Statistics for Quantifying Heterogeneity in Univariate and Multivariate Meta-Analyses of Binary Data: The Case of Meta-Analyses for Diagnostic Accuracy

• Activity Number: 194
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Yan Zhou*+ and Nandini Dendukuri
• Companies: McGill University and McGill University
• Keywords: Heterogeneity ; Meta-analysis ; Binomial distribution ; Univariate ; Bivariate ; Dichotomous diagnostic test

Abstract

Heterogeneity is common in diagnostic meta-analyses. So far the unexplained heterogeneity across studies has been quantified by either using the I^2 statistic based on a random effects model for a univariate parameter (i.e. either one of the sensitivity or the specificity) or visually examining the data in receiver-operating-characteristic space. Focusing on diagnostic tests, we derived improved estimates of the I^2 for heterogeneity in a dichotomous outcome. We show that the currently used estimate of the 'typical' within-study variance proposed by Higgins and Thompson is not able to properly account for the variability of the within-study variance across studies for dichotomous variables. Therefore, when the between-study variance is large, the 'typical' within-study variance under-estimates the expected within-study variance and the corresponding I^2 is over-estimated. We propose to use the expected value of the within-study variation in the estimator of I^2. We also derive a bivariate I^2 which is able to account for the correlation between parameters. We illustrate the performance of the new estimates using simulated data as well as two real datasets of diagnostic tests.