Abstract:Background Previous research has implicated demographic, psychological, behavioral, and cognitive variables in the onset and maintenance of pediatric overweight/obesity. No adequately-powered study has simultaneously modeled these variables to assess their relative associations with body mass index (BMI; kg/m 2 ) in a nationally representative sample of youth. Methods Multiple machine learning regression approaches were employed to estimate the relative importance of 43 demographic, psychological, behavioral, and cognitive variables previously associated with BMI in youth to elucidate the associations of both fixed (e.g. demographics) and potentially modifiable (e.g. psychological/behavioral) variables with BMI in a diverse representative sample of youth. The primary analyses consisted of 9–10 year olds divided into a training ( n = 2724) and test ( n = 1123) sets. Secondary analyses were conducted by sex, ethnicity, and race. Results The full sample model captured 12% of the variance in both the training and test sets, suggesting good generalizability. Stimulant medications and demographic factors were most strongly associated with BMI. Lower attention problems and matrix reasoning (i.e. nonverbal abstract problem solving and inductive reasoning) and higher social problems and screen time were robust positive correlates in the primary analyses and in analyses separated by sex. Conclusions Beyond demographics and stimulant use, this study highlights abstract reasoning as an important cognitive variable and reaffirms social problems and screen time as significant correlates of BMI and as modifiable therapeutic targets. Prospective data are needed to understand the predictive power of these variables for BMI gain.

Demographic, psychological, behavioral, and cognitive correlates of BMI in youth: Findings from the Adolescent Brain Cognitive Development (ABCD) study