Title : Head motion and fMRI data quality in youth with eating disorders: differential contributions of BMI, age, and impulsivity
Abstract:
Motivation: Participant head motion during fMRI contributes to image censoring and poses concerns with data quality and integrity. Individual characteristics such as age and BMI have consistently been shown as correlates of head motion. Impulsivity may also contribute to differences in head motion. We aimed to evaluate data censoring due to head motion in youth with binge/purge type eating disorders (ED), a population characterized by reward system differences and tendency towards impulsivity.
Methods: Using data from the Naltrexone Neuroimaging Randomized Controlled Trial, we tested three binomial regression models on their ability to account for data censoring due to head motion (i.e., censor fraction, or proportion of outlier TRs). Our base model included characteristic variables of age, sex, and BMI z-score. Model 2 included all variables from the base model and added measures of trait impulsivity (e.g., BIS-Brief and UPPS-P scores). Model 3 included all variables from model 2 and added diagnostic variables (e.g., ED subtype and ADHD history). An ANODEV nested model comparison was used to evaluate model fit.
Results: Across all models, BMI was a strong, consistent, and positive predictor of censoring (β ≈ 0.72–0.80, p < 0.001). Age did not significantly predict censor fraction (β ≈ 0.002–0.034, p > 0.1), nor did ED subtype. Impulsivity measures showed differing effects, with BIS-Brief positively associated with censor fraction and UPPS-P negatively associated. Surprisingly, ADHD showed a moderate inverse effect, as those that had an ADHD diagnosis showed decreased censoring (β ≈ -0.99, p <0.05). Compared to the base model, the addition of impulsivity measures did not significantly improve model fit (p = 0.16), and the addition of diagnostic variables only marginally improved fit (p = 0.07).
Conclusion: Our data show consistency with findings that BMI is a primary predictor of head motion during fMRI. Age was not significantly related to censor fraction, likely due to limited age range in our sample. Effects of impulsivity became most evident in the final model, suggesting shared variance with diagnostic variables; however, these effects were relatively minimal overall. The negative association between ADHD diagnosis and censor fraction was unexpected and suggests nuances that may warrant further investigation. Overall, participant characteristics alone accounted for greater variance in censor fraction than impulsivity or diagnostic factors.
