Longitudinal Associations Between Symptoms of ADHD and BMI From Late Childhood to Early Adulthood

Participants were drawn from the Tracking Adolescents’ Individual Lives Survey (TRAILS), a prospective cohort study of Dutch adolescents consisting of a population cohort (n = 2230) and a clinical cohort (n = 543) of adolescents (52.5% male) with follow-up assessments done biannually or triennially for the past 19 years.¹⁹  The clinical cohort was set up to complement the population sample to increase the variance in problem behaviors by oversampling children at high risk of psychopathology, on the basis of their being in contact with outpatient services before age 11. We used 5 measurement waves from TRAILS. The average age of participants in the sample was 11.1 (SD = 0.54) years at baseline and 22.2 (SD = 0.68) years at time (wave) 5 (see Table 1). Each study wave was approved by the Dutch Central Committee on Research Involving Human Subjects and conducted according to the principles of the Declaration of Helsinki. Informed consent was obtained from parents and adolescents.

ADHD symptoms and depressive symptoms were assessed by using the Achenbach System of Empirically Based Assessment tools.²⁰  We used the average of parent and self-report.²¹  BMI was computed on the basis of objectively measured height and weight. Family functioning was measured by the general functioning scale of the Family Assessment Device, completed by parents.²²  Extensive description of these measurement instruments, as well as of the measurement of SES, pubertal status, medication use, and imputation procedure for missing data are available in the Supplemental Information.

Intraclass correlations were computed for item mean scores of ADHD symptom domains, depressive symptoms, family functioning scores, and BMI, as measures of the within-person variability over time and the between-person variation.

The longitudinal relationships between the ADHD symptom domains and BMI were examined by using the random intercept cross-lagged panel model (RI-CLPM) in a multigroup analyses of male and female participants. The RI-CLPM is an extension of the traditional cross-lagged panel model in which a random intercept is included in the model. The inclusion of a random intercept allows for parsing the variance of a repeatedly measured variable into a part that is stable across the measurement waves and a changing part.²³  Traditional cross-lagged models and other panel data analysis techniques fail to adequately account for trait-like stability of constructs as captured by the autoregressive relationships. As a result, the cross-lagged parameters do not represent the actual within-person relationships over time, and this may lead to a biased conclusion on longitudinal relationships.²³  Thus, the dissection of stable from variable variance over time in RI-CLPM is an important improvement relative to the traditional cross-lagged model. The core of the RI-CLPM model was composed of the relations among the study variables of interest (ie, hyperactivity and impulsivity, attention problems, and BMI) at waves 1, 2, 3, 4, and 5. The estimated model parameters were the cross-lagged effects showing how ADHD symptoms predicts BMI, and vice versa, over time, as well as the autoregressive effects and correlated changes between the 2 ADHD symptom domains and BMI at the within-person level. In addition, the stable associations between these variables of interest were estimated. In the same way, time-varying depressive symptoms and family functioning for the 5 measurement waves were incorporated in the model. In addition, baseline SES and concurrent time-varying pubertal stage and concurrent psychostimulant use were comodeled as possible confounders. Equality constraints between male and female participants on groups of parameters in the model (ie, stepwise, in blocks: effects of confounders, stable correlations among variables, residual correlations, cross-lagged paths, and autoregressive paths) were used to evaluate the equivalence of the estimates across the sexes.

Mplus version 8 was used with full information maximum likelihood estimation. Model fit was determined on the basis of the comparative fit index (CFI), the Tucker–Lewis fit index (TLI), the root mean square error of approximation (RMSEA), and the standardized root mean squared residual (SRMR), with good fit indicated by values CFI >0.95, TLI >0.05, RMSEA <0.08, and SRMR <0.05.²⁴  The χ² difference tests using the Satorra–Bentler scaling correction were used to evaluate potential model fit deterioration of increasingly stringent nested models because of increasing equality constraints on the estimated parameters for male and female participants (see Supplemental Information for additional information).

Descriptive statistics of the variables included in our model across all waves are shown in Table 1. Both hyperactivity and impulsivity and attention problems revealed a decline across development. The percentage overweight increased from 15.2% at measurement wave 1, age 11, to 30.2% at measurement wave 5, age 22. In Fig 1 we confirm that stimulant use may be an important confounder and needs to be adjusted for when studying the reciprocal effects of BMI and ADHD symptoms over time: on average, stimulant users have consistently lower BMI than nonusers, and discontinuation of such treatment is associated with increased BMI. On the basis of the intraclass correlation, 62% of the variance in the 5 measures of hyperactivity and impulsivity score was explained by stable differences between persons across the 5 measurement times, and the remainder 38% was explained by within-person variation at different measurement times. Likewise, 59% of the variance in attention problems and 47% of the variance in BMI across the measurement waves were explained by differences between persons, whereas the remainder of each by within-person variation. This shows that there is substantial within-person change in this developmental period, which is studied by fitting RI-CLPMs to the data.

The RI-CLPM with freely estimated parameters achieved a good fit [χ² (540) = 950.37, CFI = 0.98, TLI = 0.97, RSMEA = 0.023, and SRMR = 0.02] (see Supplemental Information). Next, potential similarities between male and female participants in the RI-CLPM were examined in nested multigroup analyses by testing the invariance of groups of parameter estimates across sex. The model was built up by adding equality constraints between estimates for male and female participants in a stepwise manner and evaluating model fit deterioration by using the χ² difference test (detailed procedures and model fits of this multistep procedure can be found in Supplemental Information). The fit statistics for the final constrained model were χ² (665) = 1082.31, CFI = 0.99, TLI = 0.98, RSMEA = 0.021, and SRMR = 0.03. This final model indicated that male and female individuals did not differ in the cross-lagged paths of main variables (hyperactivity and impulsivity, attention problems, and BMI), other variables (depression, family functioning), and effects of potential confounders (pubertal stage, psychostimulant use, and SES), but they differed in the stable correlations, residual correlations, and autoregressive paths (see Fig 2 for part of the final estimated model focused on hyperactivity and impulsivity, attention problems, and BMI; see Supplemental Fig 3 for the full estimated model including family functioning and depressive symptoms; see Supplemental Table 3 for associations between psychostimulant use and BMI and ADHD symptoms).

As indicated by the estimated correlations between hyperactivity and impulsivity and BMI of our final model, Fig 2 shows that, for both sexes, hyperactive and impulsive symptoms and BMI have a stable modest association between age 11 and age 22 (female: r = 0.102; male: r = 0.086). Such stable association was not found between attention problems and BMI (female: r = 0.065; male: r = 0.015). Importantly, Fig 2 also shows that there are virtually no cross-lagged paths from either hyperactive and/or impulsive and attention problems to BMI or from BMI to hyperactive and/or impulsive and attention problems, indicating the absence of potentially causal effects (β-coefficients range between −.03 and .08). One exception was a longitudinal effect in both male and female adolescents between hyperactive and impulsive symptoms at age 16 and BMI at age 19 (β = .06 in both male and female adolescents, P = .026 [male] and .029 [female]). This cross-lagged effect was studied further in post hoc analyses, making the distinction between no use of stimulants, discontinued use of stimulants, and continued use of stimulants (see Fig 1). In doing so, we checked if found cross-lagged effect of hyperactivity and impulsivity could have been confounded by a rebound effect from individuals who discontinued stimulant use. Following this, the cross-lagged increase in BMI at age 19 by hyperactivity at age 16 became nonsignificant. This also happened when we removed all individuals who discontinued medication from the sample. Finally, wee checked if the absence of longitudinal effects was explained by family functioning or depression by removing both variables at all time points from the RI-CLPM and refitting it. This was not the case, leaving unaltered the finding that there are no direct links between ADHD and BMI in this developmental period.

In the current study, we examined the longitudinal effects between ADHD symptoms and BMI from preadolescence to young adulthood, as separate from the stable associations between these variables in this period. Except for a weak positive cross-lagged effect from hyperactive and impulsive symptoms at age 16 to BMI at age 19 found in both male and female participants, there was no evidence of reciprocal effects between ADHD symptoms and BMI over time. Our data further indicated a modest and stable positive association between the hyperactive and impulsive domain of ADHD and BMI. We conclude that over the course of adolescence, the link between ADHD and BMI is highly stable and is predominant with hyperactive and impulsive symptoms rather than attention problems. Neither depressive symptoms nor poor family functioning explained the absence of effects between ADHD and BMI over time. The findings thus point to a shared genetic or familial background and/or causal effects established already earlier in childhood.

The association between ADHD and BMI, overweight, or obesity has rarely been studied in adolescence,¹⁸  and longitudinal studies are even rarer. The current study was the first longitudinal study in which researchers modeled stable relationship as separate from direct effects over time from late childhood through adolescence to young adulthood. Another study that bears resemblance to our cross-lagged approach and which, so far, is the only study in which researchers separated stable from direct effects over time was focused on childhood. The researchers found that ADHD symptoms at age 3 and 6 predicted fat mass and BMI at age 9.¹⁰  This may potentially indicate that the longitudinal effect of ADHD on BMI may be established early in childhood, which later presents as a stable correlation during the adolescent period as shown by the current findings. Of note, neither that study nor our own revealed evidence of a reverse effect (ie, high BMI leading to more hyperactivity and impulsivity or attention problems).

The absence of a reverse effect contrasts with recent findings from MR analysis, a technique using genetics as an instrumental variable for causal inference. Building on the most recent GWAS of ADHD, the first published MR study suggested stronger evidence for an effect of high BMI on ADHD liability than for an effect in the opposite direction.⁹  More recently published MR and polygenic score analyses reported bidirectional causal effect between ADHD and BMI⁹  and between ADHD and childhood obesity, but not BMI (GWAS on adult sample).^7,8  Thus, based on the combined evidence from genetic and longitudinal studies, the direction of causality between ADHD symptoms and BMI is still unclear, although findings do point to an important role of genetics in the co-occurrence of ADHD and high BMI,^6–9,25  with current evidence pointing to childhood rather than adolescence as a key developmental period in which the association between ADHD and high BMI is established.^7,10  Note, however, that MR studies are still subject to environmental confounding.^13,26 

Reviews and meta-analyses of cross-sectional data have firmly established that the co-occurrence rates of ADHD and obesity are higher in adulthood than in early life.^3,17,18  Our study was set up to identify direct effects between ADHD symptoms and BMI in between childhood and adulthood, but they were absent, so we could not bridge this gap. ADHD symptoms during adulthood are associated with poor occupational outcomes, being arrested, sedentary behavior, poor diet, financial struggle, and divorce,^27–30  all of which are associated with an unhealthy lifestyle and, consequently, overweight and obesity. Thus, it might be speculated that the stronger association between ADHD and BMI in adulthood may become established later when lifestyle habits outside the family of origin become engrained. Importantly, such changes may still be genetically driven, consistent with age-dependent genetic influences.^8,25  The mechanisms leading to higher comorbidity between ADHD and BMI in adults relative to primary school–aged children^3,18,30  needs to be studied in future research to develop effective interventions.

The current study has several key strengths. Studies on the (cross-sectional) association between ADHD and BMI tend to be focused on either childhood or adulthood, with a notable lack of studies on adolescence.¹⁸  Moreover, the longitudinal nature of our study, with 5 repeated measurement waves providing, for the first time, the bridge from childhood to adulthood; the inclusion of critical time covarying confounders; and the large sample size are important assets of the study. To avoid monoinformant bias, we used both parent- and self-reported ADHD symptoms. Another strength was the inclusion of important and time-varying confounders. Finally, the RI-CLPM model used in the current study disentangles (unlike the previous cross-lagged panel models) trait-like associations between the variables from changing relation predicted by each other over time. To the extent that direct cross-lagged associations over time can be identified, and provided that appropriate time-varying covariates are comodeled, the use of the RI-CLPM is an important first step in establishing potential causality. Their absence, as currently found for most of the estimated paths in our study, is likewise highly informative.

The absence of parent report at wave 4 and the selective attrition were limitations of the study. The missing parent report at wave 4 was imputed by using all available parent and self-report information in the other waves, estimating individual trajectories using state-of-the-art methods (see Supplemental Information). Regarding selective attrition, although the missing values were treated by using full information maximum likelihood, the findings of this study may underestimate the relationship between ADHD symptoms and BMI given that more-severely affected participants dropped out of the study somewhat more often. However, the inclusion of the TRAILS clinical cohort that oversamples individuals with ADHD symptoms and overweight makes underestimation of the associations when the aim is to generalize to the general population less likely. Given that in the current study we combined a general population sample and a clinical sample, a further limitation of this study was that the effects of stimulant use could only be studied in a relatively small proportion of the total sample. Despite this, we illustrated differences in BMI between no, continued, and discontinued medication use over time. We confirmed previous findings that stimulant use goes together with reduced BMI, including the rebound effect that discontinuation of stimulants leads to higher BMI,^14,15  such that over time, persons with ADHD who discontinued over time approached the BMI of persons with ADHD who did not use stimulants. A final limitation is that the RI-CLPM has not been developed, so far, for categorical outcome variables. Therefore, we could not analyze dichotomous outcomes (by dichotomizing ADHD scores at, eg, the 80th percentile). Given that associations may strengthen at the highest levels of symptom severity, current findings may have underestimated effects over time. Together, these 2 limitations indicate the usefulness of method development and future study of longitudinal effects in a large clinically diagnosed sample.

This study revealed no evidence of longitudinal bidirectional association where ADHD symptoms lead to greater BMI and/or high BMI leads to enhanced persistence or deterioration of ADHD symptoms during adolescence and young adulthood. Thus, our hypothesis that weight build-up may occur between childhood and adulthood through direct effects over time by ADHD symptoms and BMI reciprocally influencing one another was not born out. Rather, our main finding was that the association between hyperactive and impulsive symptoms and BMI in this developmental period is highly stable. An exception to this was a small direct effect of hyperactivity and/or impulsivity on BMI in both men and women between age 16 and 19, which requires further replication. Therefore, the outcome of this study mainly points to a shared genetic or familial background and/or potential direct causal effects already established earlier in childhood, although, direct effects may be present again in adulthood. From a clinical point of view, clinicians should be aware that adolescents presenting with overweight and obesity might have concurrent ADHD and that intervention and preventive strategies should be implemented early on in the development before the association becomes stable when evidence based interventions for ADHD may have a significant impact in terms of preventing or reducing overweight and obesity.³¹ 

This research is part of TRAILS. Participating centers of TRAILS include various departments of the University Medical Center and University of Groningen, the University of Utrecht, the Radboud Medical Center Nijmegen, and the Parnassia Bavo group, all in the Netherlands. TRAILS has been financially supported by various grants from the Netherlands Organization for Scientific Research, ZonMW, GB-MaGW, the Dutch Ministry of Justice, the European Science Foundation, the European Research Council, BBMRI-NL, and the participating universities. We are grateful to everyone who participated in this research or worked on this project to make it possible.

ADHD

attention-deficit/hyperactivity disorder

CFI

comparative fit index

GWAS

genome-wide association study

MR

Mendelian randomization

RI-CLPM

random intercept cross-lagged panel model

RMSEA

root mean square error of approximation

SES

socioeconomic status

SRMR

standardized root mean squared residual

TLI

Tucker–Lewis fit index

TRAILS

Tracking Adolescents Lives Survey