Weight Indices, Cognition, and Mental Health From Childhood to Early Adolescence

This cohort study evaluates the association between weight indices in childhood and changes in cognition and psychopathology.


Main analyses (standardized estimates).
For transparent data reporting and metaanalytical purposes, full results on standardized estimates of models presented in Figure in main text are available in a linked online data document at OSF https://tinyurl.com/Li-2024-BMICogPsychas eTable 1 (body mass index [BMI]) and eTable 2 (waist circumference [WC]).

Main analyses (unstandardized estimates).
Full results on unstandardized estimates, referenced in main text, are available in eTable 3 (BMI) and eTable 4 (WC) in the linked online data document.These unstandardized models estimated changes in outcomes associated with per unit increase in baseline BMI, WC, cognition, or psychopathology (ie, [age] × [baseline variable] interactions) as well as age-related changes in outcomes (ie, main effect of [age]) at median baseline BMI, WC, or cognition, or zero baseline psychopathology endorsement.The ratio between the two (ie, proportion of change in trajectories) was also calculated.Accompanying Johnson-Neyman plots are shown in eFigures 1 and 2 in the linked online data document.
To further characterize the observed associations between baseline BMI and longitudinal psychopathology, we implemented additional models where participants were classified as having normal weight (age-and sex-adjusted BMI percentiles  5 th to < 85 th ) or overweight/obesity (BMI percentiles  85 th ) at baseline per the Centers for Disease Control and Prevention 2000 growth charts 3 .We assessed associations between this binary weight category variable and longitudinal psychopathology.Full results are reported in eTable 5 in the linked online data document.

Sensitivity analyses (moderation by sex). Main models were modified such that the [age] × [baseline variable] interactions were extended to [age] × [baseline variable] ×
[sex] interactions in order to assess if the longitudinal associations were significantly different between sexes.Full results are available in eTable 6 (BMI) and eTable 7 (WC) in the linked online data document.

Sensitivity analyses (no report of weight-related medication use).
As weightchanging medications could confound the association between weight indices and psychopathology, we reran the main models in subgroups of participants who did not have caregiver-reported weight-related medication use in the two weeks prior to each study visit.Based on Verhaegen et al. 4 , we screened for use of antidepressants ), growth hormones, thyroid hormones (levothyroxine [Levothroid  , Levoxyl  , Synthroid  , Tirosint  , Unithroid  ]), and diabetes medications (insulin, metformin [Fortamet  , Glucophage  , Glumetza  , Riomet  ]).Full results are available in eTable 8 (BMI) and eTable 9 (WC) in the linked online data document.

Sensitivity analyses (no report of common baseline psychiatric diagnoses).
We reran the main models in subgroups of participants who did not have caregiver-reported history of ADHD, depression, bipolar disorder, anxiety, or phobias diagnoses at baseline.Full results are available in eTable 10 (BMI) and eTable 11 (WC) at the website above.

Sensitivity analyses (covarying for psychopathology in baseline cognition models).
To assess if psychopathology confounded the observed associations between baseline cognition and longitudinal BMI and WC, we extended these models by adding each of the 20 baseline psychopathology variables as covariate, running a total of 180 models each for BMI and WC.Full results are available in eTable 12 in the linked online data document

Sensitivity analyses (non-Gaussian models).
To address the possible concern that psychopathology outcomes could follow non-Gaussian distributions, we reran the main models where psychopathology outcomes were fit using Poisson, zero-inflated Poisson, negative binomial, and zero-inflated negative binomial distributions.The best model for each psychopathology outcome was selected based on convergence, Akaike information criterion, Bayesian information criterion, and log-likelihood.Results from these models are consistent with those from the main models and are available in eTable 13 in the linked online data document.
Practice effects in cognitive scores.Repeated cognitive testing may yield experience-driven performance improvement that confounds with changes due to brain and cognitive development.Although the ABCD Study cognitive assessment frequency (every other year) and tasks were designed to minimize practice effects 5 , Anokhin et al. had reported significant practice effects seen at 2-y follow-up relative to baseline 6 .Here, we replicated their analyses by comparing cognitive performance in age-matched pairs where one participant datapoint came from baseline (first assessment) and the other from 2-y follow-up (second assessment).A total of 1550 children were in the age overlap of 10.6 y (127 mo) to 11.1 y (133 mo) across the two timepoints (eMethods Figure , panel A), of which 561 pairs had matching age, sex, race and ethnicity, area deprivation index national percentile, income-to-needs ratio, baseline age-and sexadjusted BMI and WC z-scores, and baseline pubertal development scale total score established using 1-to-1 Mahalanobis distance matching (eMethods Figure , panel B).
We estimated practice effects in cognitive scores by subtracting the baseline data from the 2-y follow-up data in each matched pair, finding very similar values to the estimates reported by Anokhin et al. (eMethods Table 1) 6 .These estimates were potentially generalizable to our entire study sample (n = 5269), because the 2-y follow-up participants (who demonstrated practice effects) had similar characteristics compared to all included participants except in terms of income-to-needs ratio.Fisher's exact test or Wilcoxon rank sum test showed: sex (P > .99),race and ethnicity (P = .32),area deprivation index national percentile (P = .29),income-to-needs ratio (P = .004),baseline BMI z-score (P = .09),baseline WC z-score (P = .57).
eMethods Figure .Caption.Note the age difference shown in panel B among matched pairs was small (0.12 y, or 1.4 mo) even though there was a noticeable absolute standardized mean difference.

Table 2 )
7 ie, children gained the same amount of improvement from repeated assessments irrespective of their baseline weight status, consistent with a prior finding7.Therefore, any association between baseline weight indices and longitudinal changes in cognition would likely be driven by age-related developmental changes and not practice effects.

Table 2 .
Pearson correlations between baseline weight and practice effects