Empirical Study of Bayesian Methods for Meta-Analyses of Diagnostic Test Data Using a Large Database of Studies in the Medical Literature

• Activity Number: 118
• Type: Invited
• Date/Time: Monday, August 4, 2008 : 10:30 AM to 12:20 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): Christopher H. Schmid*+
• Companies: Tufts Medical Center
• Keywords: Meta-analysis ; Diagnostic tests ; Bayesian ; Meta-regression ; ROC ; Hierarchical model

Abstract

Researchers apply a variety of statistical methods for meta-analysis of data from diagnostic test studies. Typically, each study provides a 2x2 table of true and false positives and negatives from which sensitivity and specificity are derived. Because of the bivariate structure of the data and the typically substantial between-study heterogeneity, Bayesian bivariate models are attractive. It has recently been shown that the bivariate normal model is equivalent to a special case of the hierarchical summary ROC (HSROC) model. We examine the performance of Bayesian forms of these models, contrasting them with simpler models of diagnostic odds ratios and univariate sensitivity and specificity in 265 meta-analyses of diagnostic tests from the medical literature. Special attention focuses on the effect of covariates in meta-regression models in attempting to understand the heterogeneity.