Assessing publication bias in coordinate-based meta-analysis techniques?

fMRI is an important neuroimaging technique to localize brain functions. Since publications of single fMRI studies have flourished, it is increasingly recognized that progress in understanding human brain function requires the integration of data across studies. Meta-analysis is a promising tool to achieve this. Furthermore, meta-analysis can address some of the shortcomings of fMRI studies such as lack of power and the large multiple testing problem. Recently coordinate-based methods have been specifically developed for fMRI data that combine the limited amount of voxels that survive a statistical threshold or peaks into a summary map. The most frequently employed meta-analysis toolbox for fMRI data is Activation Likelihood Estimation(ALE) (Eickhoff et al., 2009). ALE verifies whether the congruency of location of activation across studies is larger than can be expected by chance. However, no procedures exist to assess publication bias for coordinate-based meta-analyses of fMRI studies. To validate the ALE toolbox, we study how many studies without activation in a location of interest can be added to the meta-analysis before the congruency of activation is no longer statistically significant in this location. On the one hand, we want this number to be sufficiently large as this reflects the robustness of the effect against publication bias. On the other hand, a large number might indicate that a small amount of studies drives the analysis. We simulated data-sets, consisting of studies with activation and up to 100 null studies with no activation at a predefined location in the brain. The number of null-studies that can be added is evaluated in function of different multiple testing corrections and sample sizes.

http://ift.tt/2pBEymN