Introduction
ADHD is characterized by symptoms of inattention, impulsivity, and hyperactivity (DSM-5; APA, 2013), and has been associated with high BMI, across the lifespan (Cortese et al., Reference Cortese, Moreira-Maia, Fleur, Morcillo-Peñalver, Rohde and Faraone2016; Liu et al., Reference Liu, Schoeler, Davies, Peyre, Lim, Barker, Llewellyn, Dudbridge and Pingault2021; Nigg et al., Reference Nigg, Johnstone, Musser, Long, Willoughby and Shannon2016). A possible mechanism linking ADHD and BMI is binge eating. Binge eating have been commonly reported in both individuals with ADHD and obesity (Cortese, Bernardina, & Mouren, Reference Cortese, Bernardina and Mouren2007; de Zwaan, Reference de Zwaan2001; Svedlund, Norring, Ginsberg, & von Hausswolff-Juhlin, Reference Svedlund, Norring, Ginsberg and von Hausswolff-Juhlin2017). Understanding the mechanisms underlying the links between ADHD, binge eating, and BMI is a priority both for people with lived experience of neurodiversity and for public health (Cortese et al., Reference Cortese, Bernardina and Mouren2007; Kaisari, Dourish, & Higgs, Reference Kaisari, Dourish and Higgs2017; Keller, Herle, Mandy, & Leno, Reference Keller, Herle, Mandy and Leno2024).
The association between ADHD and BMI has been well-established (Cortese et al., Reference Cortese, Moreira-Maia, Fleur, Morcillo-Peñalver, Rohde and Faraone2016). Phenotypic studies have shown that ADHD and its symptoms are associated with higher BMI and an increased risk of overweight from childhood through early adulthood (Fliers et al., Reference Fliers, Buitelaar, Maras, Bul, Höhle, Faraone, Franke and Rommelse2013; Fuemmeler et al., Reference Fuemmeler, Østbye, Yang, McClernon and Kollins2011). Mendelian randomization analyses using polygenic scores have indicated bidirectional causal effects between ADHD and BMI, with genetic liability to ADHD influencing BMI and, conversely, genetic liability to BMI influencing ADHD (Liu et al., Reference Liu, Schoeler, Davies, Peyre, Lim, Barker, Llewellyn, Dudbridge and Pingault2021).
Binge eating symptoms has been proposed as an important risk factor for high BMI among individuals with ADHD (Cortese et al., Reference Cortese, Bernardina and Mouren2007; Svedlund et al., Reference Svedlund, Norring, Ginsberg and von Hausswolff-Juhlin2017). Difficulties in executive functioning and impulsivity may contribute to the higher prevalence of disordered eating behaviors in individuals with ADHD (Keller et al., Reference Keller, Herle, Mandy and Leno2024; Martin et al., Reference Martin, Dourish, Hook, Chamberlain and Higgs2022; Nazar et al., Reference Nazar, de Sousa Pinna, Suwwan, Duchesne, Freitas, Sergeant and Mattos2016), and such behaviors may also serve as coping mechanisms in individuals with neurodevelopmental disorders (e.g. Makin et al., Reference Makin, Zesch, Meyer, Mondelli and Tchanturia2025; Pruccoli et al., Reference Pruccoli, Guardi, La Tempa, Valeriani, Chiavarino and Parmeggiani2023). Compared to its prevalence in the general population (1.4%), ADHD is more common among individuals with binge-eating-related disorders, with 26–31% of these individuals also scoring above the threshold for ADHD (Svedlund et al., Reference Svedlund, Norring, Ginsberg and von Hausswolff-Juhlin2017). In addition to studies conducted with clinical samples, binge eating symptoms are also associated with self-reported ADHD symptoms (Hanson, Phillips, Hughes, & Corson, Reference Hanson, Phillips, Hughes and Corson2020). Prospective associations have also been reported between ADHD and binge eating symptoms; for example, childhood ADHD (before age 12 years) predicts binge eating behaviors in adolescence (ages 14–16 years) and adulthood (over age 18 years) (Levin & Rawana, Reference Levin and Rawana2016; Sonneville et al., Reference Sonneville, Calzo, Horton, Field, Crosby, Solmi and Micali2015).
It is therefore possible that binge eating symptoms partially mediate the relationship between ADHD and BMI, with ADHD symptoms potentially leading to higher BMI through binge eating. Some empirical evidence has provided preliminary support for the mediating role of binge eating symptoms in the relationship between ADHD and BMI. For example, in children aged 12–15 years with obesity, impulsivity was associated with impulsive eating (Nederkoorn et al., Reference Nederkoorn, Braet, Van Eijs, Tanghe and Jansen2006). Similarly, a study of adults aged 25–46 years found that ADHD influences overweight by increasing overeating behaviors, including binge eating symptoms (Davis et al., Reference Davis, Levitan, Smith, Tweed and Curtis2006).
It is noteworthy that mechanisms underlying the associations between binge eating symptoms and the two core subdomains of ADHD (hyperactivity-impulsivity and inattention) may differ. Both subdomains have been associated with binge eating symptoms: inattention with deficits in attentional control and hyperactivity-impulsivity with impaired behavioral control (Tistarelli et al., Reference Tistarelli, Fagnani, Troianiello, Stazi and Adriani2020). However, existing research has primarily focused on the association between hyperactivity-impulsivity and binge eating symptoms (e.g. Giel et al., Reference Giel, Teufel, Junne, Zipfel and Schag2017; Schag et al., Reference Schag, Schönleber, Teufel, Zipfel and Giel2013), whereas the role of inattention has received less attention. The inattention subdomain may also be linked to binge eating symptoms due to factors such as the impaired perception or utilization of interoceptive signals to guide appetitive behavior. Supporting this, studies have reported associations between inattention symptoms and disordered eating risk, suggesting that inattention, like hyperactivity–impulsivity, may increase vulnerability to binge eating symptoms (Martin et al., Reference Martin, Dourish, Rotshtein, Spetter and Higgs2019; Martin, Schell, Srivastav, & Racine, Reference Martin, Schell, Srivastav and Racine2020). This highlights the need for further exploration into the distinct roles of different ADHD subdomains in relation to binge eating symptoms.
Understanding underlying associations between neurodiversity and disordered eating was a key outcome of lived experience priority-setting exercises (Keller et al., Reference Keller, Herle, Mandy and Leno2024; Thomas, Cooper, & Jones, Reference Thomas, Cooper and Jones2025). The current study aims to explore the predictive role of ADHD in childhood, adolescence, and early adulthood on BMI at age 26 years, and examine the mediating role of binge eating symptoms at age 21 years. To investigate whether potential mediation effects are confounded by genetic factors, genetically informative mediation models were used (Rosenström et al., Reference Rosenström, Czajkowski, Ystrom, Krueger, Aggen, Gillespie, Eilertsen, Reichborn-Kjennerud and Torvik2019). Based on previous research, we hypothesized that binge eating symptoms at age 21 years would partially mediate the association between ADHD and adult BMI at age 26 years, after controlling for shared genetic influences.
Method
This research was preregistered on the Open Science Framework (OSF) website, with a registration DOI of 10.17605/OSF.IO/24GEW.
Participants
The current study used data from the Twins Early Development Study (TEDS), a longitudinal twin cohort following participants from ages 4 to 26 years (Lockhart et al., Reference Lockhart, Bright, Ahmadzadeh, Breen, Bristow, Boyd, Downs, Hotopf, Palaiologou, Rimfeld, Maxwell, Malanchini, McAdams, McMillan, Plomin and Eley2023; Rimfeld et al., Reference Rimfeld, Malanchini, Spargo, Spickernell, Selzam, McMillan, Dale, Eley and Plomin2019; Trouton, Spinath, & Plomin, Reference Trouton, Spinath and Plomin2002). TEDS includes twins born in England and Wales between 1994 and 1996. The initial sample comprised 13,926 twin pairs. Participants with severe medical conditions, extreme adverse perinatal circumstances, missing essential background variables, or missing zygosity or gender information were excluded. After screening, 4,363 monozygotic (MZ) and 8,575 dizygotic (DZ) twins remained, with males representing 49.9% of the sample. Over 7,500 twin pairs provided ADHD data in early childhood, and more than 4,500 continued to do so in early adulthood. The sample size for each measure is shown in Supplementary Table S1.
Measures
ADHD
Latent factors were specified for ADHD, indicated by multiple measurements (see Supplementary Table S2). In childhood, ADHD symptoms were measured using the parent-report Strengths and Difficulties Questionnaire (SDQ; Goodman, Reference Goodman1997) at ages 4, 7, and 9 years, as well as the parent-reported Conners’ Rating Scales-Revised (CRS-R; Conners, Epstein, Angold, & Klaric, Reference Conners, Epstein, Angold and Klaric2003) at age 8 years. During adolescence, ADHD symptoms were assessed using the parent-reported Conners scale and SDQ at ages 12, 14, and 16 years. During early adulthood (21 years old), ADHD symptoms were assessed with the parent-reported and self-reported SDQ and parent-reported Conners.
We further divided the measured ADHD symptoms into the inattention subdomain and the hyperactivity-impulsivity subdomain based on the content of the questionnaire items. The specific measurements of the ADHD subdomains were consistent with those of overall ADHD. The only exception was the measurement of childhood inattention: to ensure a good fit, it was necessary to exclude the age 4 SDQ. Childhood inattention therefore comprised inattention items from the SDQ at ages 7 and 9 years, and the Conners at age 8 years.
Binge eating symptoms
Binge eating symptoms were measured by a brief 12-item questionnaire on eating disorders at 21 years old. This scale was developed by Garner (Reference Garner1991). We selected the three items of the binge eating subscale in this questionnaire, including ‘I stuff myself with food’, ‘I think about binging (overeating)’, and ‘I eat or drink in secrecy’.
BMI
Height and weight were self-reported at the age of 26 years, and BMI was calculated using the formula weight (kg)/height (m)2 (Bray, Reference Bray2011).
Covariates
In this study, we controlled for participants’ sex and age. In addition, given the relative stability of BMI across development, baseline BMI was controlled for in each model: BMI measured at age 4 years in the childhood model, age 12 years in the adolescent model, and age 21 years in the adulthood model.
Data analysis
The phenotypic mediation model
Mediation of the association between ADHD and BMI via binge eating symptoms was tested in lavaan (version 0.6–20) using models including the direct effect of ADHD on BMI and the indirect effect through binge eating symptoms (see Figure 1). The indirect effect was estimated as the product of the ADHD → binge eating symptoms and binge eating symptoms → BMI paths, with significance assessed using 5,000 bootstrap samples. Because the ADHD construct included repeated measures of the same questionnaire (e.g. SDQ) and questionnaires with substantial item overlap (Conners and SDQ), we allowed (1) residual correlations between the Conners and SDQ reported by the same informant at the same time point, and (2) residual correlations between adjacent time points for repeated measurements of the same questionnaire by the same informant. This approach has been supported in previous research (Landis, Edwards, & Cortina, Reference Landis, Edwards, Cortina, Lance and Vandenberg2009).
The phenotypic factor mediation model. (a) This model illustrates the mediating role of binge eating symptoms in the relationship between ADHD and BMI. ADHD symptoms were assessed using multiple measures at each stage. Binge eating symptoms and BMI were measured at ages 21 and 26, respectively. a: direct path from ADHD to binge eating; b: direct path from binge eating to BMI; c’: direct path from ADHD to BMI; Indirect effect = a × b; Total effect = c′ + (a × b). (b) This shows the coefficients for the phenotypic factor mediation models in childhood, adolescence, and adulthood, respectively.
Figure 1. Long description
The figure consists of two panels, a and b, each displaying a path diagram.
Panel a illustrates the conceptual mediation model. At the top left, an oval labeled A D H D serves as the primary predictor, with four arrows pointing downward to rectangular indicators labeled A D H D 1, A D H D 2, A D H D 3, and A D H D 4. A horizontal arrow labeled c prime points from A D H D to an oval on the top right labeled B M I. Below B M I is a rectangular indicator also labeled B M I. Between A D H D and B M I, an oval labeled Binge eating is positioned lower in the center. An arrow labeled a points from A D H D to Binge eating, and an arrow labeled b points from Binge eating to B M I. A rectangular indicator labeled B E sits below the Binge eating oval.
Panel b displays the same structural layout but replaces the path labels with numerical coefficients for childhood, adolescence, and adulthood, separated by forward slashes.
* The direct path from A D H D to B M I shows coefficients 0.49 asterisk in blue, 0.26 asterisk, and 0.07 asterisk.
* The path from A D H D to Binge eating shows coefficients 0.07 in blue, 0.17 asterisk, and 0.44 asterisk.
* The path from Binge eating to B M I shows coefficients 0.44 asterisk in blue, 0.31 asterisk, and 0.06 asterisk.
In all paths, the first value representing childhood is highlighted in blue text.
Multivariate genetic model fitting
In twin studies, monozygotic twins share 100% of their genetic material, whereas dizygotic twins share (on average) 50% of their genetic variance. When phenotypic similarity between monozygotic twins is higher than that of dizygotic twins, it indicates genetic influence (Rijsdijk & Sham, Reference Rijsdijk and Sham2002). Therefore, analysis of twin data makes it possible to estimate the extent of genetic and environmental influences on traits.
Specifically, we divided the variances and covariances of ADHD, binge eating symptoms, and BMI, as well as the residual variances of their indicator variables, into latent additive genetic components (A), shared environmental components (C), dominant genetic components (D), and non-shared environmental components (E). The A represents the cumulative effects of multiple genetic loci across the genome. D refers to the effects of alleles that can ‘mask’ or ‘dominate’ the effect of another. The C represents environmental influences that make family members similar to one another, while E includes environmental factors that differentiate family members. MZ twins typically share 100% of their A and dominant genetic components (D), whereas DZ twins typically share 50% of their A and 25% of their D. D components and C components cannot be estimated at the same time in the classical twin design, because when r(MZ) > 2 × r(DZ), the data suggest the influence of D; conversely, when r(MZ) < 2 × r(DZ), it suggests the influence of C (Boomsma, Busjahn, & Peltonen, Reference Boomsma, Busjahn and Peltonen2002; Knopik, Neiderhiser, DeFries, & Plomin, Reference Knopik, Neiderhiser, DeFries and Plomin2017).
Previous research in the same sample, using these variables, has indicated that ADHD measured at different stages follows an A, D, and E structure (You et al., Reference You, Oginni, Rijsdijk, Lim, Zavos and McAdams2024). We therefore adopted the A, D, and E structure for the ADHD variables in this study. In addition, we also included sibling interaction effects in line with previous research on ADHD (Merwood et al., Reference Merwood, Chen, Rijsdijk, Skirrow, Larsson, Thapar, Kuntsi and Asherson2014; Rietveld, Posthuma, Dolan, & Boomsma, Reference Rietveld, Posthuma, Dolan and Boomsma2003; You et al., Reference You, Oginni, Rijsdijk, Lim, Zavos and McAdams2024). For binge eating symptoms and BMI, the r(MZ) was closer to 2 × r(DZ), suggesting that the contributions of C or D were small. We further validated the etiological structure of binge eating symptoms and BMI within the univariate twin models (see Supplementary Tables S3–S4). The results suggested that in the ACE model, the C components had no significant effects on either binge eating symptoms or BMI. Therefore, we tested simplified models in which the C components were removed. After dropping the C components, the simplified models were not significantly different from the full ACE models (p > 0.05). Therefore, in the subsequent analyses, binge eating symptoms and BMI were modeled using an A, E structure.
Biometric models
To examine the etiological structure between ADHD, binge eating symptoms, and BMI, we constructed several biometric genetic models in OpenMx (version 2.15.5) using full-information maximum likelihood (FIML): a Cholesky decomposition model, an independent pathway model, and a modified biometric mediation model. The first two models specify etiological overlaps between variables, whereas the latter model includes directional paths. In the Cholesky decomposition model, the first common genetic and environmental factors (A1 and E1) influenced all three traits (ADHD, binge eating symptoms, and BMI), indicating their shared genetic and environmental effects. The second common genetic and environmental factors (A2 and E2) influenced binge eating symptoms and BMI, representing the residual shared genetic effects between the two traits after accounting for A1 and E1. The third genetic and environmental factors (A3 and E3) influenced only BMI, representing the genetic and environmental effects unique to BMI. In addition, each ADHD measurement was influenced by specific A, D, and E components (see Figure 2a). In the independent pathway model, the variables were influenced by a common set of genetic and environmental factors (see Figure 2b). The A and E components loaded on all three traits, while each observed variable was also affected by variable-specific A, DADHD, and E components (see Figure 2b). In the modified biometric mediation model, the mediation was built on a three-variable correlation model by adding directional paths between variables. On this basis, a direct path from ADHD to BMI and an indirect path from ADHD to BMI via binge eating symptoms were specified (see Figure 2c). Additionally, in the modified biometric mediation model, the correlations in the common A factors among ADHD, binge eating symptoms, and BMI were estimated, although shared E was not modeled to allow for the estimation of the mediation paths. Each ADHD measurement was also influenced by specific A, D, and E components (see Figure 2d).
The biometric models. (a) The Cholesky decomposition model of ADHD, binge eating symptoms, and BMI. This model indicates that genetic and environmental effects influence the variables sequentially. (b) The independent pathway model of ADHD, binge eating symptoms, and BMI. This model indicates that the variables were influenced by a common set of genetic and environmental factors. (c) The biometric mediation model of ADHD, binge eating symptoms, and BMI. This model indicates that ADHD influences BMI through binge eating symptoms. (d) The modified biometric mediation model of ADHD, binge eating symptoms, and BMI. This model indicates that ADHD influences BMI through binge eating symptoms after controlling for shared genetic effects. A: additive genetic components; D: dominant genetic components; E: non-shared environmental components. ADHD symptoms were assessed separately in childhood, adolescence, and adulthood, with each stage incorporating multiple measures. Binge eating symptoms were measured at age 21 years. BMI was measured at age 26 years.
Figure 2. Long description
Four path diagrams illustrate biometric models.
Panel a, Cholesky decomposition model. At the top, latent factors A 1, D, E 1, A 2, E 2, A 3, and E 3 point downward. A 1, D, and E 1 point to A D H D. A 1, A 2, and E 2 point to Binge eating. A 1, A 2, A 3, E 2, and E 3 point to B M I. Below these, A D H D points to four indicators labeled A D H D 1 through A D H D 4, each influenced by individual A, D, and E circles. Binge eating and B M I point to single B E and B M I indicators.
Panel b, Independent pathway model. Top latent factors D, A, and E point directly to all three primary variables: A D H D, Binge eating, and B M I. The indicator structure at the bottom remains the same as panel a.
Panel c, Biometric mediation model. Latent factors A, D, and E point to A D H D. Separate A and E factors point to Binge eating, and separate A and E factors point to B M I. Horizontal arrows show A D H D pointing to Binge eating, and Binge eating pointing to B M I.
Panel d, Modified biometric mediation model. Similar to panel c, but with curved double-headed arrows connecting the top A factors for A D H D, Binge eating, and B M I, labeled a sub 1 2, a sub 1 3, and a sub 2 3, indicating shared genetic effects.
Model fit was assessed using the Root Mean Square Error of Approximation (RMSEA; Steiger & Lind, Reference Steiger and Lind1980), the Tucker-Lewis Index (TLI; Tucker & Lewis, Reference Tucker and Lewis1973), and the Comparative Fit Index (CFI; Bentler, Reference Bentler1990). A well-fitting model was considered to meet the criteria of RMSEA <0.08, TLI > 0.90, and CFI > 0.90. Model comparisons for non-nested models were conducted using the Akaike Information Criterion (AIC) and AIC weights (Akaike, Reference Akaike, rov and Caski1973, Reference Akaike1974), as well as the −2 Log-Likelihood statistic (−2LL). Generally, a smaller AIC that differs by more than 3 provides some support for a better model fit (Burnham & Anderson, Reference Burnham and Anderson2004). The comparison of −2LL between the nested models was conducted using a chi-square distribution. A nonsignificant p-value indicates that, after removing the paths, the more simplified model does not fit the data significantly worse than the original model.
To examine how ADHD in childhood, adolescence, and adulthood predicts BMI through the mediating effect of binge eating symptoms at each developmental stage, as well as how genetic and environmental factors influence these mediating effects, the above analyses were repeated separately for each developmental period. Additionally, we conducted the exploratory analyses separately for the two ADHD subdomains, inattention and hyperactivity-impulsivity.
Results
Descriptive statistics
Descriptive statistics of the observed variables are shown in Supplementary Table S5. In general, symptoms of parent reported ADHD decreased with age from a mean of 4.02–1.96 from ages 4 to 21 years, respectively. The average binge eating symptoms score, measured at age 21 years, was 3.75. The mean BMI in adulthood (age 26 years) was 24.74, with 23.0% of participants classified as overweight (25 ≤ BMI < 30) and 13.8% classified as obese (BMI ≥ 30). Phenotypic correlations among ADHD, binge eating symptoms, and BMI are shown in Supplementary Table S6. ADHD symptoms in adulthood (age 21 years) showed the highest correlation with binge eating symptoms at age 21 years (r = 0.25), while childhood ADHD symptoms (r = 0.06) and adolescent ADHD symptoms (r = 0.10) had significantly lower correlations with binge eating symptoms at age 21 years. Therefore, overall correlations between ADHD symptoms and binge eating symptoms across all time points were small, with only a slight increase between symptoms measured at the same time point. ADHD symptoms at each age also showed significant but small correlations with BMI at age 26 years (r = 0.11–0.18). The correlation between binge eating symptoms and BMI was 0.24.
Phenotypic mediation models
In the results of the phenotypic mediation models (see Table 1 and Supplementary Table S7), the paths from ADHD to binge eating symptoms were not significant in childhood, but were significant in adolescence (estimate = 0.17, 95% CI: 0.10–0.25) and adulthood (estimate = 0.44, 95% CI: 0.38–0.53). Paths from ADHD to BMI were significant at all three stages, with the estimates of 0.49 (95% CI: 0.30–0.70) in childhood, 0.26 (95% CI: 0.15–0.38) in adolescence, and 0.07 (95% CI: 0.02–0.13) in adulthood. The indirect associations between ADHD and adult BMI through binge eating symptoms were significant in adolescence and adulthood, but not childhood. Specifically, during adolescence, the mediated effect was 0.05 (95% CI: 0.03–0.08); during adulthood, the mediated effect was 0.03 (95% CI: 0.01–0.05).
The phenotypic mediation model of ADHD, binge eating symptoms, and BMI
Table 1. Long description
The table is organized into three main sections representing life stages, each detailing direct and indirect effects with unstandardized estimates, 95 percent C Is (Lower and Upper), S E, and standardized estimates.
1. Childhood A D H D:
* Direct effect of A D H D to B M I: Unstandardized 0.49, Standardized 0.13.
* Direct effect of A D H D to Binge eating: Unstandardized 0.07, Standardized 0.03.
* Direct effect of Binge eating to B M I: Unstandardized 0.44, Standardized 0.23.
* Indirect effect (A D H D to Binge eating to B M I): Unstandardized 0.03, Standardized 0.01.
2. Adolescent A D H D:
* Direct effect of A D H D to B M I: Unstandardized 0.26, Standardized 0.10.
* Direct effect of A D H D to Binge eating: Unstandardized 0.17, Standardized 0.10.
* Direct effect of Binge eating to B M I: Unstandardized 0.31, Standardized 0.20.
* Indirect effect (A D H D to Binge eating to B M I): Unstandardized 0.05, Standardized 0.02.
3. Adulthood A D H D:
* Direct effect of A D H D to B M I: Unstandardized 0.07, Standardized 0.05.
* Direct effect of A D H D to Binge eating: Unstandardized 0.44, Standardized 0.30.
* Direct effect of Binge eating to B M I: Unstandardized 0.06, Standardized 0.06.
* Indirect effect (A D H D to Binge eating to B M I): Unstandardized 0.03, Standardized 0.02.
Note: CI, confidence interval; SE, standard error.
Biometric models
Supplementary Table S6 presents the estimates of A, D, and E for each variable from the best-fitting models, and Supplementary Table S4 shows the estimates of A and E for binge eating symptoms and BMI from the univariate twin models. ADHD symptoms were consistently influenced by genetic factors. In childhood, dominant genetic effects were the primary influence (d 2 = 0.75), whereas in adolescence and adulthood, additive genetic influences increased (h 2 = 0.73 and 0.43, respectively). The proportion of variance accounted for by non-shared environmental factors was small to moderate from childhood to adulthood (e 2 = 0.14–0.26). In addition, binge eating symptoms were influenced by substantial environmental effects (e 2 ≈ 0.67), along with moderate additive genetic effects (h 2 ≈ 0.33). For BMI, in the estimates from the univariate twin models, additive genetic effects were the primary influence (h 2 = 0.75), while the proportion of variance accounted for by non-shared environmental factors was small (e 2 = 0.25). The heritability estimates for BMI from the best-fitting multivariate models (Supplementary Table S6) were somewhat reduced compared to univariate estimates as BMI at the previous time point was included in analyses to control for previous BMI levels.
The results of model comparisons are shown in Tables 2 and 3 and Supplementary Table S8. The AIC differences between the modified biometric mediation model (childhood: AIC = 264,124; adolescence: AIC = 267,719) and the Cholesky decomposition model (childhood: AIC = 264,124; adolescence: AIC = 267,720) were <1 in childhood and adolescence. This suggests no strong evidence to support either model. In the modified biometric mediation models, genetic correlations were observed among the variables. After accounting for genetic overlap, only the paths from binge eating symptoms to BMI were significant. The Cholesky decomposition model makes no assumptions about direction of effect between variables; based on parsimony, it was selected as the best-fitting model. This model included significant shared genetic components among ADHD, binge eating symptoms, and BMI. Specifically, the first genetic factor (A1) significantly loaded on ADHD (childhood: a 11 = 0.03, 95% CI: 0.02–0.05; adolescence: a 11 = 0.73, 95% CI: 0.62–0.84), binge eating symptoms (childhood: a 12 = 0.16, 95% CI: 0.06–0.26; adolescence: a 12 = 0.01, 95% CI: 0.01–0.02), and BMI (childhood: a 13 = 0.63, 95% CI: 0.42–0.74; adolescence: a 13 = 0.04, 95% CI: 0.02–0.06). The second genetic factor (A2) influenced binge eating symptoms (childhood: a 22 = 0.17, 95% CI: 0.07–0.28; adolescence: a 22 = 0.32, 95% CI: 0.27–0.36) and BMI (childhood: a 23 = 0.11, 95% CI: 0.004–0.31; adolescence: a 23 = 0.05, 95% CI: 0.02–0.09), representing the residual shared genetic effects between the two after accounting for A1. Apart from A1 and A2, no unique genetic effects (A3) on BMI were observed in childhood, whereas it was still observed in adolescence (a 33 = 0.54, 95% CI: 0.49–0.58). Regarding environmental influences, common non-shared environmental influences among the traits were small and nonsignificant. This indicates that the E on the variables were independent of each other, with estimates of 21% (95% CI: 0.20–0.23), 67% (95% CI: 0.63–0.71), and 26% (95% CI: 0.23–0.29) for ADHD, binge eating symptoms, and BMI in childhood, respectively, and 14% (95% CI: 0.13–0.15), 67% (95% CI: 0.63–0.71), and 36% (95% CI: 0.33–0.40) in adolescence. Therefore, the associations between childhood and adolescent ADHD, adulthood binge eating symptoms, and BMI were largely explained by common genetic factors, although the pathway from binge eating symptoms to BMI may also be significant.
Fit statistics for Cholesky decomposition models, independent pathway models, and modified biometric mediation models at each stage_ADHD
Table 2. Long description
The table presents fit statistics across ten columns: Model, -2 L L, ep, A I C (weight), df, C F I, T F I, R M S E A, p, and Nested p. It is divided into three sections:
1. Childhood biometric model: The Cholesky model (bold) and Modified Biometric mediation model show the best fit. The Cholesky model has -2 L L of 310,669, ep of 37, A I C of 310,744 (0.26), and C F I of 0.98. The Modified Biometric mediation model has an A I C weight of 0.71.
2. Adolescence biometric model: The Cholesky model (bold) is the best fit with -2 L L of 321,419, ep of 41, A I C of 321,501 (0.62), C F I of 0.9, and R M S E A of 0.04.
3. Adulthood biometric model: Both the Cholesky and Independent pathway models (bold) show identical best-fit statistics: -2 L L of 215,429, ep of 30, A I C of 215,489 (0.40), C F I of 0.95, and R M S E A of 0.02.
Across all stages, the p-values for model fit are 0, and R M S E A values range from 0.01 to 0.04, indicating good to acceptable fit. Initialisms include A I C (Akaike information criterion), C F I (comparative fit index), T F I (Tucker-Lewis fit index), and R M S E A (root mean square error of approximation).
Note: -2LL, minus 2 × log-likelihood of the comparison model; ep, estimated parameters of the comparison model; AIC, Akaike information criterion; df, degrees in freedom of the comparison model; CFI, comparative fit index; TFI, Tucker-Lewis fit index; RMSEA, root mean square error of approximation; p: p refers to the significance level of the chi-square test of model fit; Nested_p: Nested_p indicates whether the simplified model fits significantly worse than the baseline model. When p > 0.05, it means there is no significant difference between the two models.
Modified biometric mediation model_nested: we tested a simplified model in which the paths from ADHD to binge eating symptoms and ADHD to BMI were dropped. This simplified model was nested within the full modified biometric mediation model.
The models in bold type are the best-fitting models.
Parameter estimates (95% confidence intervals) in every stage from the best-fitting model
Table 3. Long description
The table contains four rows of data corresponding to different developmental models: Childhood_a, Adolescence_a, Adulthood_a, and Adulthood_b.
Columns are labeled with parameter codes: a sub 11, a sub 12, a sub 13, a sub 22, a sub 23, a sub 33, e sub 11, e sub 12, e sub 13, e sub 22, e sub 23, e sub 33, d sub 11, and Path sub 2-3.
Key data points include:
* Childhood_a: a sub 11 is 0.03, a sub 13 is 0.63, e sub 22 is 0.67, and d sub 11 is 0.75.
* Adolescence_a: a sub 11 is 0.73, a sub 33 is 0.54, e sub 22 is 0.67, and e sub 33 is 0.36.
* Adulthood_a: a sub 11 is 0.43, a sub 22 is 0.29, e sub 22 is 0.67, and e sub 33 is 0.73.
* Adulthood_b: a sub 11 is 0.54, a sub 33 is 0.53, e sub 22 is 0.53, and e sub 33 is 0.47.
Values in parentheses represent 95 percent confidence intervals. For example, Childhood_a d sub 11 is 0.75 with an interval of 0.72 to 0.77. Parameters a sub 11 through a sub 33 represent genetic factor loadings, e sub 11 through e sub 33 represent non-shared environmental factors, and d sub 11 represents dominant genetic effects on A D H D.
Note: In childhood and adolescence, the best-fitting models are Cholesky decomposition models (Childhood_a and Adolescence_a). In adulthood, the best-fitting models are either Cholesky decomposition models (Adulthood_a) or independent pathway models (Adulthood_b).
d 11 represents the dominant genetic effect on ADHD.
Path2–3 represents the standardized path from binge eating symptoms to BMI.
If the best-fitting model is the Cholesky decomposition model, a 11 represents the first genetic factor (A1) loading on ADHD, a 12 represents the first genetic factor (A1) loading on binge eating symptoms, a 13 represents the first genetic factor (A1) loading on BMI, a 22 represents the second genetic factor (A2) loading on binge eating symptoms, a 23 represents the second genetic factor (A2) loading on BMI, and a 33 represents the third genetic factor (A3) loading on BMI; e 11 represents the first non-shared environmental factor (E1) loading on ADHD, e 12 represents the first non-shared environmental factor (E1) loading on binge eating symptoms, e 13 represents the first non-shared environmental factor (E1) loading on BMI, e 22 represents the second non-shared environmental factor (E2) loading on binge eating symptoms, e 23 represents the second non-shared environmental factor (E2) loading on BMI, and e 33 represents the third non-shared environmental factor (E3) loading on BMI (see Figure 2a).
If the best-fitting model is the independent pathway model, a 11 represents the additive genetic effect on inattention, a 22 represents the additive genetic effect on binge eating symptoms, a 33 represents the additive genetic effect on BMI, e 11 represents the non-shared environmental effect on inattention, e 22 represents the non-shared environmental effect on binge eating symptoms, and e 33 represents the non-shared environmental effect on BMI (see Figure 2b).
In adulthood, the Cholesky decomposition model and the independent pathway model with AICs of 215,429 demonstrated the best fit to the data (see Tables 1 and 3) . Specifically, in the Cholesky decomposition model, a common genetic factor (A1) significantly influenced ADHD (a 11 = 0.43, 95% CI: 0.27–0.58), binge eating symptoms (a 12 = 0.03, 95% CI: 0.02–0.06), and BMI (a 13 = 0.01, 95% CI: 0.001–0.03). In addition, another common genetic factor (A2) was identified, representing the residual shared genetic effects between binge eating symptoms (a 22 = 0.29, 95% CI: 0.24–0.33) and BMI (a 23 = 0.01, 95% CI: 0.00–0.02) after accounting for A1. Finally, there was also a significant genetic effect (A3) that influenced only BMI (a 33 = 0.26, 95% CI: 0.20–0.31). Regarding environmental influences, it accounted for 26% (95% CI: 0.23–0.28), 67% (95% CI: 0.63–0.71), and 73% (95% CI: 0.67–0.79) of the variance in ADHD, binge eating symptoms, and BMI, respectively. While the common non-shared environmental influences among the traits were small and nonsignificant. In the independent pathway model, common genetic and environmental factors influenced all three traits. The common genetic factor (A) explained 21% (95% CI: 0.15–0.27), 32% (95% CI: 0.23–0.41), and 25% (95% CI: 0.20–0.31) of the variance in inattention, binge eating symptoms, and BMI, respectively. The common non-shared environmental factor (C) accounted for 14% (95% CI: 0.13–0.15), 67% (95% CI: 0.63–0.71), and 37% (95% CI: 0.33–0.41) of the variance in ADHD, binge eating symptoms, and BMI. Overall, the best adulthood models, after accounting for shared genetic and environmental influences, included no paths between the phenotypes.
Models of ADHD subdomains
Phenotypic mediation models among ADHD subdomains, binge eating symptoms, and BMI showed significant mediating effects in line with results for the overall ADHD measure. However, in the biometric models, no mediating effects were found. The associations between ADHD subdomains and binge eating symptoms, as well as between ADHD subdomains and BMI, were primarily explained by common genetic factors. For details, please refer to the Supplementary Materials (see Supplementary Texts S1–S2; Supplementary Tables S8–S18).
Discussion
This study is the first to explore how ADHD symptoms at different developmental stages are associated with adult BMI, with a particular focus on the mediating role of binge eating symptoms. At the phenotypic level, our mediation model indicated that ADHD in adolescence and adulthood influenced subsequent BMI via binge eating symptoms, but this mediating effect was no longer significant after accounting for shared genetic factors. These results suggest that the observed phenotypic mediation effects might be largely due to shared genetic influences underlying these three traits.
In the phenotypic mediation models, significant direct effects from ADHD, hyperactivity–impulsivity, and inattention to binge eating symptoms were observed in adolescence and adulthood. This may be because, during childhood, diet is more strongly influenced by parents, whereas from adolescence onward, food choice transitions from primarily depending on caregivers to adolescents taking more responsibility for food acquisition and preparation (Neufeld et al., Reference Neufeld, Andrade, Suleiman, Barker, Beal, Blum, Demmler, Dogra, Hardy-Johnson, Lahiri, Larson, Roberto, Rodríguez-Ramírez, Sethi, Shamah-Levy, Strömmer, Tumilowicz, Weller and Zou2022). Therefore, fewer instances of binge eating symptoms associated with ADHD symptoms are observed during childhood, with such associations becoming more apparent only after adolescence begins. In addition, except for inattention in adulthood, the direct effects of ADHD and ADHD subdomains on BMI were significant across developmental stages. These findings are consistent with previous studies suggesting that individuals with ADHD may experience more difficulties with impulse inhibition, which has been linked to disordered eating behaviors and obesity (e.g. Davis et al., Reference Davis, Levitan, Smith, Tweed and Curtis2006; Nederkoorn et al., Reference Nederkoorn, Braet, Van Eijs, Tanghe and Jansen2006). Moreover, our results indicate that inattention, and not just hyperactivity–impulsivity, predicts binge eating symptoms, potentially reflecting deficits in attentional control that could also impair the regulation of eating behavior. Individuals with ADHD are more likely to be influenced by immediate reinforcement contingencies in their environment (Tripp & Wickens, Reference Tripp and Wickens2008), which is associated with the difficulty in sustaining attention. Binge eating symptoms can provide immediate reward, which may explain why individuals with higher levels of inattention symptoms are more likely to engage in binge eating symptoms as a form of immediate reinforcement. Furthermore, we observed that binge eating symptoms mediated the pathway from adolescence and early adulthood ADHD to BMI. This finding may be interpreted in a traditional mediation model as indicating that earlier ADHD symptoms continue to influence BMI at age 26 years through binge eating symptoms and behaviors in early adulthood. This highlights the longitudinal associations between ADHD symptoms, eating behaviors, and BMI across developmental stages.
In the biometric models, shared genetic risk (e.g. Capusan et al., Reference Capusan, Yao, Kuja-Halkola, Bulik, Thornton, Bendtsen, Marteinsdottir, Thorsell and Larsson2017; Cortese & Peñalver, Reference Cortese and Peñalver2010; Karhunen et al., Reference Karhunen, Bond, Zuber, Hurtig, Moilanen, Järvelin, Evangelou and Rodriguez2021) contributed to correlations between early ADHD symptoms and subsequent binge eating symptoms and BMI. This suggests that the mediation effects observed in the phenotypical models may in fact reflect confounding by shared genetic factors. Evidence from molecular genetics, behavioral, and neuroimaging research also supports the phenotypic associations and shared genetic influences identified in the current twin study. For example, a significant positive genetic correlation between ADHD and BMI was observed in a genome-wide association study using linkage disequilibrium score regression based on genome-wide association study data for ADHD (Demontis et al., Reference Demontis, Walters, Martin, Mattheisen, Als, Agerbo, Baldursson, Belliveau, Bybjerg-Grauholm, Bækvad-Hansen, Cerrato, Chambert, Churchhouse, Dumont, Eriksson, Gandal, Goldstein, Grasby, Grove and Neale2019). In addition, the first case–control genome-wide meta-analysis of binge eating identified six risk loci, including loci associated with higher BMI and impulsivity-related traits (Termorshuizen et al., Reference Termorshuizen, Davies, Lee, Dennis, Hübel, Johnson, Lu, Munn-Chernoff, Peters, Qi, Schaumberg, Signer, Singh, ter Kuile, Thornton, Xu, Yao, Yilmaz, Zhang and Coleman2025). Further evidence for shared mechanisms comes from pathway analyses. Specifically, enrichment analysis revealed that the Dopamine-DARPP32 Feedback in cAMP Signaling pathway, a component of the dopaminergic system, was enriched in ADHD-BMI results (Mota et al., Reference Mota, Poelmans, Klein, Torrico, Fernàndez-Castillo, Cormand, Reif, Franke and Arias Vásquez2020). In total, these molecular findings reveal an interconnected genetic architecture linking ADHD, binge eating symptoms, and BMI. Beyond genetics, behavioral and neuroimaging studies further support the existence of shared mechanisms linking these three traits. Specifically, research has shown that these three phenotypes share functional abnormalities in neural systems involved in reward processing, response inhibition, emotional processing, and emotion regulation (Seymour, Reinblatt, Benson, & Carnell, Reference Seymour, Reinblatt, Benson and Carnell2015). At the behavioral level, these shared neural mechanisms may be reflected in common characteristics, such as emotion dysregulation (El Archi et al., Reference El Archi, Cortese, Ballon, Réveillère, De Luca, Barrault and Brunault2020) and deficits in executive functioning (Ptacek et al., Reference Ptacek, Stefano, Weissenberger, Akotia, Raboch, Papezova, Domkarova, Stepankova and Goetz2016), which may contribute to the overlap among these phenotypes. Together, these findings provide new insights into the nature of the comorbidity among ADHD, binge eating symptoms, and elevated BMI, and may help inform the development of targeted prevention and intervention strategies.
The current study makes several important contributions. A key strength is the developmental perspective taken. We used the same sample from childhood to adulthood to examine how ADHD symptoms at different stages influence subsequent eating behaviors and BMI. Second, this study demonstrates that the mediating pathway, frequently reported in previous research, whereby early ADHD influences BMI through impaired impulse control over food, may in fact be confounded by genetic factors rather than representing a true symptom-level mediation.
Nevertheless, this study also has some limitations. For example, in many longitudinal studies, including the TEDS database, individuals with more severe early symptoms were more likely to drop out over time (Rimfeld et al., Reference Rimfeld, Malanchini, Spargo, Spickernell, Selzam, McMillan, Dale, Eley and Plomin2019; Teague et al., Reference Teague, Youssef, Macdonald, Sciberras, Shatte, Fuller-Tyszkiewicz, Greenwood, McIntosh, Olsson and Hutchinson2018), and this attrition may also have contributed to the absence of detectable mediation effects in the modified biometric mediation models. Moreover, the measure of eating behaviors used in the current study consisted of three items assessing binge eating symptoms. However, there may be other eating behaviors that link ADHD and BMI. A person with poor impulse control may be more prone to eat more than they need even if they do not binge eat, and such overeating without loss of control has also been associated with weight status (Neumark-Sztainer et al., Reference Neumark-Sztainer, Berge, Pisetsky, MacLehose, Loth and Goldschmidt2015). Therefore, future research could further examine how ADHD influences other types of eating behaviors apart from binge eating symptoms. In addition, the observed associations may have been confounded by other psychiatric disorders or treatments. For example, this study did not account for psychiatric or medical comorbidities associated with ADHD that may independently influence BMI. These comorbidities may have contributed to the observed associations between ADHD, binge eating symptoms, and BMI. Furthermore, the analyses did not consider the influence of ADHD diagnosis or pharmacological treatment. Pharmacological treatments for ADHD, such as methylphenidate, have well-documented effects on reducing appetite, body weight, and height growth (Carucci et al., Reference Carucci, Balia, Gagliano, Lampis, Buitelaar, Danckaerts, Dittmann, Garas, Hollis, Inglis, Konrad, Kovshoff, Liddle, McCarthy, Nagy, Panei, Romaniello, Usala and Wong2021; Zhang, Du, & Zhuang, Reference Zhang, Du and Zhuang2010). Individuals with more severe ADHD symptoms who receive pharmacological treatment may experience treatment-related changes in eating behaviors, which could confound the associations between ADHD symptoms and eating behaviors. Lastly, the current study relied on self-reported data. Self-reported BMI has been shown in previous studies to have systematic bias, with lower BMI values tending to be overestimated and higher BMI values tending to be underestimated (Stommel & Schoenborn, Reference Stommel and Schoenborn2009). This pattern, which may bias BMI estimates toward the mean, could also affect the detection of associations between ADHD, binge eating symptoms, and BMI. These factors should be considered in future studies.
Conclusion
This study investigated the associations between ADHD symptoms at different developmental stages, binge eating symptoms at age 21 years, and BMI at age 26 years. The results showed that in the phenotypic mediation model, adolescent and adult ADHD influenced subsequent BMI through the mediation of binge eating symptoms. However, this mediation effect was in fact attributable to shared genetic factors in biometric models. Thus, the associations between ADHD and binge eating symptoms, as well as between ADHD and BMI, are explained by shared genetic effects. This twin study reveals shared liability across ADHD, binge eating symptoms, and BMI. Intervention strategies could focus more on common underlying behavioral and self-regulatory mechanisms across these traits, as well as placing more emphasis on symptom patterns within families.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726105935.
Acknowledgments
The authors gratefully acknowledge the ongoing contribution of the TEDS participants and their families.
Funding statement
Y.Y. is supported by the China Scholarship Council; T.A.M. is supported by a Wellcome Trust Senior Research Fellowship awarded to T.A.M. (Grant 220382/Z/20/Z). TEDS is supported by a program grant (MR/V012878/1) to Professor Thalia Eley from the UK Medical Research Council (previously MR/M021475/1 awarded to Professor Robert Plomin).
Competing interests
The authors declare none.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Open access
