Extending Egger's Regression: Detecting Outcome Reporting Bias in Meta-Analysis on Dependent Multiple Outcomes

Outcome reporting bias (ORB) is a form of publication bias resulting from a primary study authors' reporting results for significant outcomes in a meta-analysis (Rothstein et al., 2006). The current study extended Egger's regression test while handling the dependence from multiple effect sizes per study. We compared methods for handling within-study dependence simultaneously (multivariate) versus separately (univariate) and compared the performance of estimation models including RVE, MLMA, and MLMA-RVE. In addition to real data analysis examples, the simulation study explored five characteristics of meta-analytic data in their impact on results. According to the findings, the multivariate model enhances power and the use of RVE or MLMA+RVE with multivariate models worked best. More results, limitations, and directions for future research are discussed.