Introduction
The controversy of whether attention deficit hyperactivity disorder (ADHD) should be regarded as a category or as a dimension has been a key issue in the literature for over a decade (Carson, Reference Carson1991; Clark et al. Reference Clark, Watson and Reynolds1995). A categorical view proposes that ADHD differs qualitatively from variation in the normal range, with the discontinuity in underlying risk processes around the diagnostic boundary. The dimensional view, by contrast, sees ADHD differing from normality only in degree, with a spectrum of severity expressing continuity in risk–disorder relationships (Coghill & Sonuga-Barke, Reference Coghill and Sonuga-Barke2012). In the former, diagnostic thresholds are conceptualized to reflect ‘natural’ boundaries linked to underlying causes; in the latter, they reflect clinical and societal burden. Although commonsense may indicate for both clinicians and researchers that ADHD might be best conceptualized as a dimension rather than a category, scientific evidence supporting commonsense reasoning has been restricted to behavioral genetic (Levy et al. Reference Levy, Hay, McStephen, Wood and Waldman1997; Polderman et al. Reference Polderman, Derks, Hudziak, Verhulst, Posthuma and Boomsma2007; Larsson et al. Reference Larsson, Anckarsater, Rastam, Chang and Lichtenstein2012), taxonomic (Haslam et al. Reference Haslam, Williams, Prior, Haslam, Graetz and Sawyer2006; Frazier et al. Reference Frazier, Youngstrom and Naugle2007; Marcus & Barry, Reference Marcus and Barry2011; Marcus et al. Reference Marcus, Norris and Coccaro2012) and neuroimaging (Shaw et al. Reference Shaw, Gilliam, Liverpool, Weddle, Malek, Sharp, Greenstein, Evans, Rapoport and Giedd2011) approaches. Studies investigating this issue from a neurocognitive perspective are lacking.
ADHD children differ from typically developing peers in a variety of cognitive domains (Sonuga-Barke, Reference Sonuga-Barke2005). Deficits in inhibitory-based executive function (IB-EF; Barkley, Reference Barkley1997; Quay, Reference Quay1997; Wood et al. Reference Wood, Buitelaar, Rijsdijk, Asherson and Kuntsi2010) and basic information processing (BIP; Sergeant, Reference Sergeant2000; Castellanos et al. Reference Castellanos, Sonuga-Barke, Scheres, Di Martino, Hyde and Walters2005; Rommelse et al. Reference Rommelse, Altink, de Sonneville, Buschgens, Buitelaar, Oosterlaan and Sergeant2007; Bitsakou et al. Reference Bitsakou, Psychogiou, Thompson and Sonuga-Barke2008; Mulder et al. Reference Mulder, Bos, Weusten, van Belle, van Dijk, Simen, van Engeland and Durston2010) are particularly consistent among studies and are considered central to ADHD neuropsychology (Barkley, Reference Barkley1997; Quay, Reference Quay1997; Sergeant et al. Reference Sergeant, Geurts, Huijbregts, Scheres and Oosterlaan2003; Nigg et al. Reference Nigg, Willcutt, Doyle and Sonuga-Barke2005; Sonuga-Barke, Reference Sonuga-Barke2005; Wood et al. Reference Wood, Buitelaar, Rijsdijk, Asherson and Kuntsi2010; Killeen et al. Reference Killeen, Russell and Sergeant2013). BIP refers to a set of lower-order cognitive processes that are involved in processing very basic information involved in almost all neurocognitive tasks such as encoding and efficiency of stimulus transmission and motor function (Parisi, Reference Parisi1997; Sergeant, Reference Sergeant2000). IB-EF reflects top-down cognitive processes linked to the ability to inhibit an inappropriate pre-potent or dominant response in favor of a more appropriate alternative (Barkley, Reference Barkley1997). BIP and IB-EF are frequently measured using forced-choice reaction time (RT) paradigms requiring a quick decision in which both accuracy and response speed are measured. In these tests, children with ADHD often respond less accurately, more slowly and have more variable RTs than their peers (Kuntsi & Klein, Reference Kuntsi and Klein2012). However, the psychological significance of these differences are obscured by the fact that errors and RTs on these tasks are not independent and can reflect multiple interactive processes, such as stimulus encoding, processing efficiency, motor preparation and output and speed–accuracy trade-off (Ratcliff & McKoon, Reference Ratcliff and McKoon1988). Sequential sampling statistical models such as diffusion models (DMs) can differentiate between these different elements by simultaneously analyzing RTs and accuracy over time. In the current study, we extracted such DM parameter estimates to provide more interpretable assessments of BIP and IB-EF.
Previous DM analyses have found that subjects with ADHD are less efficient in basic processing efficiency, that is in the DM parameter drift rate (Huang-Pollock et al. Reference Huang-Pollock, Karalunas, Tam and Moore2012; Karalunas et al. Reference Karalunas, Huang-Pollock and Nigg2012; Karalunas & Huang-Pollock, Reference Karalunas and Huang-Pollock2013). Indeed, some studies suggest that executive deficits in ADHD may be totally attributable to underlying BIP deficits (Karalunas & Huang-Pollock, Reference Karalunas and Huang-Pollock2013). Concurring with these findings, a previous study from our group showed that children with ADHD have deficits in encoding/motor-function and decision-making processes (Salum et al. Reference Salum, Sergeant, Sonuga-Barke, Vandekerckhove, Gadelha, Pan, Moriyama, Graeff-Martins, de Alvarenga, do Rosario, Manfro, Polanczyk and Rohde2014).
We therefore used the documented association between BIP deficits and ADHD subjects from our previous study to formally test competing models of ADHD (e.g. dimensional and categorical). We built on this literature to examine the presence of BIP and IB-EF across the clinical and subclinical parts of the ADHD spectrum and to test between categorical and dimensional models of the neuropsychological basis of ADHD. The categorical model predicts a discontinuity in the pattern of associations between symptom severity and cognitive impairment around the diagnostic boundary, that is deficits would be present in full ADHD cases and absent in sub-boundary cases. Therefore, we would expect a non-linear relationship between increases in symptom severity and the scale of cognitive deficits. By contrast, the dimensional model predicts no such discontinuity, with cognitive deficits present in subclinical and in clinical cases to a degree proportionate to the severity of the ADHD presentation. Thus, the symptom severity and cognitive deficit association is predicted to take a linear form.
To test these different predictions, we selected different samples of children not meeting a full ADHD diagnosis but with different levels of symptoms (asymptomatic, subthreshold minimal and subthreshold moderate) and compared these with a group of children with ADHD defined by DSM. Using this approach, we were able to test between the categorical and the dimensional models relating ADHD to BIP and IB-EF. We adopted a group-based approach rather than exploring associations across ADHD symptom levels as a continuous trait because we wanted to maintain the clinical relevance of the analysis by including a group that met the full diagnostic criteria. As currently conceptualized, ADHD is not defined simply as high levels of ADHD symptoms, but requires synthetic diagnostic rules to be met in relation to different dimensions of ADHD, age of onset restrictions and also related impairment.
In keeping with the previous literature supporting the dimensional model, our main predictions were that: (i) deficits in neurocognitive functions would be observed at subclinical and full clinical levels of ADHD and (ii) deficits would increase in a linear way across groups with increasingly severe presentations of ADHD. We selected groups excluding children with other psychiatric disorders and exposure to medication. Medication and co-morbidity have proven to affect both BIP and IB-EF in previous studies (Oosterlaan et al. Reference Oosterlaan, Logan and Sergeant1998, Reference Oosterlaan, Scheres and Sergeant2005; Rhodes et al. Reference Rhodes, Coghill and Matthews2006, Reference Rhodes, Park, Seth and Coghill2012; Coghill et al. Reference Coghill, Rhodes and Matthews2007; Chamberlain et al. Reference Chamberlain, Robbins, Winder-Rhodes, Muller, Sahakian, Blackwell and Barnett2011; Swanson et al. Reference Swanson, Baler and Volkow2011; Wang et al. Reference Wang, Volkow, Wigal, Kollins, Newcorn, Telang, Logan, Jayne, Wong, Han, Fowler, Zhu and Swanson2013). Therefore, this sampling strategy provided a strong design for testing this specific hypothesis. It decreases the possibility that a categorical nature would be revealed by factors influencing cognition in ways not specifically linked to ADHD symptom severity.
Method
Participants
The sample is part of a large community school-based study of children aged 6 to 12 years from 57 schools in two Brazilian cities: Porto Alegre (n = 22) and São Paulo (n = 35). The ethics committee of the University of São Paulo approved the study. We obtained written consent from parents of all participants, and verbal assent from all children.
During the screening phase at their school registry day, 9937 informants were interviewed using the Family History Survey (FHS; Weissman et al. Reference Weissman, Wickramaratne, Adams, Wolk, Verdeli and Olfson2000). From this pool, we recruited two subgroups using a random-selection (n = 958) and high-risk group selection procedure (n = 1514), resulting in a total sample of 2512 subjects (for further details, see Salum et al. Reference Salum, Mogg, Bradley, Gadelha, Pan, Tamanaha, Moriyama, Graeff-Martins, Jarros, Polanczyk, do Rosario, Leibenluft, Rohde, Manfro and Pine2013). From these 2512 subjects, 2002 (74%) had data available for the tasks used in this study. Subjects who did not perform the tasks did not differ from those who did regarding psychopathology (all p > 0.05; data available upon request). A total of 144 subjects were excluded for poor task compliance, but again no differences were detected between them and the remaining sample regarding psychopathology (all p > 0.05; data not shown).
The aim of the current study was to select from the remaining sample (n = 1858) specific groups of participants with different levels of symptom severity across the spectrum of ADHD symptoms (from non-clinical to clinical) but without co-morbidity with other psychiatric disorders. The ADHD section of the Development and Well-Being Assessment (DAWBA; Goodman et al. Reference Goodman, Ford, Richards, Gatward and Meltzer2000) (no skipping rules) was used to assess inattention and hyperactivity/impulsivity in the total sample. Each of the 18 ADHD symptoms has a three-option response scale ‘No more than other’, ‘A little more than others’ and ‘A lot more than others’, representing a score of 0, 1 and 2 respectively. Groups were constructed based on similar symptom cut-offs suggested by previous studies exploring subthreshold ADHD (Faraone et al. Reference Faraone, Biederman, Doyle, Murray, Petty, Adamson and Seidman2006; Shaw et al. Reference Shaw, Gilliam, Liverpool, Weddle, Malek, Sharp, Greenstein, Evans, Rapoport and Giedd2011). Therefore, we selected overlapping hierarchically defined sets of groups as follows.
Inattention
(1) Asymptomatic: randomly selected subjects, scoring 0 on inattentive symptoms; (2) subthreshold minimal symptoms: subjects scoring from 1 to 5 on the inattentive symptoms scale (maximum of two full ADHD symptoms); (3) subthreshold moderate symptoms: subjects scoring from 6 to 11 on inattentive symptoms (maximum of five full ADHD symptoms in typically developing children); (4) predominantly inattentive ADHD subtype (n = 36) or a combined ADHD subtype (n = 17) (full ADHD DSM-IV diagnoses including impairment).
Hyperactivity/impulsivity
(1) Asymptomatic: randomly selected subjects, scoring 0 on the hyperactive/impulsive symptom scale; (2) subthreshold minimal symptoms: subjects scoring from 1 to 5 on the hyperactive/impulsive symptom scale (maximum of two full ADHD symptoms); (3) subthreshold moderate symptoms: subjects scoring from 6 to 11 on hyperactive/impulsive symptom scale (maximum of five full ADHD symptoms in typically developing children); (4) predominantly hyperactive/impulsive ADHD subtype (n = 15) or a combined ADHD subtype (n = 17) (full ADHD DSM-IV diagnoses). Inattentive and hyperactive/impulsive groups were created independently, and hyperactivity/impulsivity was allowed to vary freely within the inattentive group, and vice versa.
We excluded from all analyses participants who had ever received any psychiatric medication with or without a psychiatric diagnosis (n = 74, 4%; of whom 36 had ADHD), with an IQ below 70 (n = 36; 1.8%), co-morbid conduct/oppositional disorder (n = 127; 6.8%), anxiety (n = 100; 5.4%), depression (n = 64; 3.5%), mania (n = 3; 0.2%), psychoses (n = 1; 0.1%), pervasive developmental (n = 9; 0.6%), tics (n = 15; 0.8%) and eating (n = 8; 0.5%) disorders.
Psychiatric diagnosis
Our diagnosis was based on the DAWBA (Goodman et al. Reference Goodman, Ford, Richards, Gatward and Meltzer2000), a structured interview administered to biological parents by trained lay-interviewers. The DAWBA is a reliable and widely used psychiatric instrument translated into 22 languages. A comparison of levels of agreement for ADHD between the DAWBA and the Child and Adolescent Psychiatric Assessment (CAPA) and between the DAWBA and the Diagnostic Interview Schedule for Children (DISC) resulted in moderate agreement (0.49 and 0.57 respectively), similar to the κ value of 0.52 obtained by Angold et al. (Reference Angold, Erkanli, Copeland, Goodman, Fisher and Costello2012) when comparing CAPA and DISC. The DAWBA is well suited for epidemiological research, given that it is shorter than alternative interviews. Nine psychiatrists performed the rating procedures. All were trained and supervised by a senior child psychiatrist. A second child psychiatrist rated a total of 200 interviews and the between-rater κ values for ADHD was high (0.72).
Neurocognitive tasks
Two tasks were used to assess BIP and IB-EF: a simple two-choice reaction time task (2C-RT) and a conflict control task (CCT; Hogan et al. Reference Hogan, Vargha-Khadem, Kirkham and Baldeweg2005). The 2C-RT measures the ability of the participant to perform basic perceptual decisions about the direction an arrow on the screen is pointing. It has minimal executive component. A total of 100 arrow stimuli were presented, half requiring a left and half requiring a right button press. The CCT builds on the 2C-RT and includes a second inhibitory executive component requiring participants to occasionally suppress a dominant tendency to respond to the actual direction of an arrow and to initiate a response indicating the opposite direction. This requirement was indicated by a change in the color of the arrow (a ‘conflict’ effect). There were 75 congruent trials with green arrows (participants had to press the button indicating the actual direction of the arrow) and 25 incongruent trials, when red arrows were presented (and participants had to respond in the opposite direction to that indicated by the arrows presented). These two tasks were used to derive BIP variables using DMs (2C-RT and CCT) and IB-EF measured in the context of BIP deficits.
Deriving indices of BIP
BIP variables were derived from DM analysis (Ratcliff & McKoon, Reference Ratcliff and McKoon1988; White et al. Reference White, Ratcliff, Vasey and McKoon2010) from both 2C-RT and CCT data. The parameters of this model provide information about: processing efficiency [drift rate (v), the rate at which an individual is able to acquire information to make a forced choice response; that is the neural signal-to-noise ratio], speed–accuracy trade-off [boundary separation (a), response caution or impulsive response style] and encoding/motor function [non-decision time (Ter), the time it takes to complete all other information processes not involved in stimulus discrimination] (Huang-Pollock et al. Reference Huang-Pollock, Karalunas, Tam and Moore2012). Both processing efficiency and encoding/motor function fluctuate from trial to trial in the course of the experiment, also providing parameters of BIP variability (Q and e respectively).
Data points in the DM analysis that are outliers (outside the range of normal observations) or likely to represent contaminants (random guesses and fast guesses) were detected and filtered out using exponentially weighted moving average (EWMA). Further details about these methods can be found in Vandekerckhove & Tuerlinckx (Reference Vandekerckhove and Tuerlinckx2007). A total of 36 participants (1.6%) had more than 50% of missing/outlier responses and were excluded from the analysis. In addition, 108 (4.3%) were excluded because of failures in estimating the parameters with the Markov chain Monte Carlo method (Vandekerckhove & Tuerlinckx, Reference Vandekerckhove and Tuerlinckx2007). There were no between-group differences in the number of non-included subjects in the analyses (all p values > 0.05).
Estimating IB-EF (taking account of BIP)
IB-EF was estimated by subtracting mean drift rates (i.e. processing efficiency) of incongruent trials from mean drift rates of congruent trials (White et al. Reference White, Ratcliff, Vasey and McKoon2010). This model assumes that, on the incongruent trials of CCT, there is a ‘conflict effect’; that is an initial accumulation of evidence towards the wrong boundary (the conflict effect), which is followed for an accumulation of evidence towards the correct boundary. Congruent trials do not have conflict and therefore the subtraction (incongruent minus congruent) provides a way of estimating the IB-EF more reliably and taking account of BIP (i.e. above and beyond deficits in BIP).
Statistical analysis
To investigate our first hypothesis, we performed a series of MANCOVAs using BIP DM variables derived from 2C-RT and CCT. The results from the MANCOVAs and IB-EF derived measures from the CCT were further decomposed with ANCOVAs and between-group differences were analyzed using two specific contrasts: (1) differences from asymptomatic individuals and (2) differences from clinically defined ADHD subjects. All models controlled for site, gender, age and IQ. We tested linear, quadratic and cubic trends among the means of the four groups in the ANCOVAs for all BIP and IB-EF parameters using polynomial contrasts to test our second hypothesis. Groups were formed to have a very similar range of increments of symptoms from the asymptomatic group.
All variables were z transformed and normalized using the van den Waerden transformation (Lehmann, Reference Lehmann1975). Effect sizes were defined in terms of percentage of explained variance and 1, 9 and 25% were defined respectively as small, medium and large effects corresponding to 0.01, 0.06 and 0.14 partial eta square (η p 2) values (Cohen, Reference Cohen1988). DM analysis was implemented using hierarchical DMs for two-choice response times. Details about the estimation of DM outcome variables can be found elsewhere (Vandekerckhove et al. Reference Vandekerckhove, Tuerlinckx and Lee2011). All tests were two-tailed, with an α value of 0.05.
Results
Sample characteristics are given in Table 1. Groups did not differ in terms of gender (all p values > 0.05), and age differences were minimal. Analyses for inattentive and hyperactive ADHD symptoms are depicted in Tables 2 and 3 respectively.
Sample description and classical task measures
ADHD, Attention deficit hyperactivity disorder; IQ, intelligence quotient; SES, socio-economic status; 2C-RT, two-choice reaction time task; CCT, conflict control task; s.d., standard deviation.
Outliers defined as RTs < 100 ms or 3 s.d.s above the subject's mean (for descriptive purposes).
Post-hoc ANCOVAs showing differences between groups of inattention in diffusion model (DM) parameters for the two-choice reaction time task (2C-RT) and the conflict control task (CCT)
ADHD, Attention deficit hyperactivity disorder (predominantly inattentive or combined subtypes); BIP, basic information processing; IB-EF, inhibitory-based executive function; s.e., standard error; Asym, asymptomatic; Min, minimal; Mod, moderate; Q, trial-to-trial variability in Ter; Ter, mean non-decision time (encoding/motor function); a, boundary separation (speed/accuracy trade-off); e, trial-to-trial variability in v; v, mean drift rate (processing efficiency); (c), congruent trial; (i) incongruent trial.
MANCOVAs: BIP (2C-RT) F 15,3246 = 3.07/p < 0.001/η p 2 = 0.014; BIP (CCT) F 15,3246 = 2.63/p = 0.001/η p 2 = 0.012.
Polynomial contrasts: differences between asymptomatic and non-clinical groups are in italics.
Estimated marginal means for z scores (corrected for age and IQ).
Analysis repeated only in the randomly selected subjects resulted in very similar results.
Bold values indicate statistically significant results.
Post-hoc ANCOVAs showing differences between groups of hyperactivity/impulsivity in diffusion model (DM) parameters for the two-choice reaction time task (2C-RT) and the conflict control task (CCT)
ADHD, Attention deficit hyperactivity disorder (predominantly hyperactive/impulsive or combined subtypes); IB-EF, inhibitory-based executive function; BIP, basic information processing; s.e., standard error; Asym, asymptomatic; Min, minimal; Mod, moderate; Q, trial-to-trial variability in Ter; Ter, mean non-decision time (encoding/motor function); a, boundary separation (speed/accuracy trade-off); e, trial-to-trial variability in v; v, mean drift rate (processing efficiency); (c), congruent trial; (i) incongruent trial.
MANCOVAs: BIP (2C-RT) F 15,3126 = 2.07/p = 0.009/η p 2 = 0.010; BIP (CCT) F 15,3126 = 1.56/p = 0.077/η p 2 = 0.007.
Polynomial contrasts: differences between asymptomatic and non-clinical groups are in italics.
Estimated marginal means for z scores (corrected for age and IQ).
Analysis repeated only in the randomly selected subjects resulted in very similar results.
Bold values indicate statistically significant results.
For inattention, we found diminished processing efficiency (v) in children with minimal inattentive symptoms in both tasks compared to asymptomatic children. In addition to lower processing efficiency, those with moderate levels of inattention also had a higher trial-to-trial variability in encoding/motor function (Q) in both tasks compared to asymptomatic participants. In addition to the deficits presented by the moderate group, full ADHD inattentive/combined participants also had higher variability in processing efficiency (E) and faster encoding/motor function (Ter) in the 2C-RT compared to asymptomatic individuals. For all between-group differences, we found a significant linear trend supporting a dose–response relationship between inattentive symptoms and BIP (Table 2, Fig. 1). No between-group differences in IB-EF were found (Table 2).
Between-group differences in mean drift rates (v) in the two-choice reaction time task (2C-RT) and the conflict control task (CCT) for inattention and hyperactivity/impulsivity symptoms. For the CCT, mean drift rates from congruent trials were used. Asym, asymptomatic; Min, minimal; Mod, moderate; ADHD, attention deficit hyperactivity disorder; s.e., standard error. a Significant differences from typically developing children (TDC). b Significant differences from ADHD.
With respect to hyperactivity/impulsivity, the minimal hyperactivity group had diminished processing efficiency (v) when compared to the asymptomatic group, a pattern similar to that seen in the moderate symptoms group and the group with full-blown ADHD in the 2C-RT task. Those with moderate symptoms and those with ADHD also had a faster encoding/motor function (Q) only in the 2C-RT. The overall MANCOVA for the CCT was not significant and therefore no further ANCOVAs for the CCT were performed (Table 3, Fig. 1). Again, all between-group differences had a significant linear trend and therefore support a dose–response relationship between hyperactive symptoms and BIP. No between-group differences were detected for IB-EF (Table 3).
Additional analyses revealed that our results were unaffected by controlling for subthreshold oppositional defiant disorder (ODD) symptoms based on the Child Behavior Checklist (CBCL; Krol et al. Reference Krol, De Bruyn, Coolen and van Aarle2006) (Supplementary Tables S1 and S2). In addition, in a supplementary linear model using ADHD as a continuous variable, contrasts revealed that, for all significant associations between ADHD and neurocognitive variables, a linear trend could be detected and no other higher-order link function was statistically significant (Supplementary Table S3). An exception is the encoding/motor function (Ter) in the 2C-RT, where no trend was detected.
Discussion
In this study, we were able to demonstrate that deficits in information processing were found in subjects with minimal, moderate and clinical ADHD for both inattention (in both tasks) and hyperactivity/impulsive dimensions (in the 2C-RT). No significant associations emerged for IB-EF. Crucially, these trends followed a linear function and therefore there was no evidence for a discontinuity in the neurocognitive association around the categorical boundary. The current study provides evidence that deficient processing efficiency is a key element in ADHD neuropsychology as the syndrome meeting the full criteria and is also strongly implicated in both inattention and hyperactivity problems even at minimal levels. These results therefore demonstrate that neuropsychological deficits operate at several levels of the ADHD symptom spectrum and provide neurocognitive support for a dimensional model of ADHD.
We have demonstrated the value of the DM perspective for understanding the origins of the inaccurate, slow and variable performance in ADHD subjects, which indicates impairments in the basic processing of information, particularly processing efficiency. Our results are in agreement with a growing body of literature indicating that differences in inhibitory function and other executive measures become non-significant after controlling for baseline measures of BIP (Rommelse et al. Reference Rommelse, Altink, de Sonneville, Buschgens, Buitelaar, Oosterlaan and Sergeant2007; Karalunas & Huang-Pollock, Reference Karalunas and Huang-Pollock2013), or following the introduction of incentives (Konrad et al. Reference Konrad, Gauggel, Manz and Scholl2000; Slusarek et al. Reference Slusarek, Velling, Bunk and Eggers2001; Kuntsi et al. Reference Kuntsi, Wood, van der Meere and Asherson2009), and contradict inhibitory theories (Barkley, Reference Barkley1997; Quay, Reference Quay1997; Wood et al. Reference Wood, Buitelaar, Rijsdijk, Asherson and Kuntsi2010). Given empirical evidence supporting the association between ADHD and BIP, we postulate that these neurocognitive processes are indicators of underlying ADHD neural dysfunction. Our results suggest that neural dysfunctions related to basic processing deficits are present across the ADHD spectrum even where only minimal levels of symptoms are present.
Other evidence for dimensionality arises from several sources. Psychometric studies have used several statistical techniques such as latent class analysis (Hudziak et al. Reference Hudziak, Heath, Madden, Reich, Bucholz, Slutske, Bierut, Neuman and Todd1998; Neuman et al. Reference Neuman, Todd, Heath, Reich, Hudziak, Bucholz, Madden, Begleiter, Porjesz, Kuperman, Hesselbrock and Reich1999), factor mixture models (Lubke et al. Reference Lubke, Hudziak, Derks, van Bijsterveldt and Boomsma2009; Ranby et al. Reference Ranby, Boynton, Kollins, McClernon, Yang and Fuemmeler2012) and different taxonomic approaches (Haslam et al. Reference Haslam, Williams, Prior, Haslam, Graetz and Sawyer2006; Frazier et al. Reference Frazier, Youngstrom and Naugle2007; Marcus & Barry, Reference Marcus and Barry2011; Marcus et al. Reference Marcus, Norris and Coccaro2012). Behavioral genetics studies also support dimensionality (Levy et al. Reference Levy, Hay, McStephen, Wood and Waldman1997; Polderman et al. Reference Polderman, Derks, Hudziak, Verhulst, Posthuma and Boomsma2007; Larsson et al. Reference Larsson, Anckarsater, Rastam, Chang and Lichtenstein2012). Our results are in agreement with other studies that directly investigated etiological and pathophysiological markers of ADHD. Evidence from epidemiological studies (Scahill et al. Reference Scahill, Schwab-Stone, Merikangas, Leckman, Zhang and Kasl1999), structural neuroimaging studies (Shaw et al. Reference Shaw, Gilliam, Liverpool, Weddle, Malek, Sharp, Greenstein, Evans, Rapoport and Giedd2011), clinical trials (Rapoport et al. Reference Rapoport, Buchsbaum, Weingartner, Zahn, Ludlow and Mikkelsen1980; Surman et al. Reference Surman, Hammerness, Petty, Doyle, Chu, Gebhard, Williams and Biederman2010) and personality traits (Cho et al. Reference Cho, Kim, Kim, Rohde, Hwang, Chungh, Shin, Lyoo, Go, Lee and Kim2009; Faraone et al. Reference Faraone, Kunwar, Adamson and Biederman2009) suggest similar patterns of effects between subthreshold and clinical cases.
This evidence that ADHD is best seen as a dimension rather than a category has potentially important clinical implications. Of note, we underscore the implications to our understanding of the etiology of ADHD. Dimensional phenotypes cannot arise from a single dichotomous causal factor and are most typically the result of an interaction of multiple etiological factors (Haslam, Reference Haslam1997). In addition, the clinical dilemma relating to where to put the threshold designating the categorical diagnosis (Matte et al. Reference Matte, Rohde and Grevet2012) is inherent to a dimensional approach. Pragmatically, we will still need practical decision rules for clinical purposes and the ‘thresholds’ decision will need to be addressed for ADHD, as in hypertension and levels of cholesterol. Therefore, focusing on defining these ‘thresholds’ is a crucial step for us to better stratify risk and start doing rational stepped care.
Our study has some limitations. First, other scales, such as the Strengths and Weaknesses of ADHD Symptoms and Normal Behavior (SWAN), that have a more appropriate normal distribution of its scores in the population could have been more sensitive to between-group differences in inattention and hyperactivity (Polderman et al. Reference Polderman, Derks, Hudziak, Verhulst, Posthuma and Boomsma2007). Second, our analyses were limited to the evaluation of BIP and IB-EF and the results do not necessarily imply that ADHD is a continuous disorder per se. Other neurocognitive domains, such as temporal processing and delay aversion, might be associated with a discontinuity within the ADHD spectrum. Nevertheless, our study shows that, for those specific measures, there is a clear linear relationship with symptom severity. Third, because of sample size issues we were not able to investigate different levels of the spectrum inside the clinical group, and a post-hoc analysis revealed that our study might have been underpowered to detect between-group differences in some BIP variables and in IB-EF. Our study also has notable strengths. It has a strong design to account for the impact of co-morbid problems, medication profiles and referral bias. Furthermore, all the effects reported are above and beyond effects of age, gender, IQ and investigational site.
In conjunction with accumulating evidence, our findings suggest that research in neurobiology of ADHD may benefit from a change in focus from extreme group comparisons to dimensional designs (Frazier et al. Reference Frazier, Youngstrom and Naugle2007). This approach may even facilitate scientific discoveries on the neurobiology of inattention and/or hyperactivity/impulsivity problems.
Supplementary material
For supplementary material accompanying this paper visit http://dx.doi.org/10.1017/S0033291714000919.
Acknowledgments
We thank the children and families for their participation, which made this research possible; the other members of the high-risk cohort research team (E. C. Miguel, R. Affonseca-Bressan, M. C. do Rosário, A. C. Tamanaha, P. Pan, P. G. de Alvarenga and H. Brentani); the collaborators for the neuropsychological evaluation (B. S. Scarpato, S. L. R. do Valle and C. Araújo); R. Goodman for his research support regarding the DAWBA instrument procedures; and B. Fleitlich-Bilyk for her clinical supervision.
This work is supported by the National Institute of Developmental Psychiatry for Children and Adolescents, a science and technology institute funded by the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq; National Council for Scientific and Technological Development; grant no. 573974/2008-0) and the Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP; Research Support Foundation of the State of São Paulo; grant no. 2008/57896-8).
Declaration of Interest
G. A. Salum is in receipt of a CAPES/FAPERGS post-doctoral scholarship and declares no potential conflicts of interest. E. Sonuga-Barke is a member of an advisory board to Shire, Flynn Pharma, UCB Pharma and AstraZeneca. He has served as speaker and consultant for Shire and UCB Pharma. He receives/has received current/recent research support from Janssen Cilag, Shire, Qbtech and Flynn Pharma, and conference support from Shire. J. Sergeant is a member of an advisory board to Lilly and Shire. He has received research funding from Lilly, and speaker fees from Lilly, Janssen-Cilag, Novartis and Shire. J. Vandekerckhove declares no potential conflicts of interest. A. Gadelha and T. S. Moriyama are in receipt of CAPES Ph.D. scholarships and receive continuous medical education support from Astra Zeneca, Eli-Lilly and Janssen-Cilag. A. S. Graeff-Martins is in receipt of a CNPq post-doctoral fellowship and declares no potential conflicts of interest. G. G. Manfro receives research support from Brazilian government institutions (CNPQ, FAPERGS and FIPE-HCPA) and declares no potential conflicts of interest. G. V. Polanczyk has served as a speaker and/or consultant to Eli-Lilly, Novartis and Shire Pharmaceuticals, has developed educational material for Janssen-Cilag, and receives unrestricted research support from Novartis and CNPq. L. A. P. Rohde was on the speakers' bureau and/or acted as consultant for Eli-Lilly, Janssen-Cilag, Novartis and Shire in the past 3 years. He also received travel awards (air tickets and hotel costs) from Novartis and Janssen-Cilag in 2010 for taking part in two child psychiatric meetings. The ADHD and Juvenile Bipolar Disorder Outpatient Programs chaired by L. A. P. Rohde received unrestricted educational and research support from the pharmaceutical companies Abbott, Eli-Lilly, Janssen-Cilag, Novartis and Shire in the past 3 years.
