B1879
Title: Alternative tests and measures for between-study inconsistency in meta-analysis
Authors: Lifeng Lin - University of Arizona (United States) [presenting]
Abstract: Meta-analysis is a widely used tool to combine research findings from multiple studies in many disciplines. A critical issue in meta-analysis practice is addressing the inconsistencies between different studies' results. Such inconsistency could arise due to several factors, such as heterogeneity in baseline characteristics of individual studies' populations, different methods applied by research teams, and outlying effects of a few studies. In the current literature, the Q and $I^2$ statistics are commonly used to test for and quantify the between-study heterogeneity, respectively, but they are not powerful in many cases, particularly when the number of studies is small. Alternative Q-like statistics are proposed for assessing inconsistency. Formal statistical tests are built on these statistics, including a hybrid test. The corresponding measures for inconsistency are also studied. The various tests are compared using comprehensive simulation studies with many scenarios of between-study distributions. The hybrid test is found to have relatively high power across various settings. Case studies are given to illustrate the proposed methods' real-world performance.
