The Hierarchical Taxonomy of Psychopathology in Adolescents: Support for a Neurodevelopmental Spectrum Without ADHD

Abstract

Using the Hierarchical Taxonomy of Psychopathology (HiTOP) as our framework, the current study examines how 13 common psychological disorders can be grouped into different spectra in two groups of adolescents: a community sample (N = 951), and a clinic-referred sample (N = 173). Scores for the disorders were obtained using the parent version of the Child and Adolescent PsychProfiler. Taken together, the findings across the two samples for factor structure, reliability, and discriminant and concurrent validity indicate the most support for a three-factor CFA oblique model with primary factors for neurodevelopment disorders (that include Specific Learning Disorder, Autism Spectrum Disorder, Language Disorder, and Speech Sound Disorder), internalizing disorder problems (that include Generalized Anxiety Disorder, Persistent Depressive Disorder, Separation Anxiety Disorder, Obsessive–Compulsive Disorder, Posttraumatic Stress Disorder, Anorexia Nervosa, and Bulimia Nervosa), and externalizing disorder problems [(that include Attention Deficit/Hyperactivity Disorder (ADHD), and Oppositional Defiant Disorder/Conduct Disorder (ODD/CD)], with a covariance for the error variance for Anorexia Nervosa and Bulimia Nervosa. Additionally, the analysis for Sample 2 supports the concurrent validity of the factors in this model. A modification of this model, with ADHD cross-loading on the neurodevelopment disorders factor, did not produce an admissible solution. The findings indicate support for a neurodevelopmental spectrum in the HiTOP model, with ADHD and ODD/CD showing stronger statistical association with the externalizing factor than with the neurodevelopmental factor in the models tested. This finding pertains to dimensional structure and does not invalidate the neurodevelopmental classification of ADHD in DSM-5-TR.

1. Introduction

The taxonomy of psychological disorders is an ongoing area of intense research. Although the traditional conceptualization of psychopathology, like that in DSM-5-TR [1], is categorical, alternative dimensional conceptualizations of psychopathology are now gaining traction. A popular influential dimensional model of psychopathology is the Hierarchical Taxonomy of Psychopathology (HiTOP; [2,3]). Data-driven and based mainly on adult studies, when initially published, it was suggested that although the model was ready for use by scientists and clinicians, there was still need for more research on the model, especially in terms of inclusion and placement of psychopathologies not included in the model [4]. Relatedly, even though neurodevelopmental disorders play critical roles in psychopathogenesis, they were not featured in the original HiTOP model. Therefore, the major goal of the current study is to examine for a group of adolescents the inclusion and location of a neurodevelopment problems spectrum within this model, and related to it, the disorders in this spectrum, especially attention deficit/hyperactivity disorder (ADHD), oppositional defiant disorder (ODD), and conduct disorder (CD).

1.1. The Original Hierarchical Taxonomy of Psychopathology (HiTOP)

Structurally, based on the results of factor analysis studies, the original HiTOP proposed by Kotov et al. [2] organized psychopathology at five different hierarchical levels, moving upwards from narrow to broader dimensional constructs of psychopathology [3,5]. At the bottom of the hierarchy are signs, symptoms, and maladaptive traits of psychopathology, such as impulsivity, dysphoria, insomnia, worry, avoidance, irritability, anxiousness, emotional lability, and inattention. At the level just above or the second level, there are empirically derived syndromes. Although HITOP syndromes do not map onto DSM-5-TR or ICD-11 [6] disorders, this level most closely corresponds to them [7]. At the level above or third level are subfactors (i.e., eating pathology, fear, distress, mania, and antisocial behaviour, sexual problems, and substance use). At the fourth level are the spectra (i.e., internalizing, thought disorder, detachment, somatoform, disinhibited externalizing, and antagonistic externalizing problems). At the very top, the broadest level or super-spectra level, is a single general or p-factor.

With reference to the different spectra, the syndromes and disorders for the internalizing spectrum include distress (e.g., depression, anxiety, and PTSD; mania [e.g., bipolar I disorder, and bipolar II disorder]), fear (e.g., panic disorder, social phobia, and specific phobia), obsessive–compulsive disorder, separation anxiety disorder, eating pathology (e.g., anorexia nervosa and bulimia nervosa) and sexual problems (e.g., hyperactive sexual desire disorder, and delayed ejaculation) [8]. The syndromes for the externalizing spectrum include disinhibition (substance use, hyperactivity, and inattention), antisocial behaviour (aggression, destruction of property, fraud, and theft), and antagonism, reflecting primarily antisocial behaviour (i.e., antisocial personality, oppositional defiant disorder, conduct disorder, narcissistic personality, and histrionic personality) [9]. The behaviour signs and syndromes for the somatoform spectrum include somatic symptom disorder and illness anxiety disorder. The syndromes for the detachment spectrum include anhedonia, suspiciousness, social withdrawal, intimacy avoidance, unassertiveness, risk aversion, and restricted affectivity [10,11]. The syndromes for the thought disorder spectrum include psychoticism, hallucinations, and delusions [10].

Although the original HiTOP model proposed by Kotov et al. [2] did not include a spectrum for neurodevelopmental disorders, a recent study involving older youths and adults found support for internalizing, externalizing, thought disorder, neurodevelopmental and cognitive difficulties, somatoform, and mania/low detachment spectra [12]. In a factor analytic study of parent ratings of items from the Child Behaviour Checklist (CBCL) [13] in 10-year-old children, and adult self-ratings of the Adult Self-Report (ASR) [14], support was found for such a spectrum [15]. The study found that for children there were spectra for internalizing, somatoform, detachment, neurodevelopmental, and externalizing; and for adults there were spectra for internalizing, somatoform, detachment, inattentive neurodevelopmental, and externalizing. The neurodevelopmental spectrum for children comprised primarily inattention (IA) and hyperactivity/impulsivity (HI), and the adult inattentive neurodevelopmental spectrum comprised primarily IA.

The disorders for the neurodevelopmental-related spectra reported by Michelini et al. [15] are not surprising, as the CBCL and the ARS have limited items to measure neurodevelopmental problems, such as language disorder, speech sound disorder, autism spectrum disorder, neurodevelopmental motor disorders, tic disorders, and specific learning disorder. Notwithstanding this, the findings in that study raise several issues that have more general relevance for the HiTOP model. First, it can be argued that the neurodevelopment spectrum and inattentive neurodevelopmental spectrum reported in that study are not reflective of bonified neurodevelopment dimensions or spectra but rather dimensions specific to ADHD. Ideally, a stronger claim for a neurodevelopment spectrum can be made if we can identify one which includes many more neurodevelopmental symptoms/syndromes, like language problems, speech sound problems, autism spectrum problems, developmental motor coordination problems, tic problems, and specific learning problems. Second, the emergence of a spectrum for the core ADHD symptoms (IA and HI) raises questions about the location of ADHD in the HiTOP model. In the initially proposed version of the HiTOP model, ADHD was in the externalizing spectrum. Given the findings reported by Michelini et al. [15], it can be speculated that the initially proposed location of ADHD in the disinhibition externalizing spectrum of the HiTOP model is open to question. It could be either part of the neurodevelopmental spectrum, or part of the externalizing spectrum, or part of both neurodevelopmental and disinhibition externalizing, or even a spectrum on its own. There is empirical support for such possibilities. With reference to a set of `validating criteria’ listed by the DSM-5 Study Group, Andrews et al. [16] concluded that although ADHD exhibited a relationship with an externalizing disorder cluster, it also showed some associations with a neurodevelopmental disorder cluster, albeit less strongly. Relatedly, although not arising from the finding in Michelini et al.’s [15] study, there are reasons to suspect that the initially proposed location of childhood disruptive disorders like ODD and CD in the externalizing spectrum is also open to question. This is because a comprehensive review by Wakschlag et al. [17] concluded that the weight of evidence from developmental, clinical, and neuroscience areas indicate that these disorders and problems are more closely related to neurodevelopmental disorders and problems. For example, Kerekes et al. [18] found strong associations for ODD and CD with autism spectrum disorder (ASD) and ADHD (HI in boys, and IA in girls). Also, genetic and environmental effects linked to ADHD and ASD also influenced ODD, and to a lesser extent, CD, especially in boys. Taking such findings into consideration, it can be argued that ODD and CD could potentially be better located in the neurodevelopmental spectrum rather than the externalizing spectrum of the HiTOP.

1.2. Limitations of Findings in Existing Literature

Overall, the existing findings in this area are limited. First, research is yet to clearly demonstrate the presence of a broadly defined neurodevelopmental spectrum in the HiTOP model. In this respect, more studies with measures covering a wider range of neurodevelopmental problems are needed to bring more clarity to this area. Second, it is not certain that if such a spectrum is present, whether ADHD is part of this spectrum or another spectrum with other externalizing disorders (externalizing spectrum), as currently proposed for the HiTOP model [2]. Third, although the HiTOP model has located childhood disruptive disorders (ODD and CD) in the externalizing spectrum, there is sufficient evidence to suspect that they might be better located in the neurodevelopmental spectrum (if present) rather than the externalizing spectrum. Fourth, to date there has been limited validation of the HiTOP model in children and adolescents [12,19]. While most studies involving children and adolescents have generally examined the hierarchal structure for a limited number of disorders [12,20], adding to the breadth of disorders examined would provide a more meaningful understanding of dimensional classification of psychopathology. With reference to children, Michelini et al.’s [15] study, discussed earlier, is a good example. At least two studies involving adolescents have examined concurrently a wide range of internalizing, externalizing, neurodevelopmental, cognitive, and eating disorders [19,21]. Hankin et al. [21] revealed a structure with a general psychopathology at the very top, nine narrowband symptom factors at the very bottom, and intermediate factors for internalizing, disinhibited externalizing, and personality dysregulation. The findings in the study by Forbes et al. [19] that included both children and adolescents, covering nearly all major forms of mental disorders and related content domains (e.g., impulsivity), indicates a hierarchical structure with 15 narrow dimensions under four higher-order domains that could be considered spectra: internalizing; externalizing; eating pathology; and uncontrollable worries, obsessions, and compulsions. Taken together, while these studies support the presence of spectra for internalizing and externalizing problems, they do not demonstrate a spectrum for neurodevelopmental problems. Clearly, considering that such a spectrum has been demonstrated for adults and children, more studies involving adolescents are needed to bring clarity to this area.

1.3. Aims of the Study

Given the limitations highlighted above, the current study has three aims. The first aim is to evaluate empirical support for the presence of a neurodevelopmental spectrum within HiTOP. This was examined in two groups of adolescents: a community sample, and a clinic-referred sample. The second and third aims, contingent on support for a neurodevelopmental spectrum, are to examine if ADHD, ODD and CD are better located in the externalizing spectrum or in the neurodevelopment spectrum. To achieve our study goals, for both study samples, we tested four CFA models in a data set that included 13 DSM-5-TR-based screening scores that cover internalizing, externalizing and neurodevelopmental disorders. They were Attention Deficit/Hyperactivity Disorder (ADHD), Oppositional Defiant Disorder plus Conduct Disorder (ODD/CD), Specific Learning Disorder (SLD), Autism Spectrum Disorder (ASD), Language Disorder (LD), Speech Sound Disorder (SSD), Generalized Anxiety Disorder (GAD), Persistent Depressive Disorder (PDD), Separation Anxiety Disorder (SAD), Obsessive–Compulsive Disorder (OCD), Posttraumatic Stress Disorder (PTSD), Anorexia Nervosa (AN), and Bulimia Nervosa (BN). The scores for these disorders were derived from parent ratings of the Child and Adolescent PsychProfiler (CAPP-PRF) [22,23]. A recent study [24] showed support for the factor structure, reliability and validity of the CAPP-PRF. Also, the associations for these disorders could be broadly grouped into internalizing, externalizing and neurodevelopmental factors.

The four factor models tested in both the community and clinic-referred samples, shown schematically in Supplementary Materials, Figure S1, were (1) a one-factor model with all 13 loading on a single factor [Model (M) 1]; (2) a two-factor oblique model, with neurodevelopmental disorders (ADHD, ODD/CD, SLD, ASD, LD and SSD), and internalizing disorders (that GAD, SAD, PDD, OCD, PTSD, AN and BN; M 2) loading onto two separate factors; (3) a three-factor oblique model, with neurodevelopment disorders (limiting to SLD, ASD, LD and SS), internalizing disorders (GAD, PDD, SAD, OCD and PTSD), and externalizing problems (covering ADHD and ODD/CD; M3) loading onto three separate factors; and (4) a revised three-factor oblique model, with ADHD cross-loading on the neurodevelopment disorders factor (M4), i.e., factors for neurodevelopmental disorders (ADHD, SLD, ASD, LD and SSD), internalizing disorders (GAD, PDD, SAD, OCD and PTSD), and externalizing disorders (ADHD and ODD/CD).

Support for Model 1 would indicate that the different psychopathologies considered in the study are best viewed in terms of a single general or p-factor and cannot be organized into different spectra below it. Support for Model 2 would indicate that the different psychopathologies examined in the current study can be grouped into two spectra, i.e., a neurodevelopmental spectrum (ADHD, ODD/CD, SLD, ASD, LD and SSD), and an internalizing problems spectrum (GAD, SAD, PDD, OCD, PTSD, AN and BN). Support for Model 3 would indicate that the different disorders can be grouped into three spectra, i.e., a neurodevelopmental spectrum without ADHD (i.e., that includes SLD, ASD, LD and SSD), an internalizing spectrum (GAD, PDD, SAD, OCD and PTSD), and an externalizing spectrum with ADHD (ADHD and ODD/CD). Support for Model 4 would indicate that the different disorders can be grouped into the same three spectra, with ADHD being part of both the externalizing spectrum and the neurodevelopmental spectrum. Based on the HiTOP model and the general literature reviewed in the introduction, we expected most support for Model 3. With reference to the aims of this study, this would mean support for a neurodevelopmental spectrum, with ADHD, ODD and CD loading on the externalizing spectrum rather than the neurodevelopmental spectrum.

2. Method

2.1. Participants

The community sample comprised 951 adolescents with parent ratings of the CAPP-PRF [23], the measure used here to obtain ratings for the 13 disorders included in the study. This measure is described in detail in the measures section. The mean age (SD; range) of participants was 14.54 years (1.66 years; 12.01 years to 17.99 years). There were 572 (60.1%) boys, (mean age = 14.54 years, SD = 1.70 years) and 372 (39.1%) girls (mean age = 14.52 years, SD = 1.60 years), and no gender information for 7 (0.7%) adolescents. There was no significant difference for age across boys and girls, t (df = 942) = 0.188, ns.

For the clinic-referred sample, there were 173 ratings of the CAPP-PRF for adolescents (age range from 12 to 17 years; mean age 14.57 years, SD = 1.84 years). There were 112 (64.7%) males, (mean age 14.40 years, SD = 2.30 years) and 61 (35.3%) females (mean age 14.69 years, SD = 1.80 years). There was no significant difference for age, t (df = 171) = 0.842, ns.

Although details were not collected, the majority of participants in both samples came from intact families, and their parents completed at least secondary education.

Supplementary Materials, Table S1, shows the frequencies of the different disorders for which adolescents in the community sample screened positive, based on the CAPP-PRF [23]. As shown, more than half the participants were screened positive for SLD. GAD, PDD, ADHD, and ODD/CD were quite prevalent (ranging from 30.1% to 40.9%). In contrast, the frequencies for positive screens for ASD, SSD, SAD, OCD, AN and BN were relatively low (ranging from 4% to 14.1%). These mean scores suggest that although this was a community sample, overall, there was a range of psychopathologies in the sample. The prevalence rates observed in the community sample are higher than would typically be expected in epidemiological studies of adolescents, particularly for SLD, PDD, and ODD/CD. The recruitment procedure is a likely contributor. The community sample was drawn from individuals who had voluntarily completed a paid online DSM-5-TR screening tool (the PsychProfiler), so the sample is best understood as a help-seeking or concern-prompted community sample rather than a general-population sample, and a degree of self-selection bias toward families with existing concerns about their adolescents’ functioning should be assumed. Supplementary Materials, Table S2, shows the frequencies of the different disorders for which adolescents in the clinic-referred sample screened positive, based on the CAPP-PRF [23]. As shown, more than half the participants screened positive for SLD and ADHD. PDD was also quite prevalent (37.6%). In contrast, the frequencies for a positive screen for ASD, SAD, OCD, AN, and BN were relatively low (ranging from 4% to 8.1%).

Supplementary Materials, Table S3, shows the mean (SD) scores for the BYI-2 constructs for the participants in the clinic-referred sample. These means scores suggest relatively high levels of psychopathology (see Table 5.19 of the BYI-2 Manual). In relation to intelligence, the mean (SD) for FSIQ was 87.77 (34.69). Taken together, these findings can be interpreted as indicating an IQ at a lower average, and noticeable levels of a range of psychopathologies in the clinic-referred sample.

2.2. Measures

As part of the study, parents of participants in the community sample completed the CAPP-PRF [23,24]. For the clinic-referred sample, parents of participants completed the CAPP-PRF, and adolescents completed the Beck Youth Inventories, Second Edition (BYI-2; [25]) and were also administrated the Wechsler Intelligence Scale for Children–Fifth Edition (WISC-V; [26]).

2.2.1. Parent Report Form of Child and Adolescent PsychProfiler (CAPP-PRF; [23])

Scores for the disorders for our CFA models were obtained using the CAPP-PRF [23]. The recent study by Langsford et al. [24] showed support for the factor structure, reliability and validity of the CAPP-PRF. The CAPP-PRF, with 17 screening scales, can simultaneously screen for 14 of the most common psychiatric, psychological, and educational disorders in children, adolescents, and adults. They are ADHD, ODD, CD, SLD, ASD, LD, SSD, GAD, PDD, SAD, PTSD, OCD, AN, and BN. In terms of the disorders modelled in the current study, all three ADHD presentations were grouped together in a single ADHD group, and because of low frequency of ODD (N = 3), ODD and CD were combined in a single group. This resulted in modelling 13 traditionally recognized internalizing, externalizing, and neurodevelopmental disorders for the CFA models. Consequently, we could only test the presence of these spectra, and not the spectra for detachment, somatoform, and thought disorder. Thus, the study is only a partial examination of the original version of the HITOP model.

In the CAPP-PRF, the different disorders are measured using items corresponding to their symptoms (including wording in most instances), as presented in DSM-5-TR. In all, the CAPP-PRF has 126 items that cover the 14 disorders, and all items are rated on a six-point Likert scale (never = 0, rarely = 1, sometimes = 2, regularly = 3, often = 4, and very often = 5). For calculating positive disorder screening scores, the item scores were recoded as follows: never, rarely and sometimes = 0; and regularly, often, and very often = 1. The summation of items within each disorder produces a screening score for that disorder, which, if met or exceeded the screening cut-off, was considered as having a ‘positive screen’. The cut-off scores for the respective disorders are identical to the symptom threshold scores for them in the DSM-5-TR. Binary “positive screen” indicators were used in preference to continuous symptom totals for three reasons. First, the CAPP-PRF was designed and validated as a disorder-level screening instrument, with cut-offs anchored to DSM-5-TR symptom thresholds; modelling outcomes at the disorder-screen level therefore preserves the clinical decision unit the instrument is intended to inform. Second, several disorders (e.g., ASD, SSD, AN, BN) had low base rates in the community sample, and the distributions of their continuous scores were strongly skewed, so the binary indicators provided a more comparable metric across disorders. Third, this scoring approach is consistent with how the CAPP-PRF is used in applied screening settings, which improves the external relevance of the structural findings. We acknowledge, however, that dichotomisation reduces variability and discards information on sub-threshold symptom severity, and consider this further in the limitations. In this study, the internal consistency reliability omega coefficient values ranged from 0.809 to 0.927, and the alpha coefficients ranged from 0.759 to 0.928 (see Supplementary Materials, Table S1, column 3 for details).

2.2.2. The Beck Youth Inventories, Second Edition (BYI-2; [25])

The Beck Youth Inventories, Second Edition [25], used for children and adolescents aged between 7 and 18 years, comprises five self-report inventories, measuring depression, anxiety, anger, disruptive behaviour, and self-concept. Each questionnaire has 20 items, resulting in 100 items in total. Individuals rate all 100 items in terms of the extent to which each statement describes them on a 4-point Likert scale (i.e., “0 = never”, “1 = sometimes”, “2 = often”, and “3 = very often”). All five inventories have good construct validity, high reliability (coefficient alpha ranging from 0.86 to 0.96), and test–retest reliability (coefficients ranging from 0.74 to 0.93) [25,26]. Thus, existing findings indicate good support for its psychometric properties (factor structure, reliability, and validity), and utility with adolescents [27].

The current study used only the depression, anxiety, anger, and disruptive behaviour inventories. Among others, the Depression Inventory includes items covering negative thoughts, feelings of guilt and sadness, and sleep issues (e.g., “I have trouble sleeping”). The Anxiety Inventory includes items covering concerns and apprehension regarding school, the future, reactions from others, losing control, and physiological anxiety symptoms (such as, “My hands shake”). The Anger Inventory focuses on feelings of hatred and anger as well as thoughts of unjust or unfair treatment (such as, “I get mad and stay mad”). The Disruptive Behaviour Inventory includes items related to behaviours and attitude associated with ODD and CD (e.g., “I hurt people”). For each inventory, item scores are summed and converted to T-Scores, with higher scores for depression, anxiety, anger, and disruptive behaviour indicating greater severity. The anxiety, depression, anger and disruptive behaviour scales have shown adequate convergent validity with scales measuring related constructs [25], and ability to distinguish between clinical and nonclinical samples [28]. Therefore, anxiety and depression scales can be considered relevant for testing the concurrent validity of an internalizing spectrum, and the disruptive behaviour scale, especially if it includes ODD and/or CD, as relevant for testing the concurrent validity of an externalizing spectrum.

2.2.3. Wechsler Intelligence Scale for Children–Fifth Edition (WISC–V; [26])

The WISC-V is an individually administered test of intelligence for children from 6 years 0 months to 16 years 11 months. It provides index scores for various cognitive areas, and a composite score that represents general intellectual ability, called the Full-Scale IQ (FSIQ). Intellectual disability, considered to be a neurodevelopmental disorder (DSM-5-TR; [1]) is characterized primarily by low FSIQ. Thus, the FSIQ can be considered relevant for testing the concurrent validity of a neurodevelopmental spectrum.

2.3. Procedure

The participants in the community sample were all individuals who provided data through the PsychProfiler measures website (https://www.psychprofiler.com). This website can be used by those interested in online screening of DSM-5-TR psychological disorders, after paying a nominal fee. The primary users are psychologists, doctors, and members of the general public. Upon completion of the PsychProfiler, individuals are requested, if they so wish, to click a statement granting permission for their data to be used for future research and validation purposes. Only adolescents with CAPP-PRF ratings and consent were included in the study.

All adolescent participants in the clinic-referred sample were seen in a clinic setting in Subiaco, Perth, Western Australia. They were attending the clinic to complete an ADHD, SLD, ASD, or intelligence assessment. These adolescents were referred to the clinic from a variety of sources (e.g., privately by their parents, through their school, or from a GP, pediatrician, or child and adolescent psychiatrist). For this sample, parents completed a suite of checklists as part of the assessment that their children were undergoing. Within the suite of checklists was the CAPP-PRF and the BYI-2. Also included with the checklists was a consent form for signing should they consent to their child’s de-identified data being used for future research purposes. Only information of adolescents whose parents and adolescents had signed the consent form were included in the sample.

For both the community and clinic-referred samples, the responses were entered into a scoring software for scoring and the summary scores. The observed indicators in all the CFA models were the total recoded positive screening scores. As previously mentioned, an adolescent received a positive screen for a “disorder” if the parents’ scores met or exceeded the DSM-5-TR cutoff for that particular disorder.

2.4. Statistical Analysis

Mplus Version 7 (Mplus, Los Angeles, CA, USA) [29] was used to compute all the CFA analyses. WLSMV extraction was used to test these models as the observed indicators for the models were binary (disorder screened positive versus disorder screened negative). Only adolescents with complete CAPP-PRF data on all 13 modelled disorders were retained for the analyses; cases with missing screening scores on any of the modelled indicators were excluded prior to estimation, which is the default pairwise/listwise handling under WLSMV in Mplus when there are no auxiliary variables. The proportion of cases excluded for missingness was small (<3% in both samples) and did not differ meaningfully by age or sex. The observed indicators were presence/absence of ADHD, ODD/CD, SLD, ASD, LD, SSD, GAD, PDD, SAD, OCD, PTSD, AN, and BN. A sequential stepwise algorithm was used to examine the fit indices of CFA models examined in the study. In step 1, the global fit of the model was examined using root mean squared error of approximation (RMSEA), the comparative fit index (CFI), and the Tucker–Lewis index (TLI). Hu and Bentler [30] have proposed that RMSEA values < 06 = good fit, <0.08 = acceptable fit, and >0.08 to 0.10 = marginal fit. For CFI and TLI, values ≥ 0.95 = good fit, and ≥0.90 = acceptable fit. As all our models were nested, we used ∆WLSMV to compare models, when deemed necessary. A significant difference would indicate better fit for the more restricted model. For the best fitting model, it was also necessary for the loadings of indicators in the model to be significant and salient, >0.30 [31], and for the factors to demonstrate acceptable discriminant validity, r < 0.80 [32], reliable omega coefficients [33,34], and concurrent validities. Although there are no universally accepted guidelines at present for interpreting omega coefficients, Watkins [35] has proposed that omega values should meet the same standards as alpha coefficients. For alpha coefficients, guidelines for acceptability have a range from 0.70 [36] to 0.96 [37]. For the current study we used omega values of 0.70 and above as acceptable. The concurrent validities of the factors in the optimum model tested was evaluated in the clinic-referred sample by correlating the latent factors in this model with the factors for anxiety, depression, anger and disruptive behaviour, as measured by the BYI-2 and WISC-V FSIQ scores.

3. Results

3.1. Sample Size Requirements in the Community and Clinic-Referred Samples

Soper’s [38] software (DanielSoper.com, Fullerton, CA, USA) for computing sample size requirements for the CFA models was used to evaluate the sample size required for the study. For both samples, the anticipated effect size was set at 0.3, power at 0.8, the number of latent variables at 4, the number of observed variables at 13, and probability at 0.05. The analysis recommended a minimum sample size of 495 for the community sample, and 166 for the clinic-referred sample. Therefore, with a sample size of N = 952 in the community sample and 173 in the clinic-referred sample, our sample sizes were adequate for the analyses involving both samples.

3.2. Fit Values of All the Models Tested in the Study for the Community Sample

Table 1. Fit values of the CFA models tested in the community sample.

	Model Fit					Model Difference (Δ)
Model (M)	WLSMVχ2	df	RMSEA (90% C.I.)	CFI	TLI	ΔM	Δdf	Δχ2
One-factor (Model 1)	903.385	65	0.116 (0.110–0.123)	0.774	0.729	-	-	-
Two-factor (Model 2)	634.099	64	0.097 (0.090–0.104)	0.847	0.813	M1–M2	1	133.629 ***
Three-factor CFA (Model 3)	481.728	62	0.084 (0.077–0.091)	0.887	0.858	M1–M3 M2–M3	3 2	253.857 *** 101.900 ***
Three-factor CFA (Model 3b) Error covariance between AN and BN	394.304	61	0.076 (0.069–0.083)	0.910	0.885	M1–M3b M2–M3b M3–M3b	4 3 1	304.785 *** 149.651 *** 48.659 ***
Three-factor CFA with ADHD cross-loading on neurodevelopmental spectrum (Model 4)	Non-admissible solution. The residual covariance matrix was not positive definite, with ODD/CD loading on its latent factor > 1 (i.e., 14.256)

Notes: WLSMVχ² = weighted least square mean and variance adjusted; RMSEA = root mean square error of approximation; CFI = comparative fit index; TLI = Tucker–Lewis index; C.I. = confidence interval. All WLSMVχ² values were significant. *** p < 0.001.

shows the fit values for all the models tested in the study for the community sample. The one-factor (M1), two-factor (M2), and three-factor (M3) models showed poor fit in terms of their CFI, TLI and RMSEA values. RMSEA values indicated a poor fit for the one-factor model, and a marginal fit for the two-factor model and three-factor model. The three-factor model showed better fit than the three-factor model and two-factor model. The output for the three-factor model indicated high covariance between the error variances for AN and BN. Consequently, we ran a modified three-factor model (M3b) in which this covariance was modelled. This is theoretically acceptable as AN and BN are highly related [39]. The modified three-factor model (M3b) indicated a mixed fit. The CFI (0.910) reached the conventional threshold for an acceptable fit, while the TLI (0.885) fell just below the 0.90 threshold and the RMSEA (0.076) was at the upper limit of acceptable fit (i.e., approaching the 0.08 boundary). These values should therefore be interpreted as indicating adequate, rather than good, model fit, and conclusions drawn from this model warrant a degree of caution. The three-factor model with cross-loadings or M3 with ADHD cross-loading on the neurodevelopmental spectrum (M4) did not produce an admissible solution as the residual covariance matrix was not positively defined, with ODD/CD loading on its latent factor at >1. Thus, although the modified three-factor model with correlated error variance between AN and BN (M3b) showed a mixed fit, it was better than the one-factor model (M1), two-factor model (M2), and three-factor model with cross-loading (M4). It was therefore considered to be our preferred model that we examined further for reliability.

3.3. Factor Loadings of the Indicators, and Discriminant Validity and Reliabilities of the Factors in the Modified Three-Factor (M3b) Model for the Community Sample

Table 2. Factor loadings, omega coefficients, and factor correlations of the three-factor model in the community sample.

	Neurodevelopment	Internalizing	Externalizing
Specific Learning Disorder	0.615
Autism Spectrum Disorder	0.850
Language Disorder	0.866
Speech Sound Disorder	0.772
Generalized Anxiety Disorder		0.939
Persistent Depressive Disorder		0.871
Separation Anxiety Disorder		0.560
Posttraumatic Stress Disorder		0.808
Obsessive–Compulsive Disorder		0.672
Anorexia Nervosa		0.516
Bulimia Nervosa		0.568
Attention Deficit/Hyperactivity Disorder			0.320
Conduct/Oppositional Defiant Disorder			0.747
Reliability
Omega	0.861	0.878	0.467
Factor correlation
Neurodevelopment	-	0.431 **	0.494 ***
Internalizing		-	0.722 ***
Externalizing			-

Notes: All factor loadings were significant (p > 0.001) and salient (loadings > 0.30). ** p < 0.01. *** p < 0.001.

shows the factor loadings for the M3b. As shown in this table, all the indicators loaded significantly and saliently (>0.30) on their respective factors. Also, based on criteria proposed by Brown [32] for evaluating discriminant validity (r < 0.80), all factors showed discriminant validity. The omega reliability coefficient values were 0.86 for the neurodevelopmental factor, 0.88 for the internalizing factor, and 0.46 for the externalizing factor. Based on a value of at least 0.70 as acceptable reliability, the omega-based reliability coefficients for the factors can be considered acceptable for neurodevelopmental and internalizing factors (spectra), and unacceptable for the externalizing factor (spectrum).

3.4. Fit Values of All the Models Tested in the Study for the Clinic-Referred Sample

Table 3. Fit values of the CFA models tested in the clinic-referred sample.

	Model Fit					Model Difference (Δ)
Model (M)	WLSMVχ2	df	RMSEA (90% C.I.)	CFI	TLI	ΔM	Δdf	Δχ2
One-factor (Model 1)	176.989	65	0.100 (0.082–0.118)	0.849	0.819	-	-	-
Two-factor (Model 2)	129.055	64	0.077 (0.057–0.096)	0.912	0.893	M1–M2	1	36.257 ***
Three-factor CFA (Model 3)	101.807	62	0.061 (0.039–0.082)	0.946	0.932	M1–M3 M2–M3	3 2	59.846 *** 22.423 ***
Three-factor CFA with ADHD cross-loading on neurodevelopmental spectrum (Model 4)	Non-admissible solution. The residual covariance matrix was not positive definite, with ODD/CD loading on its latent factor > 1 (i.e., 1.311)

Notes: WLSMVχ² = weighted least square mean and variance adjusted; RMSEA = root mean square error of approximation; CFI = comparative fit index; TLI = Tucker–Lewis index; C.I. = confidence interval. All WLSMVχ² values were significant. *** p < 0.001.

shows the fit values for all the models tested in the study for the clinic-referred sample. The one-factor model (M1) showed poor fit in terms of their CFI, TLI and RMSEA values. The two-factor model (M2) showed adequate fit in terms of its CFI and RMSRA values, and close to adequate fit for the TLI value. The three-factor model (M3), without covariance between the error variance for Anorexia Nervosa and Bulimia Nervosa, showed adequate fit in terms of its CFI and RMSRA values, and close to adequate fit for the TLI value. The three-factor model with cross-loading (M4) for ADHD on the neurodevelopmental and externalizing factors (M4) did not produce an admissible solution. Thus, the original three-factor model (M3) was adopted as the preferred model and was examined further for reliability and validity.

3.5. Factor Loadings of the Indicators, and Discriminant Validity and Reliabilities of the Factors in the Three-Factor (M3) Model for the Clinic-Referred Sample

Table 4. Factor loadings, omega coefficients, and factor correlations of the three-factor model in the clinic-referred sample.

	Neurodevelopment	Internalizing	Externalizing
Specific Learning Disorder	0.598
Autism Spectrum Disorder	0.880
Language Disorder	0.805
Speech Sound Disorder	0.919
Generalized Anxiety Disorder		0.937
Persistent Depressive Disorder		0.960
Separation Anxiety Disorder		0.582
Posttraumatic Stress Disorder		0.676
Obsessive–Compulsive Disorder		0.619
Anorexia Nervosa		0.807
Bulimia Nervosa		0.860
Attention Deficit/Hyperactivity Disorder			0.595
Conduct/Oppositional Defiant Disorder			0.841
Reliability
Omega	0.882	0.918	0.687
Factor correlation
Neurodevelopment	-	0.376 **	0.513 ***
Internalizing		-	0.699 ***
Externalizing			-

Notes: All factor loadings were significant (p > 0.001), and salient (loadings > 0.30). ** p < 0.01. *** p < 0.001.

shows the factor loadings for the three-factor (M3) model. As shown in this table, all the indicators loaded significantly and saliently (>0.30) on their respective factors. Also, all factors demonstrate acceptable discriminant validity, based on r < 0.80 [32]. The omega reliability coefficient values were 0.88 for the neurodevelopmental factor, 0.93 for the internalizing factor, and 0.69 for the externalizing factor. Based on a value of at least 0.70 as acceptable reliability, the omega-based reliability coefficients for the factors in the three-factor model (M3) can be considered acceptable for neurodevelopmental and internalizing factors, and very close to acceptable for the externalizing factor.

3.6. Concurrent Validities of the Factors in M3 in the Clinic-Referred Sample

Table 5. Path coefficient for the correlation of the BYI-2 constructs on the HiTOP factors in the three-factor model in the clinical-referred sample.

	Anxiety	Depression	Anger	Disruptive Behaviour	Full Scale IQ
Neurodevelopment	−0.185	−0.202	−0.028	0.050	−0.218 *
Internalizing	0.146 ***	0.483 ***	0.541 ***	0.437 ***	−0.026
Externalizing	0.022	0.045	0.257 **	0.444 ***	−0.097

Notes: * p < 0.05; ** p < 0.01; *** p < 0.001.

shows the correlation coefficients for the BYI-2 constructs (anxiety, depression, anger, and disruptive behaviour) and WISC-V FSIQ with the latent factors (neurodevelopment, internalizing, externalizing, and general p-factor) in the three-factor (M3) model. As shown, the neurodevelopmental factor was associated negatively and significantly with FSIQ, with a small effect size. The internalizing factor was associated positively with anxiety, and depression with small and moderate effect sizes, respectively, and with anger, and disruptive behaviour with moderate effect sizes. The externalizing factor was associated positively with anger, with a small effect size, and with disruptive behaviour, with a moderate effect size. These associations can be interpreted as supportive of concurrent validity for the neurodevelopmental and externalizing spectra, and to a lesser degree for the internalizing spectrum. It should be noted that the correlations observed here are generally small-to-moderate in magnitude and so provide only initial, rather than strong, evidence of concurrent validity; effect sizes should be interpreted with the appropriate degree of caution. One unexpected pattern was that the internalizing factor correlated more strongly with anger (r = 0.54) and disruptive behaviour (r = 0.44) than with anxiety (r = 0.15) in the clinic-referred sample. This is unlikely to reflect a failure of the internalizing factor and more likely reflects two features of this sample. First, the clinic-referred adolescents were predominantly referred for ADHD, SLD, ASD, or intellectual assessment, so internalizing symptoms in this group often co-occurred with externally directed irritability and behavioural dysregulation rather than presenting in pure form. Second, the BYI-2 anger and disruptive behaviour scales include the item content (e.g., feeling treated unfairly, irritability) that overlaps with the irritability and frustration components of adolescent depression captured by the CAPP-PRF PDD scale, which can inflate cross-domain correlations. The pattern is therefore consistent with the well-documented entanglement of irritability with internalizing distress in adolescent clinical samples rather than with a misallocation of indicators in the factor model. We further note that concurrent validity was assessed only in the clinic-referred sample, because the BYI-2 and WISC-V were administered only in that setting, and so the generalisability of these validity estimates to community adolescents remains to be established.

4. Discussion

This present study sought to identify the presence of a neurodevelopmental spectrum, and the appropriate locations for ADHD, ODD and CD in the HiTOP model in two independent samples of adolescents: a community sample, and a clinic-referred sample. In both samples, parents of adolescents completed the CAPP-PRF that provided scores for the presence or otherwise of 13 common internalizing disorders (GAD, SAD, PDD, OCD, PTSD, AN and BN), externalizing disorders (ADHD and ODD/CD), and neurodevelopmental disorders (SLD, ASD, LD and SSD). To answer our research questions, we tested and compared four structural models in both samples. For both samples, a three-factor model with spectra for neurodevelopment disorder, internalizing, and externalizing showed adequate, and better fit than a one-factor model with all the disorders loading on a single factor; a two-factor model, with neurodevelopmental and externalizing disorders, and internalizing disorders loading onto two separate factors; and a different three-factor model, with ADHD in both the externalizing spectrum and the neurodevelopmental spectrum. There was also support for the concurrent validity of the factors in this model for the clinic-referred sample. However, when considering the externalizing spectrum in this model, it may be worth keeping in mind that it showed poor internal consistency reliability (omega coefficient). As omega coefficients are related to the number of items or indicators in the specific scale, we believe that the poor internal consistency reliability found for the externalizing spectrum may be because it had only two indicators (ADHD and ODD/CD) for it. Two further considerations are likely to have contributed to this. First, ODD and CD were combined into a single ODD/CD indicator owing to the very low number of positive ODD screens (N = 3 in the community sample), which is itself a constraint on the breadth of the externalizing factor and may have obscured potentially distinct relationships of ODD and CD with the other modelled disorders. Second, dichotomising the symptom scales into positive/negative screens reduces the variance available to each indicator, which tends to depress reliability estimates for factors defined by few indicators. The omega value in the community sample (0.46) is below the conventional 0.70 threshold and the clinic-referred value (0.69) is just below it, so conclusions about the placement of ADHD and ODD/CD on the externalizing factor should be regarded as provisional and as based on factor structure and pattern of associations rather than on a strongly internally consistent externalizing composite. Replication using continuous symptom severity scores and a broader set of externalizing indicators (e.g., substance use, antisocial behaviour) will be important for testing the stability of these placements. Notwithstanding this, externalizing has consistently emerged as a separate and robust higher order factor in models of psychopathology in children, adolescents and adults [2,15,40].

Although our support for our three-factor model with a neurodevelopmental spectrum is consistent with the findings reported by Michelini et al. [15], our findings provide a more comprehensive understanding of the neurodevelopmental spectrum. More specifically, the neurodevelopmental spectrum for children reported by Michelini et al. [15] showed that it comprised primarily inattention (IA) and hyperactivity/impulsivity (HI), and that for adults (which they referred to as inattentive neurodevelopmental) comprised primarily IA. Thus, these spectra may be specific to ADHD. In contrast, in the current study and unlike Michelini et al.’s [15] study, the neurodevelopmental spectrum comprised SLD, ASD, LD, and SSD. Additionally, ADHD together with ODD/CD was in a separate factor, reflecting externalizing disorders. Also, a model in which ADHD cross-loaded on the factors for the neurodevelopment and externalizing failed to converge. Our findings therefore suggest that, at the dimensional level tested here, ADHD aligns more closely with the externalizing factor than with the neurodevelopmental spectrum, as currently proposed in the HITOP model. In this context, our findings also differed from how it is grouped in DSM-5-TR, i.e., with other neurodevelopmental disorders. Considering this, it can be speculated that even if we accept ADHD is primarily a neurodevelopmental disorder, its phenotypic expression from a dimensional viewpoint is one that aligns more with those disorders in the externalizing spectrum. This view concurs with Andrews et al.’s [16] conclusion that although ADHD exhibits some associations with a neurodevelopmental disorder cluster, it has stronger associations with an externalizing disorder cluster. Another noteworthy finding in this study is that, despite raising the possibility that ODD and CD could potentially be better located in the neurodevelopment spectrum rather than the externalizing spectrum [17,18], this was not supported. Our finding showed that they are better located with ADHD in the externalizing spectrum. Our results also align with, and extend, two recent adolescent-focused HiTOP studies. Forbes et al. [19], using a hierarchical symptom-level model in children and adolescents, recovered four higher-order domains (internalizing, externalizing, eating pathology, and uncontrollable worries/obsessions/compulsions) and placed ADHD-related content under externalizing. Hankin et al. [21] similarly identified internalizing, disinhibited externalizing, and personality dysregulation as intermediate factors in youth, with ADHD-related disinhibition again sitting on the externalizing side rather than constituting a separate neurodevelopmental domain. Neither of those studies, however, included the breadth of neurodevelopmental disorders modelled here (SLD, ASD, LD, SSD), so the present results add a complementary observation: even when an explicit neurodevelopmental factor is permitted in the model and is populated by disorders other than ADHD, ADHD and ODD/CD continue to associate more strongly with externalizing. This pattern is broadly consistent with how ICD-11 has retained ADHD within neurodevelopmental disorders while acknowledging substantial overlap with disruptive behaviour disorders, and it suggests that future revisions of HiTOP, DSM-5-TR, and ICD-11 may benefit from explicitly representing this dual nature of ADHD, as categorically neurodevelopmental in origin but dimensionally aligned with externalizing in its expressed symptom profile, rather than forcing a single placement.

Summary and Study Limitations

Despite the new and interesting findings of this study, there are limitations in the current study that need to be considered when viewing the findings and conclusions made in the study. First, although the current HiTOP model has six spectra (i.e., internalizing, thought disorder, detachment, somatoform, disinhibited externalizing, and antagonistic externalizing problems), the limitations in our data set did not allow us to include spectra for thought disorder, detachment, and somatoform, and therefore the complete HiTOP model. Also, we modelled a comprised externalizing spectrum. Second, although power analysis showed the sample size in the clinic-referred sample (N = 173) was adequate, some may consider it too low, thereby questioning the concurrent validity findings report for this sample. Third, the study did not include a wider range of external variables that could have allowed for a more comprehensive evaluation of the concurrent validity of the spectra in our preferred three-factor model. Also, this was examined only for the clinic-referred sample. Fourth, as the data for the current study was based on the CAPP-PRF and for a group of adolescents, the findings are most applicable to parent ratings of adolescents, and therefore not applicable to teacher ratings and self-ratings of adolescent or other age groups. Fifth, all data for the community sample were collected via online from all around Australia. Thus, it is uncertain if respondents gave appropriate consideration to completing their responses independently and attentively. Sixth, all data collected for the clinic-referred sample came from referrals made to the same clinic. Thus, referral filter bias (bias related to examining a sample completely from the same clinic) may have confounded the findings. Seventh, males were over-represented in both samples (approximately 60% in the community sample and 65% in the clinic-referred sample). Given well-documented gender differences in the prevalence and symptom presentation of several of the modelled disorders, including ADHD, ASD, internalizing disorders, and eating disorders, this skew could have influenced both the strength of factor loadings and the inter-factor correlations, and the findings should be replicated in samples with a more balanced gender distribution and, ideally, with sex-stratified or measurement-invariance analyses. Eighth, the design is cross-sectional, so the present findings speak to the dimensional structure of psychopathology at a single point in adolescence and not to its developmental stability or to the direction of associations among spectra over time; longitudinal HiTOP studies are needed to clarify whether the placement of ADHD, ODD and CD on the externalizing factor remains stable across adolescent development. Ninth, although the present study added four neurodevelopmental disorders beyond ADHD (SLD, ASD, LD, SSD), other neurodevelopmental conditions central to DSM-5-TR, in particular intellectual disability, tic disorders, and developmental coordination disorder, were not modelled because the CAPP-PRF does not screen for them; this limits the comprehensiveness of the neurodevelopmental spectrum recovered here and future studies should aim to model a broader set of neurodevelopmental indicators. In terms of clinical implications, despite these limitations, the present findings have practical utility for adolescent assessment. They suggest that, alongside categorical DSM-5-TR diagnoses, clinicians can usefully formulate adolescent presentations along three-dimensional spectra, neurodevelopmental, internalizing, and externalizing, which can guide transdiagnostic case formulation, prioritize multi-domain assessment in adolescents presenting with ADHD or ODD/CD, and inform the selection of broad-band rather than disorder-specific interventions for individuals whose presentations cut across DSM-5-TR boundaries. Tenth, although all factor correlations in our final model were below 0.80, meeting discriminant validity criteria, their magnitudes were relatively high. Indeed, the correlation between the internalizing and externalizing factors reached 0.722 in the community sample. Considering this, there is a high possibility that although our intention was to recruit a community sample, we recruited a sample that was highly psychopathological and not a community sample. The community sample was recruited online exclusively by a psychological clinic, and this fact was clearly displayed on the website. It is therefore possible that knowledge of this may have not only biassed relatively more participation by individuals who suspected that they were experiencing psychology problems (i.e., sponsor bias), but also how they rated themselves—generally overrating their psychological problems (response bias; [41]). This is further supported by the fact that the levels of psychopathology in this group were comparable or even higher than the clinical sample deliberately recruited in this study. Given this, it can be argued that our findings and interpretations referring to the community sample may be better interpreted as also referring to a clinical sample.

5. Summary and Conclusions

In summary, our findings indicate support for a neurodevelopmental spectrum in the HiTOP model, with ADHD and ODD/CD loading on the externalizing spectrum rather than the neurodevelopmental spectrum. We demonstrate this in two independent community and clinic-referred samples of adolescents. Our findings are consistent and extend the current HiTOP model. At a more general level, we demonstrated the applicability of the HiTOP in adolescents. Given that existing studies have already demonstrated the applicability of a comparable HiTOP model in adults [2,5] and children [15], the present results are consistent with prior findings in children and adults, providing further evidence for the applicability of the HiTOP model across different age groups. Whether the model is broadly robust across the lifespan will require additional longitudinal and cross-age studies. Overall, our findings support the view that psychopathology can be represented dimensionally [42]. It is therefore encouraging that there has been a shift from categorical diagnosis of mental disorders to precision diagnosis involving psychopathology dimensions. Indeed, the revised DSM and ICD (i.e., DSM-5-TR and ICD-11) have embraced to some degree dimensional perspectives for some disorders, such as for personality disorders in DSM-5-TR [43,44]. Indeed, it could be speculated that the dimensional approach to classification could be considered a viable alternative taxonomy for classification for psychological disorders.