Kavli Affiliate: Wesley Thompson
| Authors: Tzu-Hsuan Huang, Chun Chieh Fan, Wesley Thompson and Robert Loughnan
| Summary:
Abstract Magnetic resonance imaging (MRI) studies of the human brain are now attaining larger sample sizes with more diverse samples. However, population stratification, a key factor driving heterogeneity and confounding of associations, is seldom accounted for in neuroimaging analyses. To investigate this issue, we assessed the impact of population stratification on multimodal imaging measures using baseline data from the Adolescent Brain Cognitive Development (ABCD) StudySM (n = 10,748). Given this sociodemographically diverse sample, which broadly reflects the population composition of the United States, we performed a thorough evaluation of the impact of population stratification on derived neuroimaging metrics across five different imaging modalities: task functional MRI (task fMRI), resting state functional MRI (rsMRI), diffusion tensor images (DTI), restricted spectrum images (RSI), and structural T1 MRI (sMRI). We used parental income level as an example to highlight the impact of population stratification in confounding brain-wide associations. We show that derived metrics from structural images have up to three times more signal related to population stratification than do functional images. Controlling for population stratification in statistical models leads to a substantial reduction in the association strength between variables of interests and imaging measures, indicating the scale of potential bias. Moreover, because of unequal access to resources (such as income) across ancestral groups in United States, population stratification effects on imaging features may bias associations between parental income levels and imaging features, as we demonstrate. Our results provide a guide for researchers to critically examine the impact of population stratification and to assist in avoiding spurious brain-behavior associations. Highlights Here, we conduct a comprehensive survey of the confounding impact of population stratification in large-scale imaging studies. Morphological features from structural imaging appear to be more susceptible to the confounding effects of population stratification than do functional imaging features. The population stratification tends to inflates the association strengths between the variable of interest and imaging features. When the variable of interest is highly colinear with the population stratification, such as income levels, brain associations cannot be differentiated and may be misattributed as mediating effects. It is critical to account for population stratification in imaging analyses. Competing Interest Statement The authors have declared no competing interest.