Imaging studies conducted across multiple centres, like AIMS, are an effective way to increase recruitment rates, but bring with them their own particular operational and statistical challenges. In particular, the negative impact of the inflation of variance arising from the introduction of a between-centre factor.
Objectives:
Thus, the AIMS consortium undertook a calibration study in which a group of healthy volunteers, matched demographically to AIMS cohort, were scanned at each centre under the study protocol.
Methods:
High resolution structural images from T1-weighted and DESPOT sequences were segmented into their component tissue types with a common processing pipeline. At each intra-cerebral voxel in standard MNI space the partial volume estimates were then regressed onto a random effects model to estimate the within-centre error variance. Subsequently, based on the known sample size, type II error rate and distribution of recruitment across participating centres, power calculations estimated maps of the minimum effect size (MES) that could be observed.
Results:
Strong spatial inhomogeneity in MES was observed with segmentations derived from both sequences and a direct comparison between them informed the design of the main AIMS study.
Conclusions:
Furthermore, the MES maps provides additional information for the interpretation of regions where significant differences were observed between autistic and control participants and, moreover, gives context to the discussion of type II (false negative) errors.