Gray Matter Volume Abnormalities in ADHD: Voxel-Based Meta-Analysis Exploring the Effects of Age and Stimulant Medication

A comprehensive literature search of studies conducting voxel-based morphometry comparisons between patients with ADHD and healthy comparison subjects published between 2001—the date of the first voxel-based morphometry study in ADHD—and May 2011 was conducted using the PubMed, ScienceDirect, Web of Knowledge, and Scopus databases. The search keywords were “attention-deficit hyperactivity disorder,” “ADHD,” or “hyperkinetic,” plus “morphometry,” “voxel-based,” or “voxel-wise.” In addition, we conducted manual searches in several review articles and the reference sections of the identified articles. We excluded studies that used duplicated data sets (i.e., studies that analyzed the same data in different articles), studies that included nonclinical samples, studies that did not compare patients and healthy comparison subjects, studies that examined the neural correlates of ADHD in individuals with chromosomal abnormalities, and studies that used fewer than 10 patients. The corresponding authors were invited by e-mail to provide additional details not included in the original manuscripts. Guidelines from the Meta-Analysis of Observational Studies in Epidemiology group (30) were followed in the study.

Meta-analytical differences in global gray matter volumes were calculated using standard random-effects models with the Globals procedure in the Signed Differential Mapping (SDM) software package (www.sdmproject.com), which uses restricted maximum-likelihood estimation of the variance, a fitting method that has been recommended over previous ones for its good balance between unbiasedness and efficiency (31).

Regional differences in gray matter volume between patients and comparison subjects were also analyzed using SDM, which employs a voxel-based meta-analytic approach that is based on, and improves on, other existing methods (32, 33). The main advantage of SDM is that it uses the reported peak coordinates to recreate maps of the signed (i.e., positive and negative) volume differences between patients and comparison subjects, rather than just assessing the probability or likelihood of a peak. This unique feature makes SDM an optimal method for comparing patients and comparison subjects without biasing the results toward those brain regions that show more interstudy heterogeneity. The SDM methods have been described in detail elsewhere (32) and are briefly summarized here. First, peak coordinates of gray matter differences between patients and comparison subjects are extracted from each data set. Notably, those peaks that do not appear statistically significant at the whole brain level are excluded; that is, while different studies may employ different thresholds, we ensure that the same statistical threshold throughout the brain is used in each study. This is intended to avoid biases toward liberally thresholded brain regions, as it is not uncommon in neuroimaging studies for the statistical threshold for some regions of interest to be rather more liberal than for the rest of the brain. Second, a standard Talairach map of the differences in gray matter is separately recreated for each study by means of a Gaussian kernel, which assigns higher values to the voxels closer to peaks. This procedure includes limiting voxel values to a maximum to avoid biases toward studies reporting various coordinates in close proximity and reconstructing both increases and decreases of gray matter volume in the same map. Full width at half maximum of the kernel is 25 mm, to deal with variations in the size and shape of the original clusters, the position of the reported peak within the original clusters, or the spatial mismatch between the cortical structures of different subjects even after spatial normalization. Third, the mean map is obtained by voxel-wise calculation of the mean of the study maps, weighted by the squared root of the sample size of each study, so that studies with large sample sizes contribute more.

We also conducted additional reliability analyses to assess the robustness of the findings (32): a jackknife sensitivity analysis to assess the reproducibility of the results and descriptive analyses of quartiles. The jackknife sensitivity analysis consists in iteratively repeating the same analysis, excluding one data set at a time to establish whether the results remain significant. The analyses of quartiles describe the proportion of studies reporting changes in a particular region (i.e., 25%, 50%, 75%), regardless of p values (32, 33).

Statistical significance was determined using standard randomization tests, thus creating null distributions from which p values can be directly obtained (32). Finally, we conducted a subgroup analysis of pediatric/adolescent and adult samples, a subgroup analysis of studies conducting the additional voxel-based morphometry modulation step, and a multiple metaregression analysis with age and the percentage of patients receiving stimulant medication in each study as regressors (33).

The search yielded 17 studies. Three studies were discarded because they did not meet study criteria: one used a nonclinical sample (21), one examined the volumetric correlates of ADHD symptoms in individuals with chromosome 22q11.2 deletion syndrome (34), and one did not conduct comparative analyses between ADHD participants and healthy comparison subjects (35). After we requested and received information from the authors of the remaining 14 studies, no methodological ambiguities remained regarding the design or analysis of any of the comparisons. Therefore, 14 high-quality data sets could be included in this meta-analysis, of which five (14–18) consisted of adult ADHD samples and the remaining nine (36–43; K. Rubia et al., unpublished 2010 data) of pediatric/adolescent samples.

Combined, the studies included 378 patients with ADHD and 344 healthy comparison subjects. Patients were 202 children/adolescents (12 hyperactive; 29 inattentive; 74 combined; 87 unknown) and 176 adults (82 with combined type ADHD; 94 with unknown type). As shown in Table 1, no relevant differences between patients and comparison subjects were found in age, gender, or IQ, as the original studies were already well matched in this respect. Further details of each of the included studies, such as comorbid conditions, medication status, or diagnostic criteria, can be found at www.sdmproject.com/database.

Global gray matter volumes were available from seven (six pediatric; one adult) independent data sets (18, 36–38, 40–42). Patients with ADHD (N=191) had significantly smaller global gray matter volumes compared with healthy comparison subjects (N=179) (unbiased d=–0.28, z=–2.65, p=0.008). This difference was more pronounced when the adult study was excluded (unbiased d=–0.33, z=–2.93, p=0.003). No significant heterogeneity was found in either analysis (all data sets: Q=6.08, df=6, p=0.414; only pediatric data sets: Q=4.60, df=5, p=0.466).

Data for this analysis were obtained from all the studies included in the meta-analysis. As shown in Table 2 and Figure 1, ADHD patients had significantly smaller gray matter volumes in a large cluster with its peak in the right lentiform nucleus (Talairach coordinates: x=20, y=0, z=12; SDM value=–0.288, p<0.001; 580 voxels). This large cluster also included the caudate nucleus (230 voxels). ADHD patients also had larger gray matter volumes in the left posterior cingulate cortex/precuneus (x=–18, y=–44, z=26; SDM value=0.079, p<0.001; 35 voxels).

A whole-brain jackknife sensitivity analysis, which systematically excluded one study at a time, showed that the results in the lentiform nucleus were highly replicable, as this finding was preserved throughout all of the 14 combinations of studies. The greater volumes in the left posterior cingulate cortex/precuneus remained significant in all but two combinations of studies (Table 3).

The descriptive analyses of quartiles revealed that the smaller volume in right lentiform nucleus was detected in the median analysis, indicating that at least 50% of the included studies found a significantly smaller volume of gray matter in this region. The left posterior cingulate cortex/precuneus cluster was not detected in this analysis, indicating that less than 25% of studies found significantly larger volumes in this region.

Finally, the subgroup analysis revealed that the above findings remained unchanged when only the nine pediatric studies or only the 11 studies conducting a modulation step were analyzed. Conversely, separate analysis of the five adult studies revealed no significant differences between patients and comparison subjects (Table 3). Thus, these findings were probably driven by the pediatric studies.

Information on both age and medication was available for all 14 data sets, including 378 patients. Among these studies, 207 patients (55%) were receiving medication (methylphenidate, N=182; d-amphetamine, N=1; unidentified stimulants, N=24) at the time of their participation in the studies. The multiple metaregression analysis (33) revealed that both the mean age and the percentage of patients receiving stimulant medication were independent predictors of increasing (i.e., more normal) gray matter volume in the right basal ganglia (Figure 2). That is, age was positively and significantly correlated with gray matter volume in the right putamen (x=24, y=2, z=16; SDM value=0.778, p≤0.001; 402 voxels), controlling for the effects of stimulant medication. Similarly, the percentage of patients on stimulant medication was correlated with gray matter volume in the right caudate (x=16, y=2, z=20; SDM value=0.632, p≤0.001; 63 voxels), controlling for the effects of age.

A sensitivity analysis showed that these results remained significant in all combinations of studies, although the cluster size varied, ranging from 55 voxels (excluding Wang et al. [39]) to 487 voxels (excluding Sasayama et al. [42]) with regard to age and from four voxels (excluding Wang et al. [39]) to 122 voxels (excluding Overmeyer et al. [36]) with regard to stimulant medication.

Our subgroup analysis revealed no significant differences between patients and comparison subjects when only adult studies were included. Consistently, the metaregression analysis revealed that increasing age was correlated with progressively increasing (i.e., more normal) gray matter volume in the right putamen, over and above the effects of medication. This may suggest that over time, ADHD patients may at least partially outgrow their basal ganglia deficits. There is evidence from a longitudinal structural imaging study (22) that relative to healthy children, children with ADHD have parallel growth curves for nearly all brain regions between the ages of 5 and 18 years but a smaller brain volume overall. The only exception was the caudate, where brain abnormalities became negligible by midadolescence, around age 15 (22). The basal ganglia reach their maximum volume around age 10, after which there is a progressive decrease in gray matter volume between childhood and adulthood (55). While there are pronounced gray matter changes between childhood and adolescence in most cortical areas, the gray matter changes between adolescence and adulthood appear to be more restricted to frontal and striatal regions (55), presumably reflecting the late development of fronto-striatal pathways that mediate late-developing higher-level cognitive functions (44). These findings are parallel to those of diffusion tensor imaging studies that show a progressive increase in white matter tract development in striatal brain regions between late childhood and adulthood (56). They are also consonant with consistent evidence from functional imaging studies for progressive development in fronto-striatal brain activation between adolescence and adulthood (57). Given that striatal gray matter progressively decreases with age in normal adolescent development, it is thus possible that ADHD patients eventually “catch up” with their healthy counterparts at some point in late adolescence. A recent cross-sectional diffusion tensor imaging study (58) showed abnormal associations between age and caudate white matter tract development in ADHD patients relative to healthy comparison subjects, which was interpreted as a developmental delay that may normalizein late adolescence. A potential caveat, however, is the fact that more pediatric (N=9) than adult (N=5) studies were included in our metaregression analysis, which may have biased the findings toward greater power for pediatric studies. Nevertheless, the adult studies were generally better powered than the pediatric studies and in total included only 50 fewer patients than the pediatric studies.

The metaregression analysis revealed that the percentage of patients on stimulant medication was correlated with increasing (i.e., more normal) gray matter volume in the right caudate, over and above the effects of increasing age. Stimulant medication is known to block dopamine transporters in the basal ganglia, leading to enhanced extracellular levels of striatal dopamine, while in prefrontal regions it enhances both dopamine and noradrenaline (29). Dopamine is known to modulate the fronto-striatal neural networks of cognition and motivation that are affected in ADHD (44). fMRI studies have consistently shown that acute doses of methylphenidate enhance and normalize activation in the basal ganglia as well as their functional connectivity with fronto-cortical and cerebellar regions (2, 45, 50, 52). It is thus possible that stimulant medication “repairs” abnormal basal ganglia structure in youths with ADHD and thus elicits morphological changes as a consequence of a relative improvement in striatal dopamine levels. The possibility of a “normalization” with stimulant medication of brain abnormalities in ADHD would be consonant with individual structural imaging studies that retrospectively compared subgroups of medicated and nonmedicated children with ADHD matched on symptom severity and found more normal brain structure in medicated compared with unmedicated patients in several ADHD-relevant brain regions, including caudate volume and morphology (26, 27). Other brain regions that were found to be more normal in medicated relative to never-medicated patients were the inferior frontal, premotor, and parietal areas (23), the anterior cingulate cortex (25), the cerebellar vermis (24), and the thalamus (27). An important caveat in these studies, however, is the retrospective analysis of relatively small, even if well-matched, patient subgroups, which is confounded by selection bias (with fewer than 20 participants in the majority of studies). Our metaregression analysis extends this prior evidence by including a relatively large number of ADHD patients (N=378) and hence is thus far the statistically most powerful indication that treatment with stimulants may be associated with more normal striatal brain structure. Nevertheless, this important question of a potential normalization of brain structure with stimulant medication can be adequately addressed only in a prospective longitudinal imaging study within a randomized controlled trial.

The finding of smaller basal ganglia volumes in patients with ADHD is interesting in view of opposite findings in other fronto-striatal disorders, such as obsessive-compulsive disorder (OCD) and Tourette's syndrome, which are characterized by larger gray matter volumes in this region (32, 33, 59). These findings are parallel to results from studies using positron emission tomography showing enhanced dopamine availability in the basal ganglia in OCD and Tourette's syndrome (60, 61) but lower striatal dopamine availability in ADHD (29). Direct fMRI comparisons between ADHD and OCD have furthermore shown lower caudate activation in patients with ADHD but enhanced caudate activation in patients with OCD during salient oddball stimuli, which was also correlated with the respective disorder-relevant symptom severity (49). The inverse volume findings are supportive of the notion of an orbitofrontal-striatal dysregulation in OCD, where dorsal prefrontal regions have poor control over overactive and hyperdopaminergic basal ganglia (62, 63), as opposed to underdeveloped and hypodopaminergic fronto-striatal networks in patients with ADHD (2). Both a dysregulation of fronto-striatal networks (in OCD and Tourette's syndrome) and a developmental dysmaturation of fronto-striatal networks (in ADHD) may thus be associated with poor behavioral inhibition, which is a shared feature of all these disorders.

This study has several limitations, some inherent to all meta-analytic approaches. First, peak-based meta-analyses are based on summarized (i.e., coordinates from published studies) rather than raw statistical brain maps, and this approach may result in less accurate results (64). Second, the different studies included in this meta-analysis used different statistical thresholds. However, while thresholds involving correction for multiple comparisons are usually preferred, the inclusion of studies with more liberal thresholds is still statistically correct. Indeed, SDM preprocessing uses the coordinates of the voxels with the greatest differences to approximately recreate the statistical parametric map, but it does not make assumptions about whether or not these differences were significant. Third, while voxel-wise meta-analytic methods provide excellent control for false positive results, it is more difficult to completely avoid false negative results (64). Fourth, there are some inherent limitations to the voxel-based morphometry method, such as reduced effectiveness in detecting spatially complex and subtle group differences (65). Fifth, some of the included studies reported gray matter density rather than volume. Gray matter density might be understood as a type of volume that has not been corrected by the distorting effects of the normalization to the stereotactic space; therefore, its inclusion in the meta-analysis is valid (it is also a “volume”), although it may add a source of noise. Sixth, we were unable to assess whether ADHD symptom severity was associated with any of the reported structural changes because the included studies used a wide range of measures. Finally, the effects of medication could be assessed only by the percentage of patients receiving stimulant medication, rather than with subtler measures, such as current or lifetime stimulant doses.