Associations Between Head Motion, Age, and Psychiatric Diagnoses in a Large-Scale Developmental Sample

Many psychiatric disorders emerge during the first 2 decades of life, making it an urgent priority to discover neural mechanisms and biomarkers that might be detected in time to ameliorate or even prevent such disorders. Neuroimaging is a major approach for identifying these mechanisms and markers. In particular, for children, functional magnetic resonance imaging (fMRI) is a safe, noninvasive technique that indirectly measures neural activity via changes in blood oxygenation and flow. Historically, the main use of fMRI was to identify patterns of neural activation in tasks. More recently, it has also been used to map functional connectivity, meaning to identify similarities in fMRI signals over time in different brain regions during passive activities such as movie watching or wakeful rest. A large literature has emerged using all of these fMRI techniques to identify putative mechanisms or markers in many psychiatric disorders with childhood or adolescent onset (1,2). In a new report in Biological Psychiatry: Global Open Science, Shi et al. (3) report the results of a preregistered analysis of a large-scale dataset aimed at understanding how head motion—an important artifact in fMRI studies—may vary with diagnoses of externalizing disorders and internalizing disorders in youth.

A major interpretational challenge in fMRI studies concerns data quality because the fMRI neuroimaging signal is exquisitely sensitive to motion of both the head and the body (e.g., due to breathing, repositioning for comfort, restlessness). The large artifacts added to fMRI signals by motion are spatially structured and introduce systematic biases into the data, biases that can masquerade as neural effects. This fact, when first recognized about 15 years ago, upended the fMRI field: compared with typical young adult populations, there is greater movement in children, older adults, and most clinical populations, and these differences could confound efforts to identify functional connectivity differences as a function of psychiatric diagnosis, especially in young people. Across development, for example, the trajectory of motion is U-shaped, with high motion in young school children decreasing to low values in the late teens to 30s, followed by a gradual rise in later decades (4). After this realization, studies that more carefully controlled for motion found attenuation or elimination of prior developmental and aging effects (5, 6, 7). Clearly, care must be taken when using fMRI to study populations that systematically differ in motion.

In this work, Shi et al. (3) consider whether the quality of pediatric neuroimaging data systematically differs by psychiatric condition. The authors examine head motion during fMRI scans from the HBN (Healthy Brain Network), a large multisite project conducted around New York City. This project invites children and adolescents for psychiatric assessments, fMRI scanning (task, movie watching, and rest), and other biological and social measures, all in the service of identifying mechanisms and markers of psychiatric illness. Each participant undergoes a structured assessment by a clinician and may receive up to 10 DSM-5 diagnoses (the average is 2–3). The data are publicly available and include measures of head motion during scanning, clinician diagnoses, symptom scores from parents, scores on instruments and tests, and other demographic variables (8).

The HBN project releases new data periodically, and Shi et al. considered 2 independent batches of data, one effectively of pre-COVID-19 pandemic releases (n = 971, data up to 2020) and one post-pandemic (n = 437). Both mainly contain children of ages 5 to 16 years. While there are some baseline differences between the cohorts, the 2 groups did not differ in the principal measure of motion: the percentage of time when the scans display moderate to high levels of motion. In both datasets, linear mixed effects models were applied to determine what variables influence the levels of motion observed during fMRI and hence the data quality and reliability of inferences derived from those data. Psychiatric variables were interrogated in several ways: as internalizing/externalizing disorders, as diagnostic categories, and as parent-report symptom scales. Notably, a major strength of this work is that it began as a set of positive findings in the first cohort, which prompted planned, preregistered analyses to examine a second, independent cohort.

In both cohorts and in all analyses, the chief determinant of head motion was age, an effect often several-fold larger than any other significant effect. Interestingly, no other effect was consistent across all cohorts and analyses. The next most consistent effects, seen in both cohorts but only in some analyses, were that increased head motion was associated with lower IQ, higher body mass index (BMI), and being male. In particular, associations with psychiatric diagnoses did not replicate across cohorts, whether viewed in transdiagnostic terms of internalizing/externalizing disorders or as diagnostic categories (e.g., attention-deficit/hyperactivity disorder [ADHD], autism spectrum disorder [ASD]). However, 2 exploratory analyses revealed interesting results that did replicate in both cohorts: an association between motion and parental report of attentional problems and an interaction between age and diagnosis such that children with neurodevelopmental disorders (e.g., ASD) did not display decreasing motion with age.

What does this analysis teach us? First, age is the major determinant of head motion, swamping all other examined variables. This is a critical finding that has been reported in numerous studies, and Shi et al.’s analysis of this large-scale dataset provides strong support to reinforce this conclusion. Second, the most explanatory variables, in approximate order, were age, BMI, IQ, and sex. Third, there were no consistent associations with any psychiatric diagnosis or transdiagnostic category that had even trend-level significance in both cohorts.

The positive findings are noteworthy because they show that the study is powered adequately to detect well-known, expected effects: the decline of motion over childhood, lower motion in females, and lower motion with higher IQ (3,9,10). The exploratory finding that, cross-sectionally, children with neurodevelopmental disorders did not display decreasing motion with age is provocative and warrants further study. The finding that higher BMI is associated with lower motion in the HBN data is surprising, as several large studies with both adults and children have reported a positive correlation between BMI and head motion-positive (9,10). This finding may be related to the unusual composition of the HBN sample: approximately two-thirds of both cohorts were diagnosed with ADHD, and the analyses did not account for stimulant use.

In the context of the positive findings, the negative findings gain strength because they indicate that systematic relationships between head motion and major diagnostic categories in child psychiatry—if they exist—are likely quite small compared with other established effects. Two limitations of this welcome news deserve mention. First, this study examined one of several possible ways to estimate motion, i.e., the percentage of scans affected. Other estimates, such as average motion, might yield different results. Second, whether motion artifact—in the scans, and the ultimate cause of biased inferences—differs across populations, is a topic beyond the scope of this work.