线之间,而BREM无法获得各研究的ROC曲线。同时,MREM的参数估计不依赖于诊断阈值和构建SROC曲线方法的选择,可以获得唯一的SROC曲线,其AUC介于BREM的5种SROC曲线的AUC之间,可避免高估或低估的情形。结论 MREM能够充分利用各研究的原始数据信息,得到稳定可靠的结果,在多阈值诊断试验Meta分析中有较好的应用价值。Objective To introduce the multivariate random effects model（MREM） in the meta-analysis of diagnostic tests with multiple thresholds.Methods This paper expanded and extended the bivariate random effects model（BREM） to develop the MREM,and implemented it in the SAS Proc NLMIXED procedure.Results The MREM could obtain the study specific ROC curve for each study through empirical Bayes estimation,and the summary ROC curve located in between all study specific ROC curves evenly,while the BREM couldn＇t obtain the study specific ROC curve.In addition,in the aspect of parameters estimation,the MREM didn＇t depend on the choice of the diagnosis threshold and the type of SROC.The MREM could get only one SROC curve and its AUC was between the AUC of the 5 types of SROC from BREM,so it could avoid overestimation or underestimation.Conclusion The MREM can fully exploit the data,obtain stable and reliable results,and have a good application value in meta-analysis of diagnostic tests with multiple thresholds.

关 键 词：多变量随机效应模型 双变量随机效应模型 META分析 多阈值 诊断试验 

分 类 号：R-03[医药卫生]