EcoSta 2019: Start Registration
View Submission - EcoSta2019
Title: Bayesian selection model for publication bias correction in meta-analysis of prognostic studies Authors:  Satoshi Hattori - Osaka University (Japan) [presenting]
Abstract: Publication bias is a serious issue in conduction meta-analyses. For meta-analyses to evaluate intervention effects, the funnel plot and the trim-and-fill method are widely used due to their simplicity. However, they rely on some strong assumptions and may give us misleading insights. The Copas selection model is an alternative as a tool to quantify potential influence of publication bias on the aggregated estimates for the intervention effects. In meta-analyses of prognostic studies, the summary receiver operating characteristics (ROC) curve is widely used. A Copas-type selection model is proposed for the summary ROC curve and an inference procedure under a frequentist setting. Although it enables us to evaluate potential impacts of publication bias on the summary ROC estimation, it has a drawback that an important parameter responsible for publication bias cannot be estimated based on data and then must be treated as a sensitivity parameter. To overcome this difficulty, we propose a Bayesian inference for the Copas-type selection model and demonstrate its usefulness in practice through application to some real datasets.