Attention-deficit hyperactivity disorder symptoms and brain morphology: Examining confounding bias
Curation statements for this article:-
Curated by eLife
Evaluation Summary:
This study will be of interest to the large class of researchers who perform brain-behavior correlation analysis in the neuroimaging field, especially those related to neurodevelopment. The authors found that controlling for socioeconomic and maternal behavioral confounders, in addition to the usual demographic variables, generally attenuated such associations in ADHD using two independent large cohorts. The findings highlighted the importance of careful confounder selection and control for robust brain-behavior associations.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 agreed to share their name with the authors.)
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Medicine (eLife)
Abstract
Associations between attention-deficit/hyperactivity disorder (ADHD) and brain morphology have been reported, although with several inconsistencies. These may partly stem from confounding bias, which could distort associations and limit generalizability. We examined how associations between brain morphology and ADHD symptoms change with adjustments for potential confounders typically overlooked in the literature (aim 1), and for the intelligence quotient (IQ) and head motion, which are generally corrected for but play ambiguous roles (aim 2).
Methods:
Participants were 10-year-old children from the Adolescent Brain Cognitive Development ( N = 7722) and Generation R ( N = 2531) Studies. Cortical area, volume, and thickness were measured with MRI and ADHD symptoms with the Child Behavior Checklist. Surface-based cross-sectional analyses were run.
Results:
ADHD symptoms related to widespread cortical regions when solely adjusting for demographic factors. Additional adjustments for socioeconomic and maternal behavioral confounders (aim 1) generally attenuated associations, as cluster sizes halved and effect sizes substantially reduced. Cluster sizes further changed when including IQ and head motion (aim 2), however, we argue that adjustments might have introduced bias.
Conclusions:
Careful confounder selection and control can help identify more robust and specific regions of associations for ADHD symptoms, across two cohorts. We provided guidance to minimizing confounding bias in psychiatric neuroimaging.
Funding:
Authors are supported by an NWO-VICI grant (NWO-ZonMW: 016.VICI.170.200 to HT) for HT, LDA, SL, and the Sophia Foundation S18-20, and Erasmus University and Erasmus MC Fellowship for RLM.
Article activity feed
-
-
Evaluation Summary:
This study will be of interest to the large class of researchers who perform brain-behavior correlation analysis in the neuroimaging field, especially those related to neurodevelopment. The authors found that controlling for socioeconomic and maternal behavioral confounders, in addition to the usual demographic variables, generally attenuated such associations in ADHD using two independent large cohorts. The findings highlighted the importance of careful confounder selection and control for robust brain-behavior associations.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 agreed to share their name with the authors.)
-
Reviewer #1 (Public Review):
The authors applied FreeSurfer and performed vertex-wise linear regression models for ADHD with cortical surface and volume. They note they did not test cortical thickness, as it was found to yield null findings in a prior analysis of the Generation R data.
In initial analyses without confounders (model 1), they observed widespread associations between ADHD symptoms (measured with the Child Behavior Checklist) with surface area and volume in both datasets. In model 2, they adjusted for SES, household income, maternal education, and maternal age at childbirth. A third model added prenatal exposure to substance use (tobacco and cannabis) and postnatal maternal psychopathology. Model 2 reduced surface cluster sizes by 20% in ABCD and 49% in GenR; volume was reduced by 42% and 67%, respectively.
Model 3 reduced …
Reviewer #1 (Public Review):
The authors applied FreeSurfer and performed vertex-wise linear regression models for ADHD with cortical surface and volume. They note they did not test cortical thickness, as it was found to yield null findings in a prior analysis of the Generation R data.
In initial analyses without confounders (model 1), they observed widespread associations between ADHD symptoms (measured with the Child Behavior Checklist) with surface area and volume in both datasets. In model 2, they adjusted for SES, household income, maternal education, and maternal age at childbirth. A third model added prenatal exposure to substance use (tobacco and cannabis) and postnatal maternal psychopathology. Model 2 reduced surface cluster sizes by 20% in ABCD and 49% in GenR; volume was reduced by 42% and 67%, respectively.
Model 3 reduced surface clusters by 23% and 5% for ABCD and GerR, respectively, compared to model 2. Volume was reduced by 32% in ABCD, and increased by 7% in GenR. Similar results were obtained when ADHD was treated categorically as present or absent. A fourth model examined the complex effects of adjustment for IQ. Such adjustment further reduced the spatial extent of clusters (vs. model 3) by 23% and 57% for the area in ABCD and GenR, respectively, and 37% and 47% for volume.
This manuscript has many strengths. It is thoughtful and addresses a major challenge in the field, using large samples and rigorous methods. It is well written and mostly quite clear.
-
Reviewer #2 (Public Review):
In this study, the authors examined the associations between brain morphology and ADHD symptoms and how the adjustments for confounders change these associations. While the socioeconomic and maternal behavioral confounders were overlooked in most of prior ADHD neuroimaging studies, the authors show that controlling for these confounders in addition to the demographic variables generally attenuated the associations. The authors proposed using and building Directed Acyclic Graphs (DAGs) to identify confounders, colliders, and mediators, which helps present the research questions clearly. Notably, the authors also examined the potential role of IQ in the brain morphology-ADHD associations and concluded that IQ may be a confounder, mediator, or collider, thus the adjustments for IQ are often unnecessary.
Importan…
Reviewer #2 (Public Review):
In this study, the authors examined the associations between brain morphology and ADHD symptoms and how the adjustments for confounders change these associations. While the socioeconomic and maternal behavioral confounders were overlooked in most of prior ADHD neuroimaging studies, the authors show that controlling for these confounders in addition to the demographic variables generally attenuated the associations. The authors proposed using and building Directed Acyclic Graphs (DAGs) to identify confounders, colliders, and mediators, which helps present the research questions clearly. Notably, the authors also examined the potential role of IQ in the brain morphology-ADHD associations and concluded that IQ may be a confounder, mediator, or collider, thus the adjustments for IQ are often unnecessary.
Importantly, the authors used two independent large datasets to replicate the findings which strengthen the confidence in the results (although the significant clusters are slightly different between the two datasets). Large cohort studies could have a large number of confounds, including scanner acquisition protocol processing parameters, head motion confounds, and so on. Although this study only focused on limited confounders related to ADHD, the suggestions from authors for minimizing confounding bias are useful for future studies.
-
