Cerebellar Volume in Autism: Meta-Analysis and Analysis of the Abide Cohort

Background: Many reports suggest that cerebellar dysfunctions may be implicated in autism spectrum disorders (ASD). The cerebellum is a heavily folded region of the rombencephalon, with a number of neurons comparable to that of the neocortex. The cerebellum has been traditionally involved in the performance of precise motor behavior, but there is now evidence for its involvement in cognitive and affective functions as well.

Objectives: Our aim was to objectify a possible alteration of the cerebellar volume in ASD.

Methods:  We looked for a difference of mean or variance between the cerebellar volume distributions of ASD and controls. We first performed a meta-analysis of the literature by looking for studies reporting cerebellar MRI volume measurement on ASD patients and typical controls. We used the PRISMA guidelines for the selection of articles. We combined the effect sizes using the random effects model. We also looked for possible factors of variability across studies by making a meta-regression on age and intelligence quotient (IQ). We then tried to replicate the results of the meta-analysis by analyzing the cerebellar volume segmented with FreeSurfer from the ABIDE cohort. We analyzed the effect of the variables group, site, age, IQ and sex on cerebellar volume by both a linear model approach and a meta-analysis approach.

Results:

Meta-analysis: 16 studies met the criteria for inclusion (635 patients and 424 controls). Although the studies were highly heterogeneous (I2=72%, p<0.0001), we found a larger cerebellar volume in individuals with ASD compared with typical controls (Cohen’s d=0.32, 95% confidence interval: [ 0.08; 0.56 ]; p=0.0089). Individually, the mean statistical power that the studies had to detect such an effect size was only 22.7%. We also found a significative negative impact of age on Cohen’s d (-0.041 year-1, 95% confidence interval: [ -0.065; -0.016 ]), suggesting a faster cerebellar volume decrease in individuals with ASD during adulthood. Age and IQ were not the only factors of heterogeneity between studies, as the residual heterogeneity was still statistically significant (I2=49%, p=0.016) after covarying with these factors. We did not find a statistically significant difference in the variance between individuals with ASD and typical controls (combined log-variance ratio=0.19, 95% confidence interval: [ -0.12; 0.50 ]). ABIDE: After quality control, 341 patients and 360 controls were retained. The mean cerebellar volume did not differ significantly between individuals with ASD and controls. There was a strong impact of the site on the measured volume. We were able to see evidences for effects of age IQ and sex on cerebellar volume, but none of the effects were statistically significantly different between groups.

Conclusions:  The MRI analyses did not confirm the results of the meta-analysis. These discrepancies could be explained by a higher heterogeneity between the studies of the meta-analysis than in the sites of ABIDE, but could also result from methodological biases. Our results call into question the validity of results obtained from small, underpowered cohorts, and underline the utility of data sharing and collaborative work to obtain larger samples.