Two automated, coordinate-based meta-analyses of voxel-based morphometry (VBM) studies comparing individuals with obsessive–compulsive disorder (OCD) and healthy controls have been recently published, respectively, in the British Journal of Psychiatry 1 and Neuropsychopharmacology. 2 Surprisingly, their results are less concordant than one would have expected. We believe this is largely due to methodological differences across the studies.
In coordinate-based meta-analysis, three-dimensional brain maps are built based on the reported coordinates of voxels of peak statistical difference between groups, with higher values being assigned to voxels closer to these coordinates. The full-width at half maximum (FWHM) value of a Gaussian kernel determines the width of spatial distribution, 1,3,4 thus critically influencing the results. Radua & Mataix-Cols 1 used a 25 mm FWHM kernel, whereas Rotge et al 2 set this parameter at 12 mm. Such distinction may explain two differences between their results. First, only Radua & Mataix-Cols reported grey matter increases in the right superior parietal cortex and precuneus, although both studies took exactly the same parietal cortical coordinates (n = 4) from the individual VBM investigations. However, these parietal coordinates were not in close proximity to each other, possibly reflecting the spatial uncertainty of OCD-related abnormalities in this area. Since the width of the distribution of voxel values reflects the spatial uncertainty of significant findings, 3 the greater FWHM kernel used by Radua & Mataix-Cols possibly afforded greater sensitivity to detect parietal clusters of grey matter difference. Second, although both studies detected striatal foci of increased grey matter, Rotge et al's findings were confined to the putamen, whereas in the study by Radua & Mataix-Cols these foci spread also to the globus pallidus and caudate nucleus. The greater FWHM value used by Radua & Mataix-Cols probably explains the lower spatial resolution of the striatal foci in their meta-analysis.
Moreover, Rotge et al used the activation likelihood estimation method, 4 in which coordinates regarding increased and decreased grey matter are separately computed in independent maps. Conversely, Radua & Mataix-Cols used the signed differential mapping method, 1 in which coordinates for findings of either increased or decreased grey matter are reconstructed in the same map, thus influencing each other. Since VBM studies of OCD have identified foci of both increased and decreased grey matter in the orbitofrontal cortex, this may explain why Radua & Mataix-Cols did not reproduce Rotge et al's finding of grey matter increase in this region of critical relevance to the pathophysiology of OCD. 5
An additional source of discrepancy relates to the criteria for coordinate selection. Rotge et al included all coordinates reported in the selected studies, regardless of statistical thresholds and correction for multiple comparisons. Conversely, Radua & Mataix-Cols employed stricter criteria, thus leading to the inclusion of fewer coordinates (as detailed in their article). 1
In conclusion, these papers are an example of how methodological differences may critically influence the results of coordinate-based meta-analyses. Therefore, when performing such investigations, one should clearly justify the criteria used for coordinate selection and the choice of other methodological parameters. Future studies using such novel techniques should focus on how to foster greater methodological comparability and reproducibility of results.
Edited by Kiriakos Xenitidis and Colin Campbell