Attention deficit hyperactivity disorder (ADHD) affects around 3.4% of children globally [
] and is characterised by pervasive and impairing levels of inattention and/or hyperactivity/impulsivity [
]. DSM-5 diagnostic criteria stipulate that for a diagnosis of ADHD to be given, problems must be evident across multiple contexts. However, a sizeable proportion of children show symptoms in only one context (or according to one informant) and may be no less impaired than children showing symptoms across multiple contexts. Furthermore, some authors have argued that it may be clinically useful to conceptualise children showing behavioural problems in specific contexts as representing distinct phenotypes [
]. The extent to which children can be meaningfully distinguished on the basis of the contexts in which they display ADHD symptoms has; however, yet to be established. In this study we thus evaluated the criterion validity of context-based presentation classifications (e.g. presentation at ‘home only’, ‘school only’, ‘both home and school’). We tested whether children differing in contexts of symptom expression differ in patterns of context-specific and context-general risk factors and sequalae.
DSM-5 diagnostic criteria for ADHD state that for a diagnosis ‘several inattentive or hyperactive-impulsive symptoms are present in two or more settings (e.g., at home, school, or work; with friends or relatives; in other activities)’ [
]. To determine if symptoms are present across contexts necessitates collecting information from more than one informant. For elementary school-aged children, this effectively means parents and teachers who can provide information on behaviour in the home and at school respectively. The ‘cross-context’ requirement of DSM-5 creates a challenge, however, because parents and teachers frequently disagree on the severity of ADHD symptoms displayed by a child. In a meta-analysis of multi-informant studies, for example, the average correlation between parent and teacher- reported ADHD was only .43 for inattention and .42 for hyperactivity/impulsivity [
], with studies published since broadly replicating these figures [
]. The modest agreement between parents and teachers is not merely due to measurement error or informant biases (although both undoubtedly do contribute). Rather, evidence suggests that there are genuine differences in child behaviour across contexts/ in interaction with different informants. For example, informant unique perspectives on ADHD and related disruptive behaviour disorders show genetic influences, are stable over time, predict relevant outcomes, and can be mapped to differences in interactions with people who play different roles for the child in lab-based studies [
The most appropriate way to deal with contextual differences in ADHD symptoms in clinical practice, however, remains unclear. One suggestion is to conceptualise individuals with problems in different contexts according to different subtypes. Dirks et al. [
] argue that symptoms that occur in different contexts may constitute distinct phenotypes and that characterising these phenotypes has the potential to improve diagnosis and treatment. In this system, individuals with, for example, primarily school-based issues would be considered as a separate presentation from those with primarily home-based issues, who in turn would be considered a separate presentation from those with issues that spanned both contexts. ‘Subtyping’ of this kind is most likely to be clinically useful if individuals differing in the contexts in which their symptoms are expressed show distinct etiologies, prognoses, patterns of impairment, or treatment responses.
Evidence on the utility of distinguishing presentations on the basis of informant reports is, however, currently scant. A small number of studies have compared individuals with ‘pervasive’ ADHD, i.e. symptoms across multiple contexts to individuals who display problems in only a single context [
], with mixed results. While some studies have suggested that pervasive symptoms are associated with greater overall impairment [
], others have found no difference between individuals with pervasive versus situation-specific symptoms [
]. One of the more recent studies to compare individuals differing in symptom contexts examined predictors of home- versus school-based problems as reported by parents and teachers respectively [
]. They found that a home-based risk factor (parental stress) predicted parent-reported symptoms only. They could not, however, rule out the possibility that this association reflected the response style of the parent because parents provided data on both constructs. They also found that parent-reported severity of symptoms was associated with symptoms across both contexts whereas teacher-reported severity was not. This hinted at the possibility that home-expressed symptoms are indicative of greater overall severity of problems than school-expressed symptoms.
In addition to parental stress, there are other ‘context-specific’ and ‘context-general’ factors that merit exploration to understand whether behavioural expression across contexts may be related to different etiologies and impairments. In the home, for example, negative parenting practices such as harsh or inconsistent discipline have been identified as important risk factors for, as well as outcomes of, disruptive behaviour disorders such as ADHD, conduct disorder and oppositional defiant disorder [
]. Analogous transactions may occur in the school environment where, for example, relationships with teachers can be affected by but can also shape disruptive behaviour problems [
]. Whether context-specific risk factors explain context-specific ADHD symptom expression is, however, not yet known.
In this study, we test the possibility that situation-specific problems have context-specific correlates using a large community-based longitudinal study. For comparison, we include a ‘trait-like’ predictor of ADHD: low self-control [
], which as a ‘trait’ is by definition are assumed to be expressed across multiple contexts. Using parent- and teacher- and self-reported data from the Zurich Project on Social Development from Childhood to Adulthood [z-proso; Eisner and Ribeaud
] study, we use growth mixture modelling to first define subtypes of ADHD characterised by the contexts in which symptoms are evident and then assess whether these subtypes map to context-specific and context-general risk factors. This method allows the data to dictate categories defined by symptom trajectories over different contexts rather than imposing a priori classifications. Importantly, we use self-report measures of risk factors to ensure that any associations between risk factor and context do not simply reflect common rater bias. We focus not only on cross-sectional levels of symptoms, but patterns of symptom development over the elementary school years. This is based on past research suggesting considerable change in symptom levels over time is possible within individuals and that patterns of change meaningfully distinguish individuals [
]. We hypothesise that categories would emerge that represent unaffected individuals, individuals with home-specific presentation, individuals with school-specific presentation, and individuals with cross-context presentation. We also hypothesise that parenting would be particularly related to home-specific presentation and teacher relationships to school-specific presentation. However, we hypothesise that as a trait-like characteristic, self-control would not be differentially related to context-based presentation classifications.
In this study, we sought to establish whether developmental subtypes of ADHD could be meaningfully distinguished on the basis of the contexts in which symptoms were primarily expressed. This builds on the idea proposed in previous research that distinguishing symptom presentations by informant could have clinical value [
]. Teachers served as informants for behaviour at school while parents served as informants for behaviour at home. For inattention, we found that if symptom reports varied across contexts, this was usually due to a greater expression of inattention at school. There was some evidence that school-based symptoms were related to school but not home problems, providing criterion validity support for informant-based presentation classification. For hyperactivity/impulsivity, children who showed elevated symptoms tended to show more severe school-based symptoms. However, differences in informant reports did not map to context-based predictors in the expected manner i.e., with school-based problems being particularly related to ADHD symptoms reported by teachers; home-based problems to symptoms reported by parents, and trait-like predictors to symptoms reported by both informants.
Our approach involved using growth mixture models to summarise classes of individuals with similar developmental trajectories and inattention and hyperactivity/impulsivity symptoms. Using this method, the majority of children were found to have low levels of ADHD by both informants. However, a pattern of informant discrepancy emerged, whereby when elevated symptoms were reported, teachers generally reported higher levels than parents. Inattention and hyperactivity/impulsivity have shown differing patterns of results in terms of informant discrepancy. For example, Murray et al. [
] found in a previous study in the current sample that while teachers tend to report higher levels of inattention on average, parents are more likely to report higher levels of hyperactivity/impulsivity. There is also evidence that inattention and hyperactivity/impulsivity differ in terms of their developmental trajectories. Using a similar technique to the current study, for example, Arnold et al. [
] found that in the
Longitudinal Assessment of Manic Symptoms
sample, developmental trajectories of inattention were best summarised in terms of three trajectories, while hyperactivity/impulsivity was best characterised in terms of four. We, therefore, analysed the dimensions of inattention and hyperactivity/impulsivity; discussed in turn below.
For inattention, five classes were judged optimal in the growth mixture analyses, four of which evidenced elevated levels of symptoms at some point according to at least one informant. Five classes is a larger number than those generally identified for either inattention or hyperactivity/impulsivity in previous studies of ADHD developmental trajectories [e.g.
]. Previous studies have, however, only included symptoms as reported by a single informant in their models and could, therefore, not identify distinctions between individuals with different patterns of expression across school and home contexts. The classes identified in the current study differed in overall levels of inattention symptoms as well as in developmental and informant pattern. Four of the classes evidenced informant discrepancies in levels and/or changes in inattention symptoms over time and two of these evidenced a ‘crossing-over’ effect whereby the informant who initially reported high levels reported lower levels by the end of the studied period, and vice versa.
The four classes characterised by elevated levels of inattention at any time across the elementary school years by either informant were compared to the class characterised by consistently low levels as reported by both informants. The classes were compared on ‘home’, ‘school’ and cross-situational inattention predictors. These predictors were reported by the child in order to avoid common rater bias. When considering potential predictors of class membership individually, issues with peers, teachers, parenting and low self-control all predicted membership in the classes characterised by consistently high or increasing levels relative to the ‘consistently low’ class. However, when considering the incremental contribution of the predictors (i.e., after controlling for gender and all other predictors), only two predictors were significant and both had modest effect sizes after adjustment for other predictors. Specifically, low self-control predicted membership in the class characterised by high stable levels (OR = 1.09) while teacher problems predicted membership in the class characterised by increasing levels of symptoms (OR = 0.80).
These results support the idea that consistently high levels of symptoms may be predicted by individual ‘trait-like’ features of the child such as low self-control, whereas changing levels may derive from the onset of time-varying influences such as poor relationships with teachers. In addition, the fact that teacher but not parent factors uniquely predicted symptoms that were particularly high at school supports the criterion validity of the school-specific inattention subtype.
An analogous set of analyses were conducted for hyperactivity/impulsivity symptoms. Here, five classes were also judged optimal, four of which evidenced a discrepancy between informants. When comparing these five classes on their levels of home, school and cross-situational correlates, the class with the lowest overall levels of hyperactivity/impulsivity symptoms (in which parents reported higher levels than teachers) served as the reference category. Bivariate analyses suggested that other than the ‘moderate stable’ class, classes could generally be differentiated from the reference class on the basis of peer, parent, and teacher problems and on low self-control. Examining the unique contributions of these predictors, however, there were only two significant results, both again with modest effect sizes after controlling for other predictors. First, parenting problems predicted membership in the ‘very high increasing/primarily school’ (OR = 1.04) class while low self-control predicted membership in the ‘high increasing/ primarily school’ (OR = 1.10) category. These two classes were the most ‘severe’ classes, i.e. they appeared to show the highest overall levels of symptoms and, therefore, it is not surprising that where ADHD risk factors uniquely significantly predicted class membership, it was in these two classes. This is all the more so given that both self-control and parenting were measured at age 11, the point on these two trajectories where hyperactivity/impulsivity symptoms were at their peak. It is less clear why low self-control predicted membership in the second most severe class but not the most severe class. Possibly those with the highest levels of ADHD symptoms are poorer at accurately recognising deficits in self-control, consistent with the positive illusory bias that has been observed in youth with ADHD [
]. However, these results don’t support the hypothesised mappings of home, school and cross-context risk factors with ADHD symptom presentations in the corresponding contexts. Rather, they suggest that for hyperactivity/impulsivity, risk factors in the home are related to symptoms at school over and child low self-control and relationships with teachers.
Taken together and as anticipated, our results suggested different patterns for inattention and hyperactivity/impulsivity. Although both could be summarised in terms of five trajectory classes, the trajectories represented in the classes differed. For example, while hyperactivity/impulsivity tended to be stable across development in the home context, evidencing variability mainly at school, inattention showed changes over development in both contexts. In addition, while there were two inattention classes in which the context of greatest severity swapped over time, the context of greatest severity remained constant across development for all of the hyperactivity/impulsivity classes. Further, while the inattention classes could to some degree be mapped to context-specific and context-general predictors, the same could not be said for hyperactivity/impulsivity. Arguably this suggests that utilising specifiers to indicate the context of greatest symptom expression could be more informative for inattention than for hyperactivity/impulsivity.
While it would be premature to derive any clinical implications from the current study, our results are indicative of the potential utility of further exploring the introduction of ADHD presentations based on the context(s) in which symptoms are expressed. This would represent a more nuanced approach than the current situation in which an individual must show significant symptoms across multiple domains to receive a diagnosis. This means that individuals with severe symptoms could be missing out on support and interventions from which they could benefit if clinically significant symptoms cannot be evidenced across multiple contexts. An alternative proposal would be to utilise a single cut-off for severity but to use a specifier to identify the primary contexts in which symptoms are present. Several steps will be required to evaluate the potential clinical utility of this approach. First, the mixture analyses of the current study should be replicated in other datasets to establish which context-based presentation categories are replicable. Second, a broader range of risk factors should be analysed to assess whether individuals with presentations in different contexts (and multiple versus single contexts) appear to differ in etiology. Third, it should be evaluated whether these presentations are associated with different or more severe patterns of psychosocial impairment. For example, whether those with symptoms across multiple contexts are more prone to common ADHD comorbidities such as anxiety and depression, oppositional defiant disorder, and conduct disorder. Where possible it should be evaluated whether presentations in different contexts are related to treatment responses. For example, teacher- and parent-administered interventions are recommended as psychosocial treatments for ADHD [
]; however, the extent to which an individual benefits from one or the other could depend on whether their symptoms are more severe in the context of school or home. Finally, though our focus was on ADHD symptoms, similar differences in symptoms of related disorders may also be expected to show meaningful subtypes according to contexts or situations [
]. Thus, it would be of interest to replicate the current study with oppositional and conduct problems.
It is important to note the limitations of the current study. Though our results suggested only limited evidence for the criterion validity of informant-based developmental subtypes, it included only a handful of home- school- and cross-situational predictors and it would be beneficial to explore associations with a broader range of established context-specific and context-general ADHD predictors in future research. In addition, two of our criterion measures (teacher relationships, peer relationships), though evidencing good reliability in the current study, have undergone limited prior psychometric evaluation. More broadly, concerns are sometimes raised about the validity of self-reports at younger ages. Our parenting measure showed a slightly low reliability, only just exceeding conventionally accepted levels (Omega = .71), suggesting that its associations with trajectory classes could have been under-estimated due to reliability attenuation. However, while there is evidence that measurement error is slightly greater in the age 11 reports, there is little evidence in z-proso that the age 11 self-reports are substantially less reliable or valid than self-reports taken at ages 13,15 or 17 in general [
]. Separately, however, it has been noted that individuals high in ADHD traits may have difficulty accurately reporting on their symptoms [
]. Logically this difficulty could extend to self-reports of other constructs and differentially affect criterion associations with membership in the various ADHD symptom presentation classes identified in the current study. Nevertheless, concerns about utilising self-reports for the criterion variables must be weighed against the fact that utilising parent- or teacher- reports for these could inflate associations due to common rater bias. This issue could be addressed in future studies using an additional independent informant and behavioural measures of self-control.
A second limitation of our study was the brevity of our ADHD measure; a function of being administered as part of a large cohort study. Replication with more comprehensive ADHD measures would be valuable. Second, as discussed in the introduction, differences in informant reports are not entirely due to differences in child behaviour. Measurement error and informant biases also play a role. Future studies that can control for informant characteristics such as stress or mental health problems, especially depression [
] can better isolate context-differences that are related to the child behaviour specifically.
Finally, for our analyses predicting category membership, FIML was not available and, therefore, listwise deletion was used. These analyses could, therefore, have been affected by non-random non-response. Given the patterns of non-random non-response identified by Eisner et al. [
], the most likely impact of this is an attenuation of the associations between predictors and ADHD class membership.
Our study found some support for subtyping ADHD symptoms on the basis of the informants who provide the information about symptoms. Growth mixture analyses in a normative sample of 1388 youth identified five categories that were that were distinguishable on the basis of informant reports of developmental trajectories. These categories included presentations in which symptoms were particularly elevated in school relative to home but none where they were particularly elevated at home compared to school. This suggests that were context-specific presentations occur, severity is more likely to be greater at school than at home. The categories identified showed only a weak tendency to map to context-specific and context-general predictors of ADHD. One exception was the finding that teacher-relationships uniquely and specifically predicted a rapid increase in inattention symptoms at school only. On balance, our results point to potential value in further exploring presentations that differ according to context. In particular, future studies could evaluate whether similar categories emerge in different samples and whether they can be mapped to etiological, functional, and outcome differences.