Variation in executive function relates to BMI increases in youth who were initially of a healthy weight in the ABCD Study

Study design

The ABCD Study is a 21-site 10-year cohort study that enrolled 11,878 youth aged 9/10 years old at baseline (August 2016 to October 2018). Recruitment was tailored to match the demographic population of the US Census. Details pertaining to study design, assessments, objectives, and protocols have been published in numerous documents and on the study website (www.ABCDStudy.org). Assessments are conducted annually, but the protocol is varied. Broadly, the ABCD Study was designed to assess cognitive and health development throughout adolescence. The current article focused on assessments from baseline (9/10 years old) and the 2-year follow-up (11/12 years old), and analyses were restricted to a subset of youth who initially had a healthy weight with useable data at each time point (n = 2794, 12.4% had overweight/obesity by ages 11/12 years, 51.2% male, 72.8% White). Details pertaining to overall ABCD Study inclusion as well as exclusion criteria that were applied to obtain a sample optimal for our analyses are available in online Supporting Information. Assessments included in this article were limited to anthropometrics, demographics, and task- and questionnaire-based assessments of EF and cognition. A centralized Institutional Review Board was approved by the University of California, San Diego.

Physical health assessments

Participants were weighed to the nearest 0.1 in at each visit by a trained researcher in light clothing using a physician's scale (Detecto model 439, Webb City, Missouri). Height was assessed on the same scale, with the youth's heels against the height rod. Two measurements were acquired, and a third was taken if measurements varied by more than one-fourth of an inch or 0.1 lb. Height and weight were converted to BMI (kilograms/meters squared) and BMI percentiles according to the Centers for Disease Control and Prevention (CDC) sex-age-height-weight specific growth charts [(22)] for clinical interpretations. Unadjusted (raw) BMI values were used in the statistical analyses to evaluate within-subject change. The pubertal development scale [(23)] was administered yearly to caregivers and youth. Details regarding physical health assessments have been published elsewhere [(24)] and in online Supporting Information.

Demographics

Caregivers reported on the youth's race and ethnicity, sex at birth, date of birth, family structure, and socioeconomic status (e.g., education) of the family (see online Supporting Information for details).

EF and cognitive assessment

Biennial assessments of EF and cognition were conducted at baseline and the 2-year follow-up. EF trait-based questionnaires that measured impulsivity and (the lack of) inhibition and that measured reward responsiveness consisted of the Behavioral Inhibition System/Behavioral Approach System (BIS/BAS) and of the Urgency, Premeditation (lack of), Perseverance (lack of), Sensation Seeking, Positive Urgency, Impulsive Behavioral Scale (UPPS-P [i.e., impulsivity assessment]). EF and cognitive tasks included assessments of inhibitory control, proxies of working memory, processing speed, verbal and visual learning and memory, and vocabulary; they were assessed with the National Institutes of Health (NIH) Toolbox, the Rey Auditory Verbal Learning Test (RAVLT), and the Little Man Task. Although not all measures fit the common definition of EF, these were close approximations, so we will use the term EF to refer to both classic EF and other cognitive processes that depend on optimal EF. Details of these assessments are available elsewhere [(24)] and in online Supporting Information. 

Linear mixed-effects modeling

General results

The number of participants included for each analysis differed across tasks (see online Supporting Information for the N's included per task; Table 1 displays the demographics). At baseline, all youth had a healthy weight, but by the 2-year follow-up, 323 youth transitioned to have overweight (11.6%), and 23 transitioned to have obesity (0.8%). Additionally, by the 2-year follow-up, 151 youth had extreme weight gain (as defined in [(27)]), whereas 38 youth had extreme weight gain but were still classified as having a healthy weight at the 2-year follow-up. Notably, the demographics of the sample included in the analyses did differ from the entire sample with regard to race, ethnicity, and education. Figure 1 displays the distribution of BMI and weight status for each year of data collection.

Do underlying differences in EF relate to greater increases in BMI? A test of the dual processes model of overeating

UPPS-P

Significant EF × Time interaction effects on BMI were observed for the positive urgency subscale, in which greater positive urgency scores were associated with greater increases in the rate of change in BMI over 2 years (β = 0.12, p < 0.001, Figure 2A). No other associations were observed, and all results are reported in Table 2.

TABLE 2. Change in executive function (EF) by time on BMI from baseline to 2-year follow-up

Interaction	β	95% CI	df	t statistic	p	Significance
UPPS-P
Negative urgency × Time	0.05	[−0.02 to 0.12]	3192.98	1.36	0.17
Lack of planning × Time	0	[−0.06 to 0.07]	3096.1	0.06	0.95
Sensation seeking × Time	0	[−0.06 to 0.07]	3007.36	0.13	0.89
Positive urgency × Time	0.12	[0.06 to 0.19]	3144.44	3.63	<0.001	***a
Lack of perseverance × Time	0.04	[−0.02 to 0.1]	3052.97	1.19	0.23
BIS/BAS
BIS × Time	0.08	[0.02 to 0.15]	3146.54	2.42	0.02	*a
Drive × Time	0.08	[0.01 to 0.15]	3155.46	2.38	0.02	*a
Reward response × Time	0.07	[0.0 to 0.14]	3233.14	1.96	0.05	*
Fun seeking × Time	0.03	[−0.04 to 0.1]	3212.09	0.84	0.4
Little Man Task
Efficiency × Time	0.09	[0.02 to 0.164]	3574.21	2.51	0.01	*a
NIH Toolbox
Flanker Inhibitory Control × Time	−0.08	[−0.15 to −0.0]	2981.9	−2.1	0.04	*a
Picture Vocabulary × Time	−0.08	[−0.14 to −0.02]	2687.8	−2.4	0.02	*a
Pattern Comparison × Time	−0.07	[−0.15 to 0.0]	2875.43	−1.94	0.05
Reading Comprehension × Time	−0.08	[−0.14 to −0.01]	2645.87	−2.35	0.02	*a
Picture Sequence Memory × Time	−0.1	[−0.16 to −0.03]	2928.79	−2.7	<0.01	**a
RAVLT
Learning × Time	−0.125	[−0.195 to −0.054]	2286.903	−3.465	0.001	***a
List B correct × Time	−0.148	[−0.223 to −0.074]	2475.644	−3.898	<0.001	***a
Immediate Delay × Time	−0.143	[−0.215 to −0.072]	2325.556	−3.93	<0.001	***a
Long Delay × Time	−0.146	[−0.217 to −0.075]	2303.462	−4.016	<0.001	***a
Repetition List B × Time	0.015	[−0.066 to 0.096]	2781.802	0.356	0.722
Repetition Immediate Delay × Time	0.024	[−0.056 to 0.103]	2707.237	0.582	0.561
Repetition Long Delay × Time	−0.039	[−0.121 to 0.044]	2694.271	−0.912	0.362
Total Repetitions × Time	−0.009	[−0.085 to 0.067]	2490.833	−0.232	0.816
Intrusions List B × Time	0.022	[−0.06 to 0.104]	2826.468	0.52	0.603
Intrusions Immediate Delay × Time	−0.059	[−0.141 to 0.022]	2837.695	−1.423	0.155
Intrusions Long Delay × Time	−0.005	[−0.086 to 0.076]	2818.678	−0.13	0.896

• Note: All effects that were significant survived Benjamini-Hochberg correction. Flanker Inhibitory Control and Attention Task was abbreviated in the table to Flanker Inhibitory Control; Pattern Comparison Processing Speed was abbreviated to Pattern Comparison in the table.
• Abbreviations: BIS/BAS, Behavioral Inhibition System/Behavioral Approach System; df, degrees of freedom; RAVLT, Rey Auditory Verbal Learning Test; UPPS-P, Urgency, Premeditation (lack of), Perseverance (lack of), Sensation Seeking, Positive Urgency, Impulsive Behaviorial Scale.
• ^a Survived correction.
• *p < 0.05, **p < 0.01, ***p < 0.001.

NIH Toolbox

Significant EF × Time interaction effects on BMI were observed for four NIH Toolbox assessments, namely Flanker Inhibitory Control and Attention, Picture Vocabulary, Pattern Comparison Processing Speed, Reading Comprehension, and Picture Sequence Memory. Lower scores on the Flanker Inhibitory Control and Attention (β = −0.08, p = 0.036), Picture Vocabulary (i.e., verbal IQ, β = −0.08, p = 0.017), Reading Comprehension (e.g., language, oral reading, β = −0.08, p = 0.019), and Picture Sequence Memory (e.g., episodic memory, β = −0.1, p = 0.007) were associated with greater increases in the rate of change in BMI over 2 years (Figure 3A–D). All results are reported in Table 2.

Little Man Task

There was a significant EF × Time interaction effect on BMI for efficiency performance on the Little Man Task. Lower efficiency scores (e.g., visuospatial processing, β = −0.13, p = 0.01) were associated with greater increases in the rate of change in BMI over 2 years (Figure 4A). All results are reported in Table 2.

Does weight gain lead to variation in EF?

There was a significant BMI × Time interaction effect on the Flanker Inhibitory Control and Attention task (β = −0.60, p = 0.01). No other associations were observed (Table 3).

Differences in EF and their relation to increased weight status: evidence for the dual processes model of overeating

The dual processes model of overeating suggests that underlying differences in EF affect the ability to resist temptation and stick to longer-term plans (e.g., diet adherence, preventing overeating) [(9)]. Essentially, within this framework, the desire to consume a rewarding food is stronger than the ability to stop behavior and adhere to a healthy diet. In our study, we tested this theory by examining how variation in EF and cognition related to the rate of change in BMI over a 2-year period of pre- and early adolescence. Of particular interest was examining the relationship between impulsivity and BMI, as higher scores on impulsivity assessments have been associated with weight gain and food intake across several studies in children [(14, 15, 17, 18)]. Here, impulsivity was indirectly measured with standard and validated trait-based questionnaires (e.g., BIS/BAS, UPPS-P) and tasks (e.g., Flanker Inhibitory Control and Attention–NIH Toolbox). Our results showed that youth who scored higher on trait-based impulsivity measures (e.g., positive urgency, reward responsiveness, drive, and BIS) and lower on inhibitory control assessments (e.g., Flanker Inhibitory Control and Attention) had greater increases in the rate of change in BMI from age 9 years to age 12 years. This suggests that underlying differences in impulsivity may be facilitating increased weight gain. Although the ABCD Study did not assess food intake, we interpret increases in BMI (i.e., weight gain) to be a proxy for overeating, as an energy balance model suggests that weight gain occurs when there is an excess of calories [(29)]. However, future studies are needed to assess the direct mechanism between food intake, unhealthy weight gain, and changes to EF over time. We also acknowledge that there are other factors that may explain these relationships, as environmental factors like socioeconomic status have been associated with both obesity and EF in children [(30-32)]. Notably, our models controlled for caregiver education (a proxy for socioeconomic status), and, as such, we interpret the relationship between impulsivity and weight gain to be independent of socioeconomic status. However, future studies are needed to confirm this relationship. Importantly, our results corroborate the literature showing an association between increased BMI and sensitivity to rewards and punishment [(18)], which are facets of impulsivity.

Although much literature has focused on impulsivity (e.g., reward, inhibitory control) as it relates to food intake and unhealthy weight gain, optimal functioning of other cognitive functions is also integral to making adequate food-based decisions [(33)]. Several studies have noted correlations between these processes and obesity in animals [(34, 35)] and humans [(33, 36, 37)]. Cognitive abilities such as memory interact with basic EF domains to create adaptable responses [(38)]. For example, memories about previously learned associations help to inform consumption based on previous beliefs (such as pleasant or unpleasant experiences) or to inform diet adherence [(33)]. However, memory can help to prevent overeating by providing signals related to satiety (e.g., helping us to remember when we ate last and how full we are) [(39)]. Thus, variation in memory may contribute to issues with appetite control and cause overeating despite not being hungry. As such, this may be one explanation for why steeper increases in the rate of change in BMI over 2 years were related to lower performance on memory and learning paradigms. For example, youth with variation in memory processing may provide inaccurate information about hunger and satiety, which could lead to overeating. Hence, together with the tendency to act impulsively, underlying differences in memory and learning may contribute to unhealthy weight gain. Although our results do confer with cross-sectional studies [(36, 37, 40-42)], future studies are needed to understand the exact physiologic mechanisms or neurocognitive domains driving the associations between cognitive function and weight gain.

Weight gain and its association to changes in EF

BMI was related to the rate of change in performance on the Flanker Inhibitory Control and Attention task; as BMI increased from ages 9/10 to ages 11/12, performance decreased, providing some evidence that weight gain may cause changes to EF [(20)]. However, BMI was not related to changes in any other assessments of EF or cognition. This was somewhat surprising because animal research has shown evidence that overeating and weight gain are related to cognitive deficits via neuroinflammation during adolescence [(34, 35)]. In children, longitudinal studies have suggested that increases in BMI z scores were related to microstructural changes in the nucleus accumbens (a region involved in reward and motivation) in a circular way, such that alterations in this structure were associated with additional weight gain as well as greater intake of high-fat food [(43)]. On the other hand, 2 years of extreme weight gain was related to structural changes in brain regions associated with EF but not trait-based impulsivity [(27)]. Furthermore, in a subsample of healthy-weight youth at baseline, BMI at 9/10 years old was not related to brain structure 2 years later [(44)]. Therefore, it may be that 2 years of human adolescence may not be enough time (or not enough weight was gained among these initially healthy-weight youth) to observe consequences of weight gain on EF (as assessed by questionnaires and tasks). From a prevention and intervention perspective, this lack of findings is encouraging, suggesting that 2 years of weight gain may not lead to long-lasting effects on EF. However, adolescence is also a period of normative development in EF maturation, and it is unknown how weight gain may affect future maturation. Additionally, it is unknown how much weight gain is needed to show effects on EF. As such, future research is needed to understand (A) these associations over a longer period of time; (B) these associations among youth with greater variation in unhealthy weight gain; and (C) whether these associations are due to neuroinflammation or other mechanisms (e.g., vascular, hormonal, metabolic changes).

Implications

EF training has been shown to increase adherence to treatment programs [(28)], but results are largely short-lived as only a few studies showed 10-year effects; while EF training for weight loss is promising, many behavioral programs do not obtain long-term success. One explanation for this could be that we lack a refined understanding of the exact mechanisms of interaction between EF and weight gain and the impact of pubertal development. Therefore, our results do have relevance for treatment programs that may wish to target youth in earlier stages of weight gain, prior to large variability in puberty (although our analyses did control for pubertal effects). Long-term treatment studies are still needed to examine the effects of behavioral treatment programs. Within this vein, optimal functioning of EF can positively contribute to other health-fostering behaviors that go beyond dietary choices; therefore, training may yield crossed benefits with respect to several other quality-of-life determinants (e.g., physical activity, good sleep habits, greater social interactions, academic success) [(13, 45)]. Lastly, since children with obesity are more likely to become adults with obesity [(4)], early interventions may be necessary to prevent the cascade of adverse events often attributed to having obesity. EF training is often not incorporated into daily activities and it is short-term and time-consuming, but our findings also may have benefit forbehavioral lifestyle interventions. Therefore, early life educational programs may benefit by incorporating EF exercises into the core curriculum to strengthen EF early in childhood.

Strengths and limitations

Our study has many strengths, such as evaluating theoretical cause-versus-consequence relationships between EF and cognition and BMI over a 2-year period in a sample of youth who initially had a healthy weight. Notably, the sample used in this study was pooled from a large and geographically diverse group of participants enrolled in the ABCD Study. However, we do recognize the study's limitations. First, the ABCD Study did not collect markers of peripheral (i.e., systemic) or neural inflammation, which effectively limits the insight into the exact causal mechanisms of these relationships apart from the theoretical models [(9, 20)]. Additionally, there were no markers of insulin sensitivity or hormonal and vascular changes, and together this limits the interpretation of our results as to what mechanisms may be driving the association between weight gain and changes to inhibition. Second, the ABCD Study did not collect objective food intake assessments, so it is unknown how our findings relate to actual food intake decisions. Third, it is possible that the EF assessments in the ABCD Study examined general cognitive function, and performance on these tasks may not translate to food-based decision-making. Fourth, the ABCD Study did not assess all components of the NIH Toolbox beyond baseline. Fifth, the sample demographics included in the article differed from the larger consortium participant pool (e.g., higher percentage of youth identifying as White, higher percentage of youth living in families with higher income, lower percentage of youth identifying as Latinx), with the general ABCD Study sample also being relatively homogenous. Relatedly, excluding youth who had overweight or obesity at baseline limits the generalizability of the findings due to potential exclusion of groups who have higher rates of obesity risk (e.g., Black, Latinx). Additionally, we note the limitations of including only individuals with complete data, as this introduces another bias into the data set and dampens the generalizability of the findings. To circumvent bias of a complete case analysis, we ran additional analyses to account for all available data (see online Supporting Information), but the results remained unchanged. However, our results may not be generalizable to diverse groups, and we note that there may have been bias that excluded individuals from participation due to restricting the analyses to only healthy-weight youth. Lastly, BMI is a proxy for adiposity, and the metabolic consequences of an increase in BMI for children are unknown. The ABCD Study did not collect other markers of adiposity or markers of metabolic health.