Social profiles among youth with attention-deficit/hyperactivity disorder (ADHD): Evidence from the ABCD study

Abstract

Social functioning difficulties among youth with attention-deficit/hyperactivity disorder (ADHD) have been examined behaviorally; however, limited research has investigated brain networks associated with social difficulties among youth with ADHD. A growing body of literature supports the utility of the NIMH’s Research Domain Criteria (RDoC) framework, which emphasizes broad neurobiological based dimensions, allowing for the integration of models of both neural circuitry and behavior when examining externalizing behaviors in youth. We hypothesized that an ADHD classification system based on social functioning would better predict real-world psychosocial and academic outcomes compared to traditional Diagnostic and Statistical Manual of Mental Disorders (DSM-5) nosology of ADHD presentations. First, using data from the Adolescent Brain Cognitive Development (ABCD) Study, we identified four distinct profiles of youth with ADHD ranging from low social functioning to high social functioning. These social-data-derived profiles were linked to differential social challenges associated with caregiver income and mental health disorders. Next, our neuroimaging findings initially revealed differential patterns of functional connectivity across profiles involving attention-control, cingulo-opercular, sensorimotor networks. However, these connectivity differences were not consistently replicated, indicating that social functioning alone may not define neurobiologically distinct subgroups. Finally, in comparing our social functioning profiles to existing DSM-5 nosology with respect to real-world psychosocial outcomes, our social profiles demonstrated greater explanatory power for outcomes related to peer relationships, family conflict, and mental health. Overall, these findings emphasize the heterogeneity in social functioning among ADHD youth and suggest that while behavioral profiles are clinically meaningful, future work should integrate additional dimensions, such as executive functioning, to more precisely capture the neurobiological underpinnings of ADHD.

1. Introduction

Attention-deficit/hyperactivity disorder (ADHD) is the most commonly diagnosed neurodevelopmental disorder among children (Polanczyk et al., 2015). Extensive literature has called attention to the disorder’s inherent complexity and heterogeneity given its numerous etiological risk factors, variation in clinical presentation and associated behaviors, long-term outcomes, and comorbidities (Castellanos et al., 2006; Costa Dias et al., 2015; Luo et al., 2019). Current nosology has been ineffective in teasing apart these differences, limiting the understanding of ADHD etiology, clinical prediction, and treatment development. The Research Domain Criteria (RDoC) framework, developed by the National Institute of Mental Health (NIMH), emphasizes broad biological-based dimensions, allowing for a more comprehensive approach in the classification of psychopathology (Insel et al., 2010). Recent work has demonstrated the overriding need to take an RDoC-informed approach when examining externalizing behaviors in youth, allowing for the integration of models of both neural circuitry and behavior (Beauchaine and Constantino, 2017; Beauchaine and Hinshaw, 2020).

ADHD has been conceptualized as a disorder of behavior and executive function, and more recently emotional functioning; thus, prior studies examining heterogeneity have focused on characterizing subgroups of ADHD youth based on these dimensions (Karalunas et al., 2014; Kofler et al., 2019), leaving other pivotal aspects of this disorder unexamined. Recent work highlights the importance of integrating social functioning into both the conceptualization and clinical care of ADHD (Aduen et al., 2018; Mikami et al., 2017). Youth with ADHD have been shown not only to exhibit impairments in social cognition (Parke et al., 2021), but also in real-world social behaviors; demonstrating reduced social competence (Deboo and Prins, 2007; Nixon, 2001), having fewer reciprocated friendships (Gardner and Gerdes, 2015), and engaging in less varied social activities (Carpenter Rich et al., 2009). Importantly, 50–70 % of youth with ADHD experience peer relationship difficulties (Bagwell et al., 2001; Gardner and Gerdes, 2015; Heiman, 2005; Uekermann et al., 2010), and these social difficulties have been described as stemming from two domains. First, in their social interactions with others, youth with ADHD demonstrate both elevations in negative social behaviors (i.e., failing to wait their turn, poor sportsmanship) and an absence of positive social behaviors (i.e., prosocial skills, empathy). These social behaviors affect the formation of social bonds (Mikami et al., 2019), as captured by RDoC’s social processes construct of Affiliation and Attachment. Second, youth with ADHD display difficulties in the exchange of socially relevant information, including impairments in pragmatic use of language (i.e., following a conversation or modifying communication with others) and identifying emotions in others (Gardner and Gerdes, 2015; Mikami et al., 2019; Uekermann et al., 2010), as captured by RDoC’s social processes construct of Social Communication. In a recent neuroimaging meta-analysis, we identified convergence across social task activation patterns that correspond to the RDoC constructs of Affiliation and Attachment, Social Communication, Perception and Understanding of Self, and Perception and Understanding of Others. These results provided evidence for strong convergence for Affiliation and Attachment in the default mode network (DMN) and cingulo-opercular network (CON), and Social Communication in the dorsal attention network (DAN) and visual network (VN) (Pintos Lobo et al., 2023). Additionally, ALE meta-analyses for both RDoC constructs demonstrated convergent activation in brain regions commonly associated with the frontoparietal network (FPN).

Limited prior work has aimed to examine distinct brain networks involved in social functioning among youth with ADHD, instead focusing on disrupted cognitive abilities (i.e., executive function, sustained attention) or disrupted affective processes in both negative (e.g., irritability; (Karalunas et al., 2014) and positive (e.g., reward processing) valences (Oldehinkel et al., 2016; Zepf et al., 2019), leaving pivotal neural processes related to social functioning unexamined. The exploration of social functioning as a stand-alone domain may provide critical new insight to understanding both the conceptualization and clinical care of ADHD. Doing so may serve as a feasible first step towards future work examining a more naturalistic viewpoint integrating both social and cognitive areas within the heterogeneity of ADHD youth. Notably, this population has also been linked to disruptions in general psychosocial and academic outcomes, including difficulties forming peer relationships, elevated family conflict and mental health concerns, and lower academic performance. For instance, familial and interparental relationships are at risk for higher rates of conflict in families of children with ADHD (Muñoz-Silva et al., 2017; Weyers et al., 2019). A significant percentage of youth with ADHD meet criteria for co-occurring mental health concerns including oppositional defiant disorder (ODD), conduct disorder (CD), as well as anxiety and depressive disorders (Becker et al., 2012). Additionally, youth with ADHD are at higher risk for academic underachievement, grade retention, and school dropout (DuPaul et al., 2019; Frazier et al., 2007), as well as overall academic performance difficulties due to lower rates of on-task behavior (Kofler et al., 2008). It is evident that ADHD is a heterogeneous disorder, with diverse comorbidities, clinical profiles, and long-term trajectories (Luo et al., 2019). The reality of the impact of ADHD on psychosocial and academic outcomes alludes to the ineffectiveness of current Diagnostic and Statistical Manual of Mental Disorders (DSM-5) nosology in capturing this heterogeneity. While DSM-5 currently serves as the current standard in identifying diagnostic groups, including ADHD, it fails to consider heterogeneous subtypes of ADHD. Thus, it is imperative that a novel and robust data-driven method be proposed that has the power of better predicting psychosocial and academic functioning in youth with ADHD. Moreover, a revised data-driven categorization system based on RDoC, which employs biological domains of functioning to characterize social-related subtypes of ADHD, will help disentangle ADHD heterogeneity above and beyond what current DSM-5 nosology is able to explain.

Building upon this work, the objective of the current study was to identify profiles of social functioning among youth with ADHD and examine the neurobiological validity of these profiles via an RDoC framework. To do so, we analyzed data from the Adolescent Brain Cognitive Development (ABCD) Study® (https://abcdstudy.org; (Volkow et al., 2018), a population-based study of approximately 12,000 children across 21 sites in the U.S. (Garavan et al., 2018). We aimed to (1) identify social-data-derived profiles of youth with ADHD based on heterogeneity in social behavioral processes, (2) determine whether these profiles exhibit significant resting-state functional connectivity (rsFC) differences in RDoC-defined neural circuits, thus demonstrating the neurobiological validity of social profiles, and (3) compare social-data-derived profiles to traditional DSM-5 nosology (i.e., presentation, comorbidity) with respect to clinical utility and concurrent validity of real-world psychosocial and academic outcomes.

Considering previous findings, we expected that at least three profiles of youth with ADHD would be identified, reflecting impairment in the RDoC constructs of: Affiliation and Attachment, Social Communication, and a third profile displaying low/no social difficulties compared to typically developing (TD) youth. Of note, no specific hypothesis was made for RDoC constructs Perception and Understanding of Self and Perception and Understanding of Others due to insufficient information of the role these constructs play in ADHD youth with social functioning difficulties. Next, in evaluating the neurobiological validity of these social profiles, and based on our recent meta-analytic findings (Pintos Lobo et al., 2023), we anticipated that the profile characterized by social functioning difficulties related to Affiliation and Attachment would exhibit disrupted connectivity in the DMN, CON, and FPN, whereas the profile with social difficulties related to Social Communication would exhibit disrupted connectivity in the DAN, VN, and FPN, and the low/no social functioning difficulties profile would exhibit impairment related to attention and executive function within the FPN and ventral attention network (VAN) when compared to TD youth. Finally, we expected that these social-data-derived profiles would better predict psychosocial functioning and academic performance compared to existing DSM-5 nosology, as these profiles would better account for heterogeneity among youth with ADHD. This work demonstrates the importance of detecting and classifying subtypes of ADHD youth through the use of multi-level information, including behavioral and brain-based measures. Overall, the knowledge gained will help advance understanding of ADHD etiology and serve as a step towards disentangling the heterogeneity of ADHD, particularly in terms of social functioning.

2. Methods and materials

Fig. 1 provides an overview of our methodological approach. Our objectives were to select ADHD and TD youth from the ABCD dataset (Step 1), extract ABCD items related to social functioning from the baseline protocol (Step 2), annotate the items according to RDoC’s social processes domain constructs and real-world psychosocial and academic outcomes (Step 3), identify ADHD profiles with RDoC-based social difficulties and determine whether they exhibit significant rsFC differences in RDoC-defined neural circuits (Step 4), and compare social-data-derived profiles to DSM-5 nosology with respect to psychosocial and academic outcomes (Step 5). Herein, we describe the work carried out to test our overall hypothesis that an ADHD classification system based on social functioning better predicts real-world psychosocial and academic outcomes than existing DSM-5 nosology.

Fig. 1.

Analytic Workflow. Participants were selected from the ongoing ABCD Study via items from a behavioral measure (i.e., KSADS-5) to identify ADHD and TD groups (Step 1). Select items related to social functioning were extracted from behavioral questionnaires including the KSADS-5, CBCL, Parent Short-SRS, Youth- and Parent- prosocial behavior survey, and FES - family conflict subscale (Step 2). Social items were then annotated as pertaining to either RDoC social processes domain constructs (i.e., Affiliation & Attachment, Social Communication, Perception and Understanding of Self, Perception and Understanding of Others) or real-world psychosocial and academic outcomes (Step 3). A latent profile analysis was conducted to identify data-driven phenotypic profiles of ADHD youth with social functioning impairment. Then, profile differences in rsFC data were examined utilizing the Gordon parcellation network in relation to four social profiles of youth (Step 4). Finally, eight separate linear regression models were conducted to compare the clinical utility and concurrent validity of data-driven profiles to DSM-5 nosology on outcomes related to peer relationships, family conflict, mental health, and academic outcomes (Step 5).

2.1. Participants

Participants were selected from the ongoing ABCD Study, the largest long-term study of brain development and child health in the United States (Volkow et al., 2018). The ABCD Study recruited a population-based, demographically diverse sample (Compton et al., 2019), with approximately 12,000 youth enrolled. In the present study, we examined ABCD data acquired at baseline (i.e., youth ages 9–10 years old) to employ a well-powered characterization of social profiles among youth with ADHD and examine the neurobiological validity of these profiles, thus providing new insight into the heterogeneity among youth with ADHD and the etiology and neurobiology of social difficulties among these youth. The ABCD Study was approved by the Institutional Review Board (IRB) at each study site and centralized IRB approval was provided by the University of California San Diego. Information regarding recruitment and assessment procedures have been extensively reported elsewhere (Dick et al., 2021; Volkow et al., 2018). Data from the ABCD Study are made available by the NIMH Data Archive (NDA; https://nda.nih.gov) and the current study utilized data from the ABCD Curated Annual Release 5.1.

2.2. Measures

Youth and caregivers completed the Kiddie Schedule for Affective Disorders and Schizophrenia (KSADS) for DSM-5 (KSADS-5), a reliable and valid semi-structured interview assessing psychopathology diagnoses and symptoms in children (Kaufman et al., 2000). The KSADS-5 is a semi-structured interview used to diagnose affective and neurodevelopmental disorders. In addition to the five diagnostic supplements within this assessment, the KSADS-5 includes background items to obtain information about health, presenting complaints, psychiatric treatment, school functioning, hobbies, peer, and family relations. The Child Behavior Checklist (CBCL) (Achenbach et al., 2001), a component of the Achenbach System of Empirically Based Assessment (ASEBA), was completed by caregivers to detect internalizing and externalizing problems in youth. The Parent Short Social Responsiveness Scale (SRS) was administered to caregivers to examine the presence and severity of social impairment in youth (i.e., social awareness, social information processing, capacity for reciprocal social communication) (Reiersen et al., 2008). Youth- and caregiver-reported prosocial behavior and psychopathology, including positive and negative attributes, were assessed for youth via the Parent- and Youth- Prosocial Behavior Survey (from the Strengths and Difficulties Questionnaire; SDQ) (Goodman, 1997). Finally, caregivers and youth completed the Family Environment Scale (FES) - Family Conflict Subscale Modified from PhenX to assess the amount of openly expressed anger and conflict among family members as well as parent’s levels of acceptance and responsiveness (Lanz and Maino, 2014).

2.3. Sample selection

Youth- and caregiver-report behavioral measures were extracted from the ABCD dataset to identify youth with and without ADHD. Youth were identified as meeting criteria for ADHD if caregivers reported at least six symptoms of inattention and/or six symptoms of hyperactivity/impulsivity, as well as cross-situational impairment, on the KSADS-5. Youth were deemed TD comparison youth with three or fewer total ADHD symptoms present on the KSADS-5. A subset of 5631 youth with and without ADHD (1072 with ADHD; 51.8 % male; M_age=9.96 years) were included in the current study. Given the robust size of the ABCD dataset, and to examine the stability of our findings, the ADHD sample was then divided into two subsamples to enable split-half analysis (Karcher and Barch, 2021) in order to test hypotheses in one half of the dataset (“ADHD sample 1; n = 456”) and test for replication in the other half of the dataset (“ADHD sample 2; n = 437”). Youth with subthreshold ADHD were excluded from the study (i.e., endorsing fewer than six symptoms of inattention and/or fewer than six symptoms of hyperactivity/impulsivity on the KSADS-5). Importantly, DSM-5 criteria were utilized to identify youth meeting current diagnostic symptoms of ADHD. Additional RDoC-based criteria were implemented to further refine and identify subgroups of ADHD participants; these are described below.

2.4. Identify and annotate social items

A detailed approach described below was employed to select social items from behavioral scales (i.e., KSADS-5, CBCL, SRS, Youth- and Parent-Prosocial Behavior Survey from the SDQ, FES - Family Conflict Subscale) and subsequently annotate them as pertaining to either RDoC social processes domain constructs or real-world psychosocial and academic outcomes.

2.4.1. Social functioning via RDoC domains

Our objective was to identify social-data-derived profiles of youth with ADHD based on heterogeneity in social behavioral processes. To accomplish this, social items were identified and annotated based on the RDoC social processes domain. First, one study associate (RPL) identified ABCD items from the CBCL caregiver-report, parent short SRS, and youth- and parent-reported Prosocial Behavior Survey that related to social functioning in youth. Then, three study associates (RPL, ARL, EDM) manually annotated each social item with one or more of the constructs taken from the RDoC Social Processes domain (i.e., Affiliation and Attachment, Social Communication, Perception and Understanding of Self, and Perception and Understanding of Others). These classifications were based on what the social item was measuring within the behavioral scale and were thereafter annotated with the associated RDoC social processes subdomain. Then, a single associate (RPL) reviewed all annotations to ensure accuracy and consistency (Table 1). Any disagreements between associates were resolved following a conversation between study associates (RPL, ARL, EDM). Due to the complexity of social functioning and associated items, we allowed social items to be associated with more than one RDoC social subdomain. For example, a given item could be associated as only Affiliation and Attachment, or both Affiliation and Attachment and Perception and Understanding of Others.

Table 1.

RDoC annotation of social items.

Behavioral Scale	RDoC Construct
Child Behavior Checklist (Achenbach and Rescorla, 2001)
Clings to adults or too dependent	Affiliation & Attachment
Complains of loneliness	Affiliation & Attachment
Doesn’t get along with other kids	Affiliation & Attachment; Perception and Understanding of Others
Easily Jealous	Affiliation & Attachment; Perception and Understanding of Self; Perception and Understanding of Others
Feels others are out to get him/her	Perception and Understanding of Others
Gets teased a lot	Affiliation & Attachment
Not liked by other kids	Affiliation & Attachment
Prefers being with younger kids	Affiliation & Attachment
Speech problems	Social Communication
Parent Short Social Responsiveness Scale (Reiersen et al., 2008)
Would rather be alone than with others	Affiliation & Attachment
Is able to understand the meaning of other people’s tone of voice and facial expressions	Social Communication
Avoids eye contact or has unusual eye contact	Social Communication
Has difficulty making friends, even when trying his or her best	Affiliation & Attachment
Is regarded by other children as odd or weird	Affiliation & Attachment
Has trouble keeping up with the flow of normal conversation	Social Communication
Has difficulty relating to peers	Affiliation & Attachment; Perception and Understanding of Others
Youth Prosocial Behavior Survey (Rothenberger and Woerner, 2004)
I try to be nice to other people. I care about their feelings.	Affiliation & Attachment; Perception and Understanding of Self
I am helpful if someone is hurt, upset, or feeling sick.	Affiliation & Attachment; Perception and Understanding of Self
I often offer to help others (parents, teachers, children).	Affiliation & Attachment; Perception and Understanding of Self
Parent Prosocial Behavior Survey (Rothenberger and Woerner, 2004)
My child is… considerate of other people’s feelings.	Affiliation & Attachment
My child is… helpful if someone is hurt, upset, or feeling ill.	Affiliation & Attachment
My child is… often offers to help others (parents, teachers, other children).	Affiliation & Attachment

Note. Annotations of ABCD social items into four Research Domain Criteria (RDoC) Social Processes Constructs: Affiliation and Attachment, Social Communication; Perception and Understanding of Self, and Perception and Understanding of Others. Behavioral scale and citation included for Child Behavior Checklist, Social Problems Scale; Parent Short Social Responsiveness Scale; Youth Prosocial Behavior Survey (from the Strengths and Difficulties Questionnaire); Parent Prosocial Behavior Survey (from the Strengths and Difficulties Questionnaire);

2.4.2. Psychosocial and academic outcomes

Next, we aimed to compare social-data-derived profiles to traditional DSM-5 nosology with respect to clinical utility and concurrent validity of real-world psychosocial and academic outcomes. As described above, one study associate (RPL) identified items from the KSADS-5 background items related to peer relationships and academic achievement, and the youth- and caregiver-reported FES family conflict subscale report to assess familial conflict. These items were extracted to be used as markers of real-world psychosocial outcomes related to peer, family, mental health, and academic functioning. Then, three study associates (RPL, ARL, EDM) manually annotated additional items as related to peer relationships, family conflict, and academic performance. These classifications were based on what the social item was measuring within the behavioral scale and were therefore annotated with one of four real-world psychosocial outcomes. Then, a single associate (RPL) reviewed all annotations to ensure accuracy and consistency (Table 2). Any disagreements between associates were resolved following a conversation between study associates (RPL, ARL, EDM).

Table 2.

Annotations related to Psychosocial and Academic Outcomes.

Behavioral Scale	Outcome Variable
KSADS−5 (Kaufman and Schweder, 2004)
Does your child have a best friend?	Peer Relationships
How long has your child been friends with this best friend?	Peer Relationships
Does your child have a regular group of kids he or she hangs out with at school or in your neighborhood?	Peer Relationships
How long has your child hung out with them?	Peer Relationships
Do you like your child’s friends?	Peer Relationships
Does your child have any problems with bullying at school or in your neighborhood?	Peer Relationships
In general, how does your child do in school?	Academic Performance
What kind of grades does your child get on average?	Academic Performance
In the past year or past several months, has there been a drop in your child's grades?	Academic Performance
Does your child receive special services at school?	Academic Performance
Parent Family Environment Scale (Moos and Moos, 1994)
We fight a lot in our family	Family Conflict
Family members rarely become openly angry	Family Conflict
Family member sometimes get so angry they throw things	Family Conflict
Family members hardly ever lose their tempers	Family Conflict
Family members often criticize each other	Family Conflict
Family members sometimes hit each other	Family Conflict
If there is a disagreement in our family, we try hard to smooth things over and keep the peace	Family Conflict
Family members often try to one-up or outdo each other	Family Conflict
In our family, we believe you don’t ever get anywhere by raising your voice	Family Conflict
Youth Family Environment Scale (Moos and Moos, 1994)
We fight a lot in our family	Family Conflict
Family members rarely become openly angry	Family Conflict
Family member sometimes get so angry they throw things	Family Conflict
Family members hardly ever lose their tempers	Family Conflict
Family members often criticize each other	Family Conflict
Family members sometimes hit each other	Family Conflict
If there is a disagreement in our family, we try hard to smooth things over and keep the peace	Family Conflict
Family members often try to one-up or outdo each other	Family Conflict
In our family, we believe you don’t ever get anywhere by raising your voice	Family Conflict
Parent Child Behavior Checklist (Achenbach and Rescorla, 2001)
Internalizing Problems (includes Withdrawn, Somatic Complaints, and Anxiety/Depressed Problems)	Mental Health
Externalizing Problems (includes Delinquent and Aggressive Behaviors)	Mental Health

Note. Annotations of ABCD social items into four outcomes: Peer relationships, Family conflict, Mental health, and Academic performance. Behavioral scale and citation included for Kiddie Schedule for Affective Disorders and Schizophrenia (KSADS-5); Parent Family Environment Scale - Family Conflict Subscale Modified from PhenX (FES); Parent Child Behavior Checklist (CBCL); Youth Family Environment Scale - Family Conflict Subscale Modified from PhenX (FES).

2.5. Neuroimaging data

2.5.1. Data acquisition

Youth participants completed a baseline neuroimaging protocol that included structural magnetic resonance imaging (MRI), as well as resting-state functional MRI (fMRI) using high spatial and temporal resolution simultaneous multislice/multiband echo-planar (EPI) (Hagler et al., 2019). For Siemens scanners, fMRI scan parameters were 90 × 90 matrix, 60 slices, field of view = 216 × 216, echo time/repetition time = 30/800ms, flip angle = 52°, 2.4 mm isotropic resolution, and slice acceleration factor 6. The complete protocols for all vendors and sequences are provided by Casey and colleagues (Casey et al., 2018). Herein, participants were scanned while they completed four 5-minute resting-state BOLD fMRI scans with their eyes open and fixated on a crosshair.

2.5.2. Preprocessing

Imaging data processing was performed by the ABCD Data Analysis, Informatics, and Resource Center (DAIRC; (Hagler et al., 2019). Herein, we used the variables made available by the ABCD DAIRC through public sharing of preprocessed imaging data in partnership with the NDA. Measures of resting-state functional connectivity (rsFC) are based on a seed-based correlational approach adapted for cortical surface-based analysis. Processed rsFC data include the tabulated results of region of interest (ROI)-based analyses.

Following pre-analysis processing steps (i.e., removal of initial frames, normalization, regression, temporal filtering, and calculation of ROI-average time course), imaging data were preprocessed through surface sampling and ROI averaging. Then, correlation values were calculated for each pair of ROIs which were Fisher transformed to normally distributed z-values and averaged within or between networks to provide summary measures of network correlation strength. Fisher Z-transformed averages of all pairwise correlations within each of the 13 Gordon networks (Gordon et al., 2016) were made available. The Gordon networks include Auditory (AUD), Cingulo-Opercular (CON), Cingulo-Parietal (CPAR), Default Mode (DMN), Dorsal Attention (DAN), Fronto-Parietal (FPN), None (also referred to as “Unassigned” network), Retrosplenial-Temporal, Salience, Sensorimotor-Hand (SMH), Sensorimotor-Mouth (SMM), Ventral Attention, and Visual.

2.6. Analyses

2.6.1. Latent profile analysis

Latent profile analysis (LPA) is a statistical modeling technique that allows researchers to estimate distinct profiles, or subpopulations, based on participants’ pattern of responses across variables. LPA was applied to items in Table 1 (variables = 22 items) to identify data-driven phenotypic profiles of youth with ADHD (n = 1072) reporting social functioning difficulties and subsequently will be mapped onto RDoC’s social domain constructs. LPA was performed in R using the tidyLPA package (Rosenberg et al., 2018). Additionally, mclust was used for model-based clustering, classification, and density estimation via the Expectation-Maximization (EM) algorithm, an approach for maximum likelihood estimation (Scrucca et al., 2016). Model fit statistics, including Akaike’s Information Criteria (AIC) and Bayesian Information Criteria (BIC), were computed to compare the goodness of fit of models with different numbers of latent profiles, thereby determining the number of clusters. The model with the lowest AIC/BIC values was chosen as the optimal number of clusters. Next, metrics such as entropy and posterior probabilities were applied to measure the certainty in the classification of individuals into the identified profiles of youth and examine the likelihood of class membership across the extracted profiles.

Group differences in demographic and diagnostic variables were examined using ANOVA and Chi-square tests among derived profiles of ADHD youth, as well as the TD group. Correlations among demographic/diagnostic variables with dependent variables of interest were examined. The Benjamini-Hochberg correction was applied to control for the false discovery rate (FDR) at 0.05 due to multiple comparisons (Benjamini and Hochberg, 1995). Relevant standard covariates were included, such as age, sex, race/ethnicity, pubertal status, caregiver education, household income, neighborhood disadvantage, family, site, medication use, and comorbidity of anxiety disorder(s), conduct disorder (CD), and/or oppositional defiant disorder (ODD), as well as those deemed appropriate following correlation analyses (Heeringa and Berglund, 2020). Differences among data-driven ADHD profiles were examined, as well as differences between ADHD profiles and the TD group.

2.6.2. Resting state functional connectivity (rsFC) analyses

Resting-state functional connectivity (rsFC) data from the ABCD dataset was analyzed to examine neurobiological validity of social-data-derived ADHD profiles. Differences in rsFC were examined across data-driven ADHD profiles compared to TD controls. Herein, we examined both the 13 within-network correlations (e.g., DMN-DMN) and the 13 × 13 between-network correlation matrix for the Gordon parcels (e.g., FPN-DMN; DMN-CON) in relation to ADHD and TD groups of youth with respect to social functioning difficulties. Linear mixed-effect models (LMM) were conducted in accordance to ABCD Study team recommendations (Dick et al., 2021, Hagler et al., 2019, Heeringa and Berglund, 2020) to determine whether there were significant within- and between-group differences with respect to the 13 within-network correlations and 78 between-network correlations, respectively. Covariates (i.e., age, sex, race/ethnicity, pubertal status, caregiver education, household income, neighborhood disadvantage, family, site, medication use, and comorbidity of anxiety disorder(s), CD, and/or ODD), were included in all models as recommended for best practice (Heeringa and Berglund, 2020).

2.6.3. Comparison to DSM-5 nosology

Next, eight separate linear regression models were implemented to compare the clinical utility and concurrent validity of social-data-derived profiles to traditional DSM-5 nosology (i.e., presentations, comorbidity) as categorical variables with respect to psychosocial and academic outcomes (i.e., peer relationships, family conflict, mental health, academic performance) as continuous variables. First, data-driven ADHD profiles were the predictor variable regressed on the psychosocial and academic outcomes. Then, DSM-5 nosology, including presentations (i.e., ADHD Inattention, ADHD Hyperactivity/Impulsivity, ADHD Combined) and comorbidities (i.e., ODD, CD, anxiety disorder) was the predictor variable regressed on the psychosocial and academic outcomes. Of note, given that we did not know with certainty what the comorbidity pattern in the data would be, we included comorbidities that are known to be associated with ADHD youth, including ODD, CD, and anxiety disorder. However, given the inclusion and exclusion criteria proposed by the ABCD Study, as well as the age of the sample (9–10-year-olds), we expected that the comorbidities of greatest interest will be ODD and anxiety disorder. Finally, we compared the proportion of variance explained (using R² in outcomes) between data-driven profiles and the corresponding models including DSM-5 nosology to identify which grouping variable better predicts the likelihood of disrupted psychosocial and academic outcomes in youth with ADHD.

3. Results

3.1. Sample selection

A subset of 5624 youth with and without ADHD (1072 with ADHD; Inattentive presentation [n = 437], Hyperactive presentation [n = 128], Combined presentation [n = 507]) were included in the current study. Youth were identified as meeting criteria for ADHD if caregivers reported at least six symptoms of inattention and/or six symptoms of hyperactivity/impulsivity, as well as cross-situational impairment, on the KSADS-5. Youth were deemed TD comparison youth with three or fewer total ADHD symptoms present on the KSADS-5. Youth with subthreshold ADHD (n = 2198), as well as with insufficient reported data on symptoms and/or impairment (n = 4046) were excluded from the study. In comparing the overall sample of ABCD youth (n = 11,868) to the subset of ABCD youth with and without ADHD (n = 5624), there were no statistically significant differences with respect to sex, race, or ethnicity (all p > .05). However, differences in participant age were statistically significant (t(11,120) = -3.98, p = .001), such that the mean age of the overall sample was lower compared to the subset of youth with and without ADHD.

Missing data from variables of interest were handled via casewise deletion, resulting in a subset of 5016 youth with and without ADHD (942 with ADHD). To ensure reliability and robustness of our analyses, this sample was then randomly divided into two subsamples to enable split-half analysis in order to test hypotheses in one half of the dataset (i.e., ADHD Sample 1, n = 471; TD sample 1, n = 2037) and test for replication in the other half of the dataset (ADHD Sample 2, n = 471; TD sample 2, n = 2037). This method was employed in R utilizing a predetermined seed value for reproducibility purposes. To confirm comparability between the two subsamples, we tested for differences across key demographic and covariate variables. No statistically significant differences were observed between Sample 1 and Sample 2 on age, race, ethnicity, pubertal development, parent education, household income, or neighborhood disadvantage (all p’s > .05), supporting that the random split produced demographically similar groups. All reported analyses and results below reflect findings for ADHD Sample 1, with a comparison of Sample 1 vs. Sample 2 provided at the end of the Results. Additional replication details, analyses, and results for ADHD Sample 2 are provided in the Supplementary material.

3.2. Latent profile analysis

We computed the AIC and BIC values for each model with the expectation that the optimal model would correspond to the lowest values of these indices. However, AIC and BIC values indicated different solutions: the AIC favored a five-profile model, whereas the BIC favored a four-profile model; thus, an optimal solution could not be determined using these criteria alone. Given this, and consistent with established practices when model fit indices conflict (Nylund et al., 2007), additional fit statistics were consulted to guide model selection. Specifically, we considered the Bootstrapped Likelihood Ratio Test (BLRT), entropy, approximate weight of evidence (AWE), classification likelihood criterion (CLC), and Kullback information criterion (KIC), following procedures outlines by Akogul and Erisoglu (2017). As shown in Table 3, the BLRT indicated that the four-profile solution provided significantly better fit compared to all other profile solutions (BLRT value = 718.227, p < .001). Moreover, the four-profile solution demonstrated higher entropy, suggesting more reliable classification, and achieved a favorable balance across AWE, CLC, and KIC, collectively providing evidence for superior model fit. Based on this analytic hierarchy of evidence, the four-profile solution was identified as the best fitting model and was therefore advanced as the championed model.

Table 3.

Latent profile analysis model comparisons.

Number of profiles	AIC	AWE	BIC	Adj. BIC	CLC	KIC	Entropy	Smallest	BLRT_val	BLRT_p
2	18,656.52	19,546.38	18,934.90	18,934.90	18,524.42	18,726.52	0.9468229	0.9631843	1589.07043	0.0099
3	18,374.03	19,570.02	18,747.97	18,747.97	18,195.92	18,467.03	0.9450567	0.9669421	328.49118	0.0099
4	17,701.80	19,203.88	18,171.30	18,171.30	17,477.73	17,817.80	0.9620833	0.9272011	718.22742	0.0099
5	17,654.01	19,462.31	18,219.07	18,219.07	17,383.83	17,793.01	0.9114550	0.8550532	93.79601	0.0099

Note. Number of profiles = Number of LPA profiles estimated; Akaike information criterion (AIC); Approximate weight of evidence (AWE); Bayesian information criterion (BIC); Adj., adjusted; Classification likelihood criterion (CLC); Kullback information criterion (KIC); values for the Bootstrapped Likelihood Ratio Test (BLRT_val); p-value for the bootstrapped likelihood ratio test (BLRT_p).

To ensure clearly defined class membership, we restricted profile assignment to ADHD youth whose posterior probabilities were 0.70 or higher. Of the 471 total unique ADHD youth, 456 (96.81 %) had posterior probabilities greater than 0.70, and thus were included in the class membership. Fig. 2 illustrates the four profiles of social functioning among ADHD youth. The first profile represented 10.5 % of the sample (n = 48) and was characterized by significant levels of impairment across Affiliation and Attachment (e.g., difficulty making friends), Social Communication (e.g., difficulty understanding the meaning of other people’s tones of voice and facial expressions), and Perception and Understanding of Others (e.g., feels others are out to get them). This model was labeled “Low Social Functioning”. The second profile represented 15.4 % of the sample (n = 70) and was marked by moderate levels of impairment across Affiliation and Attachment, Social Communication, and Perception and Understanding of Others. This model was labeled “Moderate Social Functioning”. The third profile represented 20.4 % of the sample (n = 93) and was characterized by high levels of social functioning across all domains, with a particular strength in the Perception and Understanding of Others. This model was labeled “Highly Perceptive”. The fourth and final profile represented 53.7 % of the sample (n = 245) and was also marked by high levels of social functioning across all domains, with a particular strength in the domains of Affiliation and Attachment and Social Communication. This model was labeled “Highly Connected”.

Fig. 2.

Social Functioning Impairment Across Profiles. Latent profile analysis revealed four profiles of social functioning among ADHD youth in the ABCD Study. These four profiles show distinct levels of social functioning among the domains of Affiliation and Attachment, Social Communication, and Perception and Understanding of Others. Higher scores indicate higher levels of impairment among social domains. Profile 1 (Blue) corresponded to Low Social Functioning; Profile 2 (Green) corresponded to Moderate Social Functioning; Profile 3 (White) corresponded to Highly Perceptive, reflecting high overall social functioning with particular strengths in Perception and Understanding of Others; and Profile 4 (Red) corresponded to Highly Connected, reflecting high overall social functioning with particular strengths in Affiliation and Attachment and Social Communication. The number of dots for each group is due to jittering applied to avoid overlap in the visualization; this does not reflect an unequal distribution of participants across groups.

Table 4 denotes the mean estimates and standard errors for RDoC social processes domains (i.e., Affiliation and Attachment, Social Communication, and Perception and Understanding of Others) across profiles. Item-level estimates and standard errors were also calculated and included for social items included in the LPA. Table 5 illustrates demographic characteristics across the TD group, total ADHD sample, and each of the four social profiles of ADHD youth. There were no significant differences between profiles in terms of site (p = .4328), family (p = .3868), sex (p = .0529), race (p = .4583), ethnicity (p = .5327), pubertal status (p = .1184), caregiver education (p = .0639), neighborhood safety (p = .3148), medication use (p = .0669), and anxiety disorder present (p = .1234). However, there was a significant difference in terms of caregiver combined income (Cramer’s V = 0.209, p = .002), such that youth among the Highly Perceptive and Highly Connected profiles were characterized by a higher mean combined caregiver income (mean combined income in the $50,000-$74,999 range) than youth in the Low Social Functioning and Moderate Social Functioning profiles (mean combined income in the $35,000-$49,000 range). There was also a significant difference in terms of presence of conduct disorder (Cramer’s V = 0.128, p = .001) such that youth among the Low Social Functioning profile (31.2 %) had higher rates of conduct disorder compared to the Moderate Social Functioning profile (17.1 %), Highly Perceptive profile (12.9 %), and Highly Connected profile (7.3 %). Finally, there was a significant difference in terms of oppositional defiant disorder (Cramer’s V = 0.129, p = .001), such that youth among the Low Social Functioning profile (50.0 %) had higher rates of oppositional defiant disorder compared to the Moderate Social Functioning profile (30.0 %), Highly Perceptive (32.3 %), and Highly Connected profile (19.0 %).

Table 4.

Raw estimates and SE across profiles.

	Profiles
Social domains & items	Profile 1 Low Social Functioning 10.5 %	Profile 2 Moderate Social Functioning 15.4 %	Profile 3 Highly Perceptive 20.4 %	Profile 4 Highly Connected 53.7 %
	Estimate (SE)	Estimate (SE)	Estimate (SE)	Estimate (SE)
Affiliation & Attachment	0.903 (0.188)	0.677 (0.191)	−0.259 (0.088)	−0.279 (0.057)
Clings to adults or too dependent	0.493 (0.179)	0.406 (0.164)	−0.245 (0.082)	−0.125 (0.058)
Complains of loneliness	0.817 (0.189)	0.256 (0.176)	−0.203 (0.107)	−0.159 (0.066)
Gets teased a lot	0.782 (0.199)	0.928 (0.177)	−0.371 (0.068)	−0.289 (0.062)
Not liked by other kids	1.183 (0.185)	0.794 (0.196)	−0.368 (0.088)	−0.328 (0.059)
Prefers being with younger kids	0.577 (0.207)	0.567 (0.167)	−0.227 (0.089)	−0.196 (0.065)
Would rather be alone than with others	0.691 (0.143)	0.450 (0.189)	−0.075 (0.104)	−0.240 (0.059)
Has difficulty making friends, even when trying their best	1.344 (0.205)	1.010 (0.248)	−0.309 (0.077)	−0.446 (0.037)
Is regarded by other children as odd or weird	1.340 (0.200)	1.002 (0.211)	−0.278 (0.089)	−0.455 (0.049)
Considerate of other people’s feelings*	−1.135 (0.131)	0.283 (0.106)	−0.776 (0.083)	0.429 (0.043)
Helpful if someone is hurt, upset, or feeling ill*	−1.572 (0.121)	0.632 (0.001)	−1.364 (0.041)	0.632 (0.001)
Often offers to help others (parents, teachers, other children)*	−1.208 (0.163)	0.415 (0.107)	−0.729 (0.090)	0.387 (0.044)
Social Communication	0.828 (0.172)	0.506 (0.158)	−0.129 (0.103)	−0.263 (0.062)
Speech problem	0.381 (0.202)	0.325 (0.161)	−0.105 (0.096)	−0.131 (0.059)
Is able to understand the meaning of other people’s tone of voice and facial expressions	0.949 (0.138)	0.398 (0.161)	−0.102 (0.111)	−0.265 (0.068)
Avoids eye contact or has unusual eye contact	0.996 (0.168)	0.520 (0.159)	−0.083 (0.117)	−0.318 (0.062)
Has trouble keeping up with the flow of normal conversation	0.987 (0.180)	0.781 (0.153)	−0.227 (0.088)	−0.339 (0.060)
Others	0.763 (0.215)	0.423 (0.167)	−0.219 (0.088)	−0.192 (0.056)
Feels others are out to get them	0.763 (0.215)	0.423 (0.167)	−0.219 (0.088)	−0.192 (0.056)

Note. Mean Estimate, and SE (standard errors) across data-driven profiles. % = the percentage of the sample meeting criteria for the latent profile. * = Questions from the Prosocial Questionnaire were not included in the Mean estimate by RDoC domain.

Table 5.

Demographic characteristics for total sample and profiles.

			Profiles
Youth Characteristics	TD Sample N = 2037	Total ADHD Sample 1 N = 456	Profile 1 Low Social Functioning n = 48	Profile 2 Moderate Social Functioning n = 70	Profile 3 Highly Perceptive n = 93	Profile 4 Highly Connected n = 245	Statistics across ADHD profiles
Age (in months)	119.73 (7.33)	118.56 (7.39)	119.98 (7.98)	119.14 (7.33)	117.71 (7.33)	118.43 (7.32)	X2 = 74.538 p = .487
Sex
Female	1059 (51.98)	128 (28.07)	11 (22.92)	11 (15.71)	28 (30.11)	78 (31.84)	X2 = 7.939 p = .053
Male	978 (48.01)	328 (71.93)	37 (77.08)	59 (84.28)	65 (69.89)	167 (68.16)
Race
White	1629 (79.97)	369 (80.92)	41 (85.42)	54 (77.14)	80 (86.02)	194 (79.18)	X2 = 18.036 p = .458
Black/African American	242 (11.88)	68 (14.91)	6 (12.50)	14 (20.00)	12 (12.90)	36 (14.69)
Native American	16 (0.78)	1 (0.22)		1 (1.43)
Asian/Asian Indian	68 (3.33)	5 (1.09)		1 (1.43)		4 (1.63)
Other Race	77 (3.78)	13 (2.85)	1 (2.08)		1 (1.07)	11 (4.49)
Refuse to Answer	5 (0.25)
Ethnicity
Hispanic	339 (16.6)	67 (14.69)	10 (20.83)	11 (15.71)	11 (11.82)	35 (14.29)	X2 = 19.677 p = .533
Non-Hispanic	1696 (83.26)	389 (85.31)	38 (79.16)	59 (84.28)	82 (88.17)	210 (85.71)
Refuse to Answer	2 (0.09)
Pubertal Status
Prepuberty	1052 (51.64)	269 (58.99)	25 (52.08)	37 (52.86)	57 (61.29)	150 (61.22)	X2 = 18.532 p = .118
Early puberty	482 (23.66)	114 (25.00)	15 (31.25)	23 (32.86)	25 (26.88)	51 (20.82)
Mid puberty	467 (22.93)	69 (15.13)	6 (12.50)	10 (14.29)	11 (11.83)	42 (17.14)
Late puberty	35 (1.72)	3 (0.66)	2 (4.17)			1 (0.41)
Post puberty	1 (0.05)	1 (0.22)				1 (0.41)
Caregiver Education
< High school diploma	100 (4.91)	8	4 (8.33)	1 (1.4)	1 (1.07)	2 (0.82)	X2 = 52.88 p = .064
High School Diploma/GED	154 (7.56)	35	3 (6.25)	8 (11.43)	8 (8.60)	16 (6.53)
Some college	517 (25.38)	160 (35.09)	20 (46.51)	29 (41.42)	33 (35.48)	78 (31.84)
Bachelor’s Degree	680 (33.38)	135 (29.61)	12 (25.0)	18 (25.71)	28 (30.11)	77 (31.43)
Postgraduate Degree	585 (28.72)	118 (4.39)	9 (18.75)	14 (20.0)	23 (24.73)	72 (29.39)
Refuse to Answer	1 (0.05)
Caregiver Combined Income
<$5,000	53 (2.60)	13 (2.85)	7 (14.58)	3 (4.29)	2 (2.15)	1 (0.41)	X2 = 60.02 p < .01 Cohen’s d = .209
$5,000-$11,999	49 (2.41)	12 (2.63)	–	3 (4.29)	3 (3.23)	6 (2.45)
$12,000-$15,999	31 (1.52)	7 (1.54)	–	3 (4.29)	2 (2.15)	2 (0.82)
$16,000-$24,999	70 (3.44)	17 (3.73)	2 (4.17)	4 (5.71)	4 (4.30)	7 (2.86)
$25,000-$34,999	84 (4.12)	28 (6.14)	4 (8.33)	3 (4.29)	5 (5.38)	16 (6.53)
$35,000-$49,999	161 (7.90)	43 (9.43)	6 (12.50)	7 (10.00)	8 (8.60)	22 (8.98)
$50,000-$74,999	254 (12.47)	58 (12.72)	7 (14.58)	7 (10.00)	12 (12.90)	32 (13.06)
$75,000-$99,999	319 (15.66)	63 (13.82)	5 (10.42)	9 (12.86)	14 (15.05)	35 (14.29)
$100,000-$199,999	609 (29.89)	129 (28.29)	11 (22.92)	18 (25.71)	21 (22.58)	79 (32.24)
$200,000 and greater	263 (12.91)	51 (11.18)	1 (2.08)	5 (7.14)	19 (20.43)	26 (10.61)
Refuse to Answer	81 (3.98)	16 (3.51)	3 (6.25)	3 (4.29)	1 (1.07)	9 (3.67)
Don’t Know	63 (3.09)	19 (4.17)	2 (4.17)	5 (7.14)	2 (2.15)	10 (4.08)
Neighborhood Disadvantage	11.96 (2.73)	11.66 (2.82)	11.17 (2.79)	11.84 (2.65)	11.71 (3.05)	11.69 (2.79)	X2 = 39.676 p = .315
Medication Use, current	39 (1.91)	135 (29.61)	21 (43.75)	24 (34.29)	25 (26.88)	65 (26.53)	X2 = 7.03 p = .067
Comorbidity
Anxiety Disorder, present	8 (0.39)	29 (6.36)	5 (10.42)	8 (11.43)	4 (4.30)	12 (4.89)	X2 = 5.848 p = .123
Conduct Disorder, present	14 (0.69)	58 (12.72)	15 (31.25)	12 (17.14)	12 (12.90)	19 (7.75)	X2 = 22.634 p < .001 Cramer’s V = .128
Oppositional Disorder, present	42 (2.06)	122 (26.75)	24 (50.00)	21 (30.00)	30 (32.26)	47 (19.18)	X2 = 22.906 p < .001 Cramer’s V = .129

Note. Data are presented as mean (SD) or n (%). GED = General Equivalency Degree. Races not included due to zero frequency: Alaska Native, Guamanian, “don’t know race”.

3.3. Resting state functional connectivity (rsFC) results

Differences in rsFC were examined across data-driven ADHD profiles compared to TD controls via linear mixed-effects modeling (LMM) using the lmerTest package in R version 2023.6.1.524. The four LPA-driven ADHD profiles of social functioning and a TD control group were included as predictor variables, along with covariates of interest including age, sex, race, ethnicity, pubertal status, caregiver education, household income, neighborhood disadvantage, mean framewise displacement (FD), medication use, and comorbidity of anxiety disorder, conduct disorder, and oppositional defiant disorder. To address the well-documented challenges of head motion in neuroimaging research, particularly among youth with ADHD, and in recognition of differing recommendations in the literature (Couvy-Duchesne et al., 2015, Power et al., 2012, Satterthwaite et al., 2012, Thomson et al., 2024), two analytic approaches were implemented. First, consistent with emerging practices prioritizing sample retention, all participants were retained and mean framewise displacement (FD) was included as a covariate (primary analyses reported below). Second, a more conservative approach was conducted in which participants exceeding a framewise-displacement threshold of > 0.5 mm were excluded, with FD still included as a covariate.

First, analyses conducted on the full sample (Sample 1; N = 2493), which included TD controls (n = 2037), Low Social Functioning profile (n = 48), Moderate Social Functioning profile (n = 70), Highly Perceptive profile (n = 93), and Highly Connected profile (n = 245), revealed within- and between-network connectivity differences reported below. Second, we repeated the analyses after excluding participants with head motion exceeding the mean FD threshold (>0.5 mm) and including mean FD as a covariate. This exclusion criterion reduced the total sample size by approximately 21.6 %, resulting in a final analytic sample of N = 1954: TD controls (n = 1637), Low Social Functioning profile (n = 30), Moderate Social Functioning profile (n = 34), Highly Perceptive profile (n = 76), and Highly Connected profile (n = 177). In this second analysis, a more conservative analytic approach, one significant finding held in the between-network analyses (AUD and None network). All other significant findings did not survive FDR correction.

Additionally, to account for the hierarchical nature of the data, the model included random effects for site and for family to account for correlation among participant data and correlation in residuals caused by the nested structure of the ABCD data (Saragosa-Harris et al., 2021). LMMs analyzed the associations between social functioning profiles of ADHD and TD youth via (1) within-network connectivity for each of the 13 Gordon networks and (2) between-network connectivity with respect to the 78 between-network correlations. All results were false discovery rate- corrected (FDR-corrected) using the Benjamini-Hochberg procedure (Benjamini and Hochberg, 1995) to account for multiple comparisons. Model estimates quantify the magnitude and direction of the relationship between the social functioning profiles and the within-network or between-network connectivity of the 13 Gorden networks. These estimates represent the coefficients from the regression models that assess functional connectivity. Importantly, a negative estimate suggests a reduction in connectivity while a positive estimate suggests an increase in functional connectivity relative to TD controls. Additionally, the higher the absolute value of the estimate, the stronger the effect.

3.4. Within-network connectivity

Social functioning profiles of ADHD youth were examined in comparison to TD controls regarding within-network connectivity of the 13 Gordon networks, allowing for the assessment of the strength and coherence of functional connections within the same network. Supplementary Tables S10-S14 illustrate model estimates for within-network connectivity relationships to TD controls and each of the four social profiles of ADHD youth. Compared to TD controls, ADHD youth among the Low Social Functioning profile exhibited reduced within-network connectivity in the DAN (Est. = −0.547, SE = 0.177, t = -3.089, p = .002, FDR-corrected p = .046) and the SMH (Est. = −0.567, SE = 0.175, t = -3.236, p = .001, FDR-corrected p = .037; Fig. 3).

Fig. 3.

Standardized DAN and SMH Within-Network Connectivity. Estimates of standardized within-network connectivity (95 % confidence intervals) for the Dorsal Attention Network (DAN) and Sensorimotor-Hand Network (SMH) across social functioning profiles of ADHD youth compared to typically developing controls. TD = Typically Developing controls. Profile 1 = Low Social Functioning (blue); Profile 2 = Moderate Social Functioning (green); Profile 3 = Highly Perceptive (white/gray); Profile 4 = Highly Connected (red). Significant estimates are denoted with asterisks: **.

3.5. Between-network connectivity

Next, we examined whether social functioning profiles of ADHD youth compared to TD controls were associated with alterations in between-network connectivity. Supplementary Tables S15-S27 illustrate model estimates for between-network connectivity relationships to TD controls and each of the four social profiles of ADHD youth. In comparison to TD controls, ADHD youth with Low Social Functioning showed increased connectivity between the AUD and Salience networks (Est. = 0.582, SE = 0.176, t = 3.306, p = .001, FDR-corrected p = .037) compared to TD controls. Furthermore, ADHD youth among this profile exhibited decreased connectivity between the DAN and Salience (Est. = −0.579, SE = 0.178, t = -3.256, p = .001, FDR-corrected p = .037). ADHD youth among the Highly Connected profile exhibited increased connectivity between the AUD and None (Est. = 0.315, SE = 0.079, t = 3.971, p < .001, FDR-corrected p = .007), DMN and SMH (Est. = 0.301, SE = 0.079, t = 3.809, p < .001, FDR-corrected p = .009), and CON and None networks (Est. = 0.279, SE = 0.077, t = 3.613, p = .000, FDR-corrected p = .009) compared to TD controls (Fig. 4).

Fig. 4.

Standardized Between-Network Connectivity. Estimates of standardized between-network connectivity (95 % confidence intervals) across social functioning profiles of ADHD youth compared to typically developing controls. Panel A = Auditory and Salience; Panel B = Dorsal Attention and Salience; Panel C = Default Mode and Sensorimotor-Hand; Panel D = Auditory and None; Panel E = Cingulo-Opercular and None. TD = Typically Developing controls. Profile 1 = Low Social Functioning (blue); Profile 2 = Moderate Social Functioning (green); Profile 3 = Highly Perceptive (white/gray); Profile 4 = Highly Connected (red). Significant estimates are denoted with asterisks: **.

3.6. Comparison to DSM-5 Nosology

Eight separate linear regression models were constructed to compare the clinical utility and concurrent validity of social-data-derived profiles to traditional DSM-5 nosology (ADHD presentations; inattentive, hyperactivity, and combined) with respect to four outcomes: peer relationships, family conflict, mental health, and academic performance. Within each psychosocial outcome, two linear regression models were conducted. The first model examined the association between psychosocial outcomes and the social functioning profiles and included four dummy-coded social profiles as predictors. The second model examined the relationship between psychosocial outcomes and DSM-5 nosology and included three ADHD presentations as predictors. The coefficient of determination (R²) was utilized as the primary metric to evaluate the proportion of variance explained by each model. Statistical analysis was conducted using tidyverse and lmtest in R. Fig. 5 illustrates the comparison of proportion of variance explained between data-driven social profiles and traditional DSM-5 nosology to psychosocial and academic outcomes.

Fig. 5.

Comparison of proportion of variance explained between data-driven social profiles and traditional DSM-5 nosology to psychosocial and academic outcomes (Peer Relationships, Family Conflict, Mental Health, Academic Performance).

First, two separate linear regression models were constructed to investigate the association between peer relationships and (1) four social functioning profiles and (2) DSM-5 nosology. The first model aimed to assess the impact of profiles of social functioning on peer relationships. The results indicated that the overall model was statistically significant (F(3,452) = 3.387, p = 0.018, R² = 0.022, R²_adjusted = 0.015). The second model explored the relationship between peer relationships and ADHD presentations. The overall model was not statistically significant (F(2,453) = 0.114, p = 0.893, R² = 0.001, R²_adjusted = -0.004). Overall, social functioning profiles had a higher proportion of variance explained in peer relationships compared to DSM-5 nosology.

Next, two linear regression models were constructed with family conflict as the dependent variable. Results of the first model, examining the association between social functioning profiles and family conflict, was statistically significant (F(3,452) = 3.075, p = 0.027, R² = 0.020, R²_adjusted = 0.014). In the second model, where ADHD was the predictor variable, the overall model was not statistically significant (F(2,453) = 0.787, p = 0.456, R² = 0.003, R²_adjusted = -0.001). In comparing the proportion of variance explained, the model with social functioning profiles had a higher proportion of variance explained in family conflict outcomes compared to the DSM-5 model.

To assess the relationship between mental health as a dependent variable, two additional linear regression models were run. In the first model, with social functioning profiles as predictors, the overall model was statistically significant (F(3,452) = 37.45, p < .001, R² = 0.199, R²_adjusted = 0.194). The second model, assessing DSM-5 ADHD presentations, was also statistically significant (F(2,453) = 17.04, p < .001, R² = 0.069, R²_adjusted = 0.067). Overall, the model utilizing social functioning profiles had a higher proportion of variance explained in mental health outcomes compared to the DSM-5 model.

Finally, two linear regression models were executed to assess academic performance outcomes. The first model, examining the impact of social functioning profiles on academic performance, was statistically significant (F(3,447) = 3.062, p = 0.028, R² = 0.020, R²_adjusted = 0.014). The second model, exploring the relationship between academic performance and DSM-5 nosology, was also statistically significant (F(2,448) = 6.195, p = 0.002, R² = 0.027, R²_adjusted = 0.023). In comparing the proportion of variance explained, the model using ADHD presentations based on DSM-5 nosology had a higher proportion of variance explained in academic performance compared to the social functioning profiles.

3.7. Replication results

Given the robust size of the ABCD dataset, we conducted a split-half analysis by testing our hypotheses in one half of the dataset and testing for replication in the other half of the dataset, therefore not only examining the stability of our findings but also ensuring that our conclusions are scientifically rigorous. Overall, the replication test indicated that the consistency of results across ADHD Sample 1 and ADHD Sample 2 was somewhat mixed.

Consistent with ADHD Sample 1, profile assignment in ADHD Sample 2 was restricted to ADHD youth with posterior probabilities of 0.70 or higher, yielding a comparable inclusion rate of 92.8 % (n = 437). The four identified profiles of social functioning reflected a similar distribution: the Low Social Functioning profile constituted 10.5 % of the sample (n = 46), characterized by significant impairments across all RDoC domains; the Moderate Social Functioning profile represented 21.5 % of the sample (n = 94), showing moderate impairments across domains; the Highly Perceptive profile included 21.06 % of the sample (n = 92), marked by strengths in the Perception and Understanding of Others; and the Highly Connected profile encompassed 46.91 % of the sample (n = 205), characterized by strengths in Affiliation and Attachment and Social Communication. In analyzing demographic characteristics across both samples, no significant differences between profiles were found in terms of ABCD site, family, sex, race, ethnicity, pubertal status, caregiver education, neighborhood safety, and medication use. Within both ADHD samples, the Low Social Functioning profile had higher rates of oppositional defiant disorder compared to the other profiles (ADHD Sample 1 = 50 %; ADHD Sample 2 = 54.35 %). Additionally, across both samples, the Low Social Functioning profile demonstrated higher rates of the presence of conduct disorder compared to the other profiles, with the Highly Connected profile consistently exhibiting the lowest rates of conduct disorder across both samples. However, in ADHD Sample 1, findings highlighted a significant difference in combined caregiver income among the Highly Perception and Highly Connected profiles that was not apparent in ADHD Sample 2. Finally, ADHD Sample 2 found a significant difference in rates of anxiety disorder across profiles, particularly among the Low Social Functioning profile, while ADHD Sample 1 found no significant difference. Overall, Samples 1 and 2 identified similar socially derived profiles, with no demographic differences, but some variability in comorbidity.

Next, social functioning profiles of ADHD youth were examined in comparison to TD controls regarding both within- and between-network connectivity, allowing for the assessment of the strength and coherence of functional connections. In ADHD Sample 2, two between-network connectivity alterations emerged as statistically significant for the Highly Connected profile. Youth in the Highly Connected profile demonstrated increased connectivity between the CON and Ventral Attention networks (Est. = 0.288, SE = 0.085, t = 3.374, p = .001, FDR-corrected p = .034), as well as increased connectivity between the CON and DMN (Est. = 0.293, SE = 0.082, t = 3.557, p = .001, FDR-corrected p = .034). A complete list of statistically significant model estimates for between-network connectivity is provided in the Supplementary Material. Overall, the connectivity differences observed in Sample 1 were not replicated in Sample 2, suggesting that the socially derived profiles remained fairly heterogeneous in terms of within- and between-network functional connectivity.

Finally, in comparing the clinical utility and concurrent validity of social functioning profiles to traditional DSM-5 nosology, our comparison among ADHD Sample 1 and ADHD Sample 2 noted a robust replication of findings. Across both samples, social-functioning profiles explained more variance than DSM-5 nosology with respect to peer relationships, family conflict, and mental health, while the model using ADHD presentations based on DSM-5 nosology had a higher proportion of variance explained in academic performance. These findings further validate the importance in capturing disrupted psychosocial outcomes among ADHD youth above and beyond the traditional DSM-5 nosology, with the goal of providing a comprehensive understanding of the struggles ADHD youth face, particularly in the context of internalizing and externalizing mental health disorders. Additional replication details, analyses, and results for ADHD Sample 2 are provided in the Supplementary material.

4. Discussion

The current study benefits from a robust, well-powered sample drawn from the ongoing ABCD Study, the largest long-term investigation of brain development and child health in the United States, encompassing a diverse population of youth. Here, we explored the heterogeneity of ADHD through a multimodal approach, including behavioral and brain-based measures, to ultimately gain a more comprehensive understanding of ADHD heterogeneity and serve as a step towards unraveling the complexity of ADHD within the context of social functioning impairment. To this end, we utilized baseline data from the ABCD Study, examining social functioning among youth with and without ADHD. First, we identified social-data-derived profiles of youth with ADHD based on heterogeneity in social behavioral processes via latent profile analysis. LPA results revealed four distinct social functioning profiles among youth with ADHD, including (1) Low Social Functioning, (2) Moderate Social Functioning, (3) Highly Perceptive (reflecting high overall social functioning with particular strengths in Perception and Understanding of Others), and (4) Highly Connected (reflecting high overall social functioning with particular strengths in Affiliation and Attachment and Social Communication). These four profiles captured variations in social functioning, providing a nuanced understanding of the social challenges experienced by youth with ADHD. Second, we examined the neurobiological validity of these data-driven social profiles and identified rsFC differences across profiles compared to TD controls; however, these differences were not replicated in a second sample. Finally, we compared social-data-derived profiles to traditional DSM-5 nosology (i.e., presentation, comorbidity) with respect to clinical utility and concurrent validity of psychosocial and academic outcomes.

4.1. Latent profile analysis

As hypothesized, LPA results revealed profiles of youth with ADHD reflecting different levels of impairment among the RDoC constructs related to the Social Processes domain including Affiliation and Attachment, Social Communication, and Perception and Understanding of Others. The LPA recognized a four-profile solution as the model of best fit, creating separate profiles of social functioning. The Low Social Functioning profile (n = 48; 10.5 % of sample) was characterized by significant levels of impairment across the domains of Affiliation and Attachment (e.g., difficulty making friends), Social Communication (e.g., difficulty understanding the meaning of other people’s tones of voice and facial expressions), and Perception and Understanding of Others (e.g., feels others are out to get them). The Moderate Social Functioning profile (n = 70; 15.4 % of sample) was marked by moderate levels of impairment across the domains of Affiliation and Attachment, Social Communication, and Perception and Understanding of Others. The Highly Perceptive profile (n = 93; 20.4 % of sample) was characterized by high levels of social functioning across all domains, with a particular strength in Perception and Understanding of Others. Finally, the Highly Connected profile (n = 245; 53.7 % of sample), was marked by high levels of social functioning across all domains with a particular strength in the domains of Affiliation and Attachment and Social Communication. As expected, our results capture the wide spectrum of social functioning that may be present in youth with ADHD and tease apart the differences with respect to the severity of social challenges. In line with prior work examining social difficulties in ADHD youth, our findings suggest that a meaningful subset of ADHD youth experience moderate to severe social functioning difficulties, including peer rejection, having fewer friendships, and victimization. These findings help provide a more nuanced understanding of the heterogeneous social experiences among ADHD youth.

Comparison of these profiles across study covariates revealed significant differences in terms of caregiver combined income, such that youth among the Highly Perceptive and Highly Connected profile were overrepresented among higher combined caregiver income. Prior work (Cooper and Stewart, 2021, Piotrowska et al., 2023) examined the association between household income and youth outcomes (i.e., cognitive, social development, behavioral, and mental health) and found that household income has a positive causal effect on youth outcomes. Researchers suggest this association may be due to either the access to financial resources which enable caregivers to provide for their children, ensuring a better quality of life (i.e., investment theory) or the impact of economic disadvantage on the home environment that may in turn affect parenting abilities (i.e., family stress model), or a combination of both. This is in line with our findings regarding social functioning difficulties, where higher combined caregiver income was associated with decreased social functioning impairments in youth. Additionally, comparison across profiles demonstrated differences in the presence of conduct disorder (CD) and oppositional defiant disorder (ODD), such that youth among the Low Social Functioning profile were overrepresented among those having comorbidities (CD, ODD) compared to the other profiles. Prior literature examining social functioning in ADHD youth have highlighted the difficulties that youth experience in the context of Affiliation and Attachment (e.g., poor social bonds), Social Communication (e.g., difficulty with responding to social cues, impairment in pragmatic use of language), and Perception and Understanding of Others (e.g., difficulties with perspective taking). As a result of a marked impairment in these social domains, ADHD youth often experience negative outcomes, including conflicting peer relationships, leading to higher rates of social isolation and, ultimately, are at greater risk of developing adverse problems, including aggression, ODD, and CD (Pardini and Fite, 2010, Cordier et al., 2018). In accordance with prior literature (Kim et al., 2023, Noordermeer et al., 2016, Evans et al., 2020), ADHD youth exhibit higher levels of social impairment, particularly in the context of comorbidities including ODD and CD, linking this finding to higher rates of peer rejection, victimization, and overall poor social functioning. Notably, there were no significant differences between LPA-driven profiles in terms of all other covariates (i.e., sex, race, ethnicity, pubertal status, caregiver education, neighborhood safety, medication use, and anxiety disorder as a comorbidity).

4.2. Comparison to DSM-5 Nosology

Finally, linear regression models were conducted to compare the clinical utility and validity of social-data-derived profiles with traditional DSM-5 nosology (i.e., ADHD presentations: inattentive, hyperactivity, combined) with respect to four outcomes: peer relationships, family conflict, mental health, and academic performance. When analyzing the association between social functioning profiles and peer relationships, it was found that social functioning profiles explained a higher proportion of variance, compared to DSM-5 nosology, in the outcomes related to peer relationships, family conflict, and mental health outcomes. However, DSM-5 nosology explained a higher proportion of variance in academic performance outcomes, compared to social functioning profiles.

Overall, our data-driven social profiles were able to capture disrupted psychosocial outcomes among ADHD youth above and beyond the traditional DSM-5 nosology. These findings suggest a reconsideration, and possible revision, of traditional diagnostic frameworks that may better reflect the diverse range of social challenges faced by ADHD youth. Beyond current DSM-5-based ADHD nosology, future research may benefit from incorporating social functioning data to provide a more holistic approach and, subsequently, inform the development of tailored interventions to address specific areas of impairment. Importantly, mental health outcomes, including internalizing problems (e.g., withdrawal, somatic complaints, anxiety, depression) and externalizing problems (e.g., delinquent and aggressive behaviors), were better explained by our social profiles. This finding underscores the importance of considering comorbid mental health conditions as these may have significant implications for overall well-being and treatment planning among youth with ADHD.

Our results suggest that the use of social functioning profiles when assessing the presence of ADHD in youth may offer a more comprehensive understanding of the assessment and treatment of this diagnosis. Youth with ADHD face significantly more challenges compared to their TD counterparts, including forming social relationships, increased conflict reported within the familial context, as well as increased interparental conflicts, and higher rate of comorbidities (Muñoz-Silva et al., 2017, Weyers et al., 2019, Becker et al., 2012). In utilizing this data effectively, we can better explain the heterogeneous presentations of ADHD above and beyond what is currently understood by traditional DSM-5 diagnostic criteria. These findings highlight the need for a more tailored and holistic approach to diagnosis and interventions that take into account the complexity in ADHD diagnoses and the interplay between psychosocial factors that are prevalent in ADHD youth. Challenges in social functioning have significant consequences for youth with ADHD. This underscores the critical role of addressing these adversities as integral components in treatment to tailor interventions more effectively to fit individual needs and improve overall outcomes.

4.3. Replication of results

By leveraging a split-half analysis, the current study sought to ensure the stability and reliability of its findings. Overall, we observed a robust replication in the identification of socially derived profiles and their alignment with DSM-5 nosology, supporting the stability and clinical relevance of these data-driven groupings. Importantly, the four-profile solution of social functioning identified in ADHD Sample 1 was replicated in ADHD Sample 2, with nearly identical structure and distribution across profiles. This replication reinforces the reliability of the latent social profiles and highlights their potential utility for capturing meaningful variation in social functioning among youth with ADHD. Similarly, comparisons between social profiles and DSM-5 presentations yielded consistent patterns across both samples, further supporting the added value of data-driven profiles for understanding functional outcomes in ADHD. However, resting-state functional connectivity differences observed between profiles in Sample 1 were not replicated in Sample 2. This discrepancy suggests that while the behavioral expression of social functioning among youth with ADHD was consistent and clinically meaningful, these profiles may not be characterized by distinct or reproducible patterns of intrinsic brain connectivity. The replication results indicated that the biosignatures across profiles are not substantially different, reinforcing the idea that these groups, while behaviorally distinct, remain neurobiologically heterogeneous and better conceptualized as points along a spectrum rather than discrete categories. The absence of stable neural signatures highlights the challenge of mapping behavioral phenotypes onto biological substrates in ADHD. We chose to consider social functioning alone, rather than in combination with executive functioning, another critical domain relevant to ADHD. This was a deliberate decision that was intended to isolate social functioning and evaluate its standalone contribution to the pathophysiology of ADHD. We found that clean, replicable profiles of social functioning do emerge, but these do not translate into clearly differentiated neurobiological signatures. This underscores the need for future work to integrate additional dimensions, most notably executive function, to better capture the multidimensional nature of ADHD and more precisely differentiate subgroups at the neural level. Task-based neuroimaging or multimodal designs may be especially useful in identifying dynamic mechanisms underlying individual differences in social outcomes.

4.4. Limitations

Several limitations should be noted. First, the ABCD data presented in the current study are cross-sectional (taken at baseline point), which does not allow for a comprehensive understanding of the associations among the variables of interest over time. For instance, longitudinal data may provide further understanding whether rsFC differences predict changes and/or the presence of social functioning. Second, annotations of RDoC domains on social-related items were conducted manually by team associates, which may have led to some degree of subjectivity during the annotation process in the way each social item was categorized. Due to this, efforts were taken to increase reliability by conducting multiple levels of review, including collaboration between study associates followed by a single associate reviewing all annotations to ensure accuracy and consistency. Third, although the current study applied a dimensional, data-driven framework to characterize heterogeneity in social functioning, the LPA was conducted only among youth meeting full diagnostic criteria for ADHD. This decision aligned with our goal of identifying variability within clinically diagnosed youth; however, excluding youth with subthreshold ADHD symptoms and typically developing peers limits the dimensional scope. Future studies should explore whether similar profiles emerge in a broader, transdiagnostic sample to enhance generalizability and capture the full spectrum of social functioning. Fourth and finally, future research could examine subregions within subcortical structures to examine specificity of associations with social functioning impairment, achieving a more nuanced understanding of the neurobiological processes underlying ADHD heterogeneity.

5. Conclusion

Herein, a novel approach was assumed by exploring social functioning as a stand-alone domain, providing critical insights into both the conceptualization and clinical care of ADHD youth. Through latent profile analysis of ABCD Study data, we identified four distinct social functioning profiles among youth with ADHD, shedding light on the nuanced variations in social challenges experienced by these individuals. Comparisons with traditional DSM-5 nosology highlighted the superior explanatory power of data-driven social functioning profiles in understanding outcomes related to peer relationships, family conflict, and mental health, emphasizing the potential for a more comprehensive assessment and treatment approach beyond current diagnostic criteria. However, the absence of consistent resting-state functional connectivity differences across samples suggests that social functioning, when examined in isolation, may not delineate biologically distinct subgroups. While we found stable and clinically meaningful behavioral profiles, these did not translate into clearly differentiated patterns of intrinsic brain connectivity, reinforcing the neurobiological heterogeneity of ADHD. Future work should explore whether incorporating additional dimensions, most notably executive functioning, can enhance the precision of neurobiological subgrouping and further clarify the mechanisms underlying individual differences in social outcomes. Overall, our study contributes to a deeper understanding of ADHD heterogeneity and underscores the importance of considering social functioning challenges among ADHD youth in clinical theory and practice.

CRediT authorship contribution statement

Donisha D. Smith: Writing – review & editing, Data curation. Kathleen E. Feeney: Writing – review & editing. Katherine M. Schmarder: Writing – review & editing. Matthew T. Sutherland: Writing – review & editing, Funding acquisition, Data curation. Raul Gonzalez: Writing – review & editing, Funding acquisition. Erica D. Musser: Writing – review & editing, Methodology, Investigation, Conceptualization. Angela R. Laird: Writing – review & editing, Supervision, Resources, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization. Rosario Pintos Lobo: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Julio A. Peraza: Visualization, Data curation. Taylor Salo: Writing – review & editing. Alan Meca: Writing – review & editing, Methodology.

Data Statement

Data used in the preparation of this article were obtained from the Adolescent Brain Cognitive Development^SM (ABCD) Study (https://abcdstudy.org), held in the NIMH Data Archive (NDA). This is a multisite, longitudinal study designed to recruit more than 10,000 children aged 9–10 and follow them over 10 years into early adulthood. The ABCD Study® is supported by the National Institutes of Health and additional federal partners under award numbers U01DA041048, U01DA050989, U01DA051016, U01DA041022, U01DA051018, U01DA051037, U01DA050987, U01DA041174, U01DA041106, U01DA041117, U01DA041028, U01DA041134, U01DA050988, U01DA051039, U01DA041156, U01DA041025, U01DA041120, U01DA051038, U01DA041148, U01DA041093, U01DA041089, U24DA041123, U24DA041147. A full list of supporters is available at https://abcdstudy.org/federal-partners.html. A listing of participating sites and a complete listing of the study investigators can be found at https://abcdstudy.org/consortium_members/. ABCD consortium investigators designed and implemented the study and/or provided data but did not necessarily participate in the analysis or writing of this report. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH or ABCD consortium investigators.

The ABCD data repository grows and changes over time. The ABCD data used in this report came from NDA ABCD Release 5.1 (DOI: 10.15154/1520591).

Code and Materials Availability

Additional information and resources will be made available on a project page for this study at the Open Science Framework (OSF; https://osf.io). The code will be available in GitHub (https://github.com) repositories, including the fMRI preprocessing and analyses. ROI and connectivity maps will be made available in NeuroVault (neurovault.org). High-resolution figures will be made available via FigShare (https://figshare.com).

Supplementary Material

Refer to DCN Stage 2- Pintos Lobo-supp-DCN.docx for supplementary material:

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

Supplementary material

Data availability

The authors do not have permission to share data.