Predictive validity of attentional functions in differentiating children with and without ADHD: a componential analysis

Authors

  • LIANE KAUFMANN,

    1.  Division of Neuropediatrics, Department of Pediatrics IV, Innsbruck Medical University, Innsbruck, Austria
    2.  Department of Psychology, University of Salzburg, Salzburg, Austria
    3.  Department of Psychology, University for Health Sciences, Medical Informatics and Technology, Hall in Tirol, Austria
    Search for more papers by this author
  • NIKOLA ZIEREN,

    1.  Section of Neuropsychology, Department of Neurology, University Hospital of the RWTH Aachen, Aachen, Germany.
    Search for more papers by this author
  • SIBYLLE ZOTTER,

    1.  Division of Neuropediatrics, Department of Pediatrics IV, Innsbruck Medical University, Innsbruck, Austria
    Search for more papers by this author
  • DANIELA KARALL,

    1.  Division of Neuropediatrics, Department of Pediatrics IV, Innsbruck Medical University, Innsbruck, Austria
    Search for more papers by this author
  • SABINE SCHOLL-BÜRGI,

    1.  Division of Neuropediatrics, Department of Pediatrics IV, Innsbruck Medical University, Innsbruck, Austria
    Search for more papers by this author
  • EDDA HABERLANDT,

    1.  Division of Neuropediatrics, Department of Pediatrics IV, Innsbruck Medical University, Innsbruck, Austria
    Search for more papers by this author
  • BRUNO FIMM

    1.  Section of Neuropsychology, Department of Neurology, University Hospital of the RWTH Aachen, Aachen, Germany.
    Search for more papers by this author

Dr Liane Kaufmann at Department of Psychology, University for Health Sciences, Medical Informatics and Technology, A-6060 Hall in Tirol, Austria. E-mail: liane.kaufmann@umit.at

Abstract

Aim  The objective of this study was to investigate which attentional components are of predictive utility in differentiating children with attention-deficit–hyperactivity disorder, combined type (ADHD-C) from their peers without ADHD.

Methods  Thirty-four children participated in the study: 17 males with ADHD-C (mean age 10y 4mo, SD 1y 9mo) and 17 comparison children (12 males, 5 females; mean age 10y 8mo, SD 1.7y). Attentional functions were assessed using a computer-administered, child-friendly test series in German (i.e. Testbatterie zur Aufmerksamkeitsprüfung für Kinder; KITAP). The KITAP measures several attentional components, including alertness and executive attention (inhibition, divided attention, flexibility).

Results  The variable best able to discriminate between children with and without ADHD-C was found to be response time variability in a go/no go task, followed by, in order, number of errors in a divided attention task and response time variability in an alertness task. However, group discrimination was not facilitated by differences in either response latency or accuracy of response in visuospatial attention and attentional flexibility tasks.

Interpretation  The assessment of attentional functions proved to be a powerful instrument for discriminating between children with and without ADHD-C. Notably, the discriminative power of executive attention was found to be task dependent and dependent on processing demands.

Attention is one of the most fundamental cognitive processes. Over the years, different conceptualizations of attention have emerged; some of them are rather general (attention as a pool of resources),1 whereas others are more specific (attention as enhanced processing of relevant stimuli).2 The hypothesized underlying neural mechanism is the top-down guidance of attention, which has been studied extensively by means of electrophysiological3 and functional brain imaging studies.4,5 The latter reveal that neural activity in task-relevant regions is higher in response to attended than to unattended stimuli.

Attention is not a unitary construct but, rather, comprises a multicomponential set of neurocognitive functions and processes.6 According to Posner and Peterson2 and Rothbart and Posner,7 attention is best conceptualized by three interrelated neurofunctional networks, namely orienting, alerting, and executive control. The orienting network mediates visuospatial attention and is responsible for detecting sensory input. It is thought to be modulated by acetylcholine and is supported by posterior attentional networks (including the superior parietal lobe and the temporoparietal junction).8 The alerting network mediates the employment and maintenance of the alert state and is supported by frontoparietal regions of the right hemisphere, possibly modulated by noradrenergic neurotransmitters.8 The executive network is enrolled in attention shifting, inhibitory control, and conflict monitoring. It is thought to be modulated by dopamine and is subserved by prefrontal networks including the anterior cingulate cortex and the dorsolateral prefrontal cortex.6,9–11 Dysfunction or disruption of one or more of these neural networks (either directly by structural damage or indirectly through neurochemical or neurogenetic malfunctions) leads to rather circumscribed attentional disabilities.

The majority of systematic investigations of attentional functions have been conducted in adults. However, recently, Rueda et al.12 assessed developmental trajectories of attentional networks in healthy children using a version of the attention network task (ANT) adapted for use in children. Their behavioural findings suggested that the three attentional networks2 that have been repeatedly validated in adult studies7 seem to be dissociable in children as well.

A somewhat different conceptualization of attention has been proposed by van Zomeren and Brouwer,13 who draw the distinction between intensity and selectivity aspects of attention. Within this framework, intensity aspects of attention are best conceptualized by alertness and sustained attention, whereas selectivity aspects of attention comprise visuospatial, focused, and divided attention, the last of which is generally attributed to executive attention.

Attentional processing deficiencies are a defining symptom of attention-deficit–hyperactivity disorder (ADHD).14 Although inattention is a key feature of the behavioural profile of ADHD, inattention is neither formally defined nor further distinguished in cognitive terms in the diagnostic criteria of the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV).15

Although scientific and public interest in ADHD has increased considerably during the past two decades, the nature of the attentional deficiencies seen in ADHD is not well understood. The findings of recent behavioural and neuroimaging studies have been inconsistent, with some being suggestive of impaired orienting attention systems (reflected in difficulties to disengage attention),16 but others indicating that the most pronounced deficiencies are found in executive attention systems (as evidenced by impaired interference control).14,17 Alternatively, the findings of Konrad et al.18 reveal that the distinguishing feature between children with ADHD and without ADHD is not a complete failure to recruit attention networks, but rather the presence of quite differential activation patterns in response to attentional demands. In a similar vein, recent findings suggest that response latencies in children with ADHD may be more variable than in comparison children without ADHD.19,20

Therefore, our current understanding of the neurocognitive underpinnings of ADHD is incomplete. A better characterization of attentional abilities in children diagnosed with ADHD may advance our understanding of the cognitive foundations of ADHD and, moreover, could be beneficial for both differential diagnosis and intervention planning.

The main aims of the present study were therefore to (1) compare attentional profiles in children with and without ADHD, combined type (ADHD-C), using a commercially available test series21 and (2) assess which attentional components are most descriptive for children with ADHD-C. Most importantly, the study aimed to identify the tasks that are best able to discriminate between children with and without ADHD-C.

Our assumption was as follows. Compared with children without ADHD, children with ADHD-C display the most pronounced difficulties in tasks involving orienting attention16 and executive attention networks.14,17 Moreover, we hypothesize that potential performance deficits in children with ADHD-C are not restricted to performance accuracy, but are also reflected in response latencies.19,20

Method

Participants

Out of 22 participating children with ADHD-C, two (aged 11y and 7y) were excluded because in all subtests their performance was more than 2.5SD higher than the group mean (response times or errors/omissions). A further three children with ADHD-C had to be excluded because their data contained too many missing values as a result of technical problems during test administration. Thus, 17 children diagnosed with ADHD-C were included in the final analyses. The comparison group consisted of 17 children and did not differ significantly from children with ADHD-C as regards mean age (ADHD-C group: 10y 4mo, SD 1y 9mo; comparison group: 10y 9mo, SD 1y 8mo; t32=−0.782, not significant [NS]) and overall intellectual ability (mean IQ), as measured by the short form of the Weschler Intelligence Scale for Children 3rd edition (WISC-III)22 (ADHD-C group: 109.0; SD 15.5; comparison group: 106.7; SD=12.2; t32=0.495, NS).

In both groups all but one child were right-handed. Five children in the comparison group were females, whereas the ADHD group consisted only of males (which is in accordance with prevalence rates15). As males and females are known to perform equally well on attentional tasks,23 the inclusion of females in the comparison group was not expected to contribute to eventual group differences. Exclusion criteria were concomitant neuropsychiatric and neuropaediatric diseases, structural brain abnormalities, severe sensory/motor impairments, and an intellectual level below 85 (prorated IQ derived from four subtests of the German version of the WISC-III22). Children with severe behavioural problems (as indicated by parent’s or teacher’s rating) or commonly associated comorbid conditions, such as oppositional defiant disorder or anxiety disorders, were also excluded from the study.

Clinical diagnoses were based on DSM-IV criteria15 and verified by semistructured psychiatric interview and behaviour rating scales (German translation of the Conners24 Parent and Teacher Rating Scales; revised and short form [parent rating ADHD index: mean 69.12, SD 5.66, range 60–85; teacher rating ADHD index: mean 75.29, SD 5.44; range 65–90; mean ADHD index over both ratings for all children with ADHD-C >70]).The presence of attentional problems was further verified using the Child Behavior Checklist (CBCL);25 every child in the ADHD-C group achieved a T score greater than 60 on the attention subscale of the CBCL. All children in the ADHD group were classified as having ADHD-C (concomitant presence of inattention and impulsivity/hyperactivity). At the time of testing, children with ADHD-C were off medication (median time off medication 12.5h). All children with ADHD were treated with stimulants (i.e. methylphenidate).

Children were recruited from the Neuropediatric Department of Innsbruck Medical University, Austria. All children with ADHD-C were clinic-referred and tested as outpatients. In contrast, comparison children were selectively recruited at local private schools. The study was approved by the local ethics committee. All children participated after having been informed of the scope of the study. Informed consent was obtained from their parents or caregivers.

Tasks

Attentional functions were assessed using the KITAP (Testbatterie zur Aufmerksamkeitsprüfung für Kinder).21 The KITAP is a German language child-friendly, commercially available, computer-assisted test series tapping various non-verbal attentional functions. The reliability of the KITAP is satisfactory. For 8- to 10-year-olds (which is the age range most closely matching the age of our study participants), the split-half reliability of KITAP test variables examined in the present study is reported to vary between 0.77 and 0.96 (sample sizes ranging from n=212–337). Moreover, according to the test manual, the split-half reliability for two test variables (i.e. flexibility errors and go/no go errors), because of the very low error frequencies, is greatly underestimated (i.e. 0.55 and 0.66 respectively). The internal validity of the KITAP has been confirmed by conducting factorial analyses. The narrative of the KITAP is an enchanted castle, requiring the child to respond manually to different task demands (by using an external response device). The instructions emphasize the importance of both response accuracy and speed. Except for two subtests (i.e. go/no go and divided attention), stimulus durations are response bound (i.e. without response deadlines). The subtests go/no go and divided attention have fixed response times.

Each of the KITAP subtests described below can be assigned to specific attentional processes, as follows: alertness– intensity aspect of attention; visual scanning– visuospatial attention; divided attention– parallel processing of visual and acoustic stimuli; go/no go– selection and inhibition of attention; and flexibility– shifting of attentional control. Tasks tapping divided attention, inhibition, and flexibility were included with the aim of identifying the most relevant cognitive abilities subsumed under the umbrella term ‘executive control’ (which, by definition, is a key deficit associated with ADHD14,15). In addition to measuring response accuracy, median response times and the SD of response latencies were computed for all subtests. Brief descriptions of KITAP subtests and corresponding test parameters included in this study are provided below (see Table I for an overview of test parameters and associated attentional functions).

Table I.   Parameters included in the discriminant analysis
KITAP subtestKITAP variables included in further analysesPostulated attentional function
  1. Terms given in square brackets correspond to the conceptualization of attention proposed by van Zomeren and Brouwer.13 KITAP, Testbatterie zur Aufmerksamkeitsprüfung für Kinder.

AlertnessMedian response timeIntrinsic alertness [intensity aspect]
SD
FlexibilityMedian response timeExecutive attention: flexible control of attentional focus (‘set shifting’) [selectivity aspect]
Errors
Divided attentionOmissionsExecutive attention: parallel processing of visual and acoustic stimuli [selectivity aspect]
Errors
Go/no goMedian response timeExecutive attention: selection and inhibition [selectivity aspect]
Errors
Go/no goSD[intensity aspect]
Visual scanningMedian response timeVisuospatial (orienting) attention/visual search [selectivity aspect]
Omissions
Errors

Alertness

The alertness subtest required children to respond as quickly as possible by pressing a response key whenever a witch appeared in the middle of the computer screen (variable interstimulus intervals). The overall duration of the alertness subtest was response bound, the mean duration being about 1.5 minutes. Median response times and SD were recorded.

Visual scanning

Children were presented with a regular visual array (five rows of five witches riding on broomsticks) and were asked to indicate as quickly as possible the potential presence of a witch flying in the ‘wrong’ direction as indicated by their broomstick. One hundred visual arrays with a total of 50 targets were presented. The overall duration of the visual scanning subtest is response bound and age dependent. Median response times for trials without targets were recorded, as well as errors (responses to non-targets) and omissions.

Divided attention

Children were requested to simultaneously attend to visual and auditory stimuli and to indicate a longer than usual pause in the hooting of owls and/or the closing of the guardian owls’ eyes (indicating falling asleep). A total of 296 visual and 297 auditory stimuli were presented, 20 targets per modality. The overall duration of the divided attention subtest was fixed at 4.5 minutes. Errors and omissions for visual and auditory trials were recorded.

Go/no go

Children were asked to press a button whenever a bat appeared on the screen and to inhibit a response upon the appearance of a cat. Forty stimuli (20 targets and 20 non-targets) were presented. The overall duration was fixed at 3 minutes. Median response times and SD as well as errors were recorded.

Flexibility

Here, children were required to press a button on the side at which a green or a blue dragon appeared. Fifty stimuli were presented. The overall duration of the flexibility subtest was response bound, lasting approximately 2 minutes. Median response times as well as errors were recorded.

Statistical analysis

To ensure that discrimination between the two groups was not caused by single participants with extreme test results, an explorative data analysis was conducted to eliminate such outliers (see first paragraph of Method section). After this trimming procedure, relevant test parameters (according to the test authors; see Table I) were selected as predictors for a subsequent discriminant analysis classifying whether study participants belonged to the patient (ADHD-C) or comparison (non-ADHD) group. Whenever error data were analysed, the arcsine transformation was used to approximate normal distributions, whereas for response latencies a common logarithmic transformation was used.26 The validity of the resulting discriminant function was subsequently tested by means of 1000 bootstrap samples (random samples of size 34 from the original participant sample with replacement).

Following a two-step-strategy we first computed a stepwise discriminant analysis with each bootstrap sample and with the same predictors as in the original analysis to check which variables were selected as significant predictors.27 In a second step, we used another 1000 bootstrap samples to compute discriminant analyses, with the whole set of variables being entered in the equation and without any stepwise inclusion/exclusion procedure. The aim was to evaluate the relative importance of each variable in the context of the other variables. Bootstrapping led to (1) an evaluation of the predictor set being entered in the discriminant function and (2) an estimation of the distribution of relevant parameters (correlation of each variable with the discriminant function, classification rates).

Results

Discriminant analysis

A stepwise discriminant analysis was conducted with the variables listed in Table I. The criterion for inclusion of a variable into the equation was a p value of 0.05 or lower; the criterion of exclusion was a p value of the corresponding F score of 0.10 (Wilks lambda criterion). Only two variables, the SD of response latencies in the go/no go test (go/no go SD) and the errors in the divided attention test, contributed significantly to the discrimination of children with and without ADHD-C. Table II presents the standardized canonical discriminant function coefficients and the correlation coefficients between both significant KITAP variables and its discriminant function (structure matrix). As can be seen from Table II, the SD of go/no go contributed most to the discriminant factor. Table III provides an overview of descriptive results on all variables entered in the discriminant analysis.

Table II.   Standardized discrimination coefficients and pooled within-group correlations between discriminating variables and standardized canonical discriminant functions
KITAP variableDiscriminant analysis, original sampleBootstrap discriminant analysis
Standardized discriminant coefficientsDiscriminant functionaDiscriminant functionb68% confidence intervalc
  1. The table displays parameters for the stepwise discriminant analysis of the original sample and the median correlations as well as the corresponding 68% confidence intervals of 1000 bootstrap discriminant analyses with all predictors entered into the equation. Note that correlations with variables of interest (i.e. attention component) may range from +1 to −1, higher values indicating stronger correlations. Variables in bold type contribute significantly to group discrimination in the original sample. aCorrelation based on the original sample and the stepwise discriminant analysis with two significant predictors. bMedian correlation based on the correlations of 1000 bootstrap discriminant analyses with all variables in the equation. cConfidence interval representing percentiles 16 and 84 of the distribution of correlations based on the 1000 bootstrap discriminant analyses. KITAP, Testbatterie zur Aufmerksamkeitsprüfung für Kinder.

Alertness
 Median response time  0.115−0.011 to 0.229
 SD  0.2830.125 to 0.411
Divided attention
 Errors0.6400.4880.1910.056 to 0.322
 Omissions  0.1530.032 to 0.280
Go/no go
 Median response time  0.1670.038 to 0.281
 SD0.8860.7760.3200.155 to 0.451
 Errors  −0.027−0.134 to 0.076
Flexibility
 Median response time  0.1370.013 to 0.251
 Errors  0.1340.014 to 0.260
Visual scanning
 Median response time  −0.035−0.155 to 0.073
 Errors  0.1300.023 to 0.228
 Omissions  0.067−0.027 to 0.167
Table III.   Overview of descriptive results, median (SD)
KITAP variableChildren with ADHD-C (n=17)Children without ADHD-C (n=17)
  1. KITAP, Testbatterie zur Aufmerksamkeitsprüfung für Kinder; ADHD-C, attention-deficit–hyperactivity disorder, combined type.

Alertness
 Median response time, ms345.5 (89.2)309.5 (61.1)
 SD, ms112.2 (78.9)60.1 (36.3)
Flexibility
 Errors4.1 (3.7)2.2 (1.6)
 Median response time, ms1118.1 (318.1)972.8 (334.8)
Divided attention
 Errors28.2 (20.9)16.2 (17.7)
 Omissions13.7 (5.8)10.2 (9.0)
Go/no go
 Errors2.1 (2.4)2.1 (1.7)
 Median response time, ms546.2 (117.7)475.5 (103.0)
 SD, ms146.4 (47.2)100.0 (30.4)
Visual scanning
 Omissions12.5 (10.2)9.7 (7.2)
 Errors5.4 (10.5)1.8 (2.3)
 Median response time, ms5725.7 (2438.7)6206.1 (2362.3)

Bootstrap discriminant analysis

First, based on 1000 bootstrap samples from the original participant sample (n=34; with replacement), stepwise discriminant analyses with the predictor set from Table I were computed. The size of the final predictor set entered into the discriminant function varied between one and nine, with two- and three-predictor solutions predominating (one predictor, 113 analyses; two predictors, 305 analyses; three predictors, 307 analyses; four predictors, 162 analyses; five predictors, 61 analyses; six predictors, 31 analyses; seven predictors, 13 analyses; eight predictors, six analyses; nine predictors, two analyses). In the case of one selected predictor, the go/no go SD was chosen in 72 analyses. Two-predictor solutions mainly consisted of the alertness SD (105), divided attention errors (181), or the go/no go SD (180), whereas three-predictor solutions again comprised the alertness SD (181), divided attention errors (220), and the go/no go SD (142) as significant predictors. Overall, divided attention errors were selected in 627 cases, followed by the go/no go SD (535) and alertness SD (528). The remaining predictors from Table I seem to be less important (alertness median response time, 263; divided attention omissions, 246; flexibility errors, 152 etc.). Thus, the bootstrap discriminant analyses confirm that the go/no go SD and divided attention errors are the variables with the highest discriminative potential. In addition, the alertness SD is considered to add incremental validity in discriminating between groups.

Second, to obtain more information on the relative importance of each variable in the context of all variables, discriminant analyses including the whole set of 12 variables (see Table I) in the discriminant function and without any stepwise inclusion/exclusion of variables were computed for each of the 1000 bootstrap samples from the original sample. The median and 68% confidence interval of the correlation coefficients between each KITAP variable and its discriminant function (structure matrix) based on the 1000 analyses are shown in Table II. Again, the go/no go SD, followed by the alertness SD and divided attention errors, show the highest discriminative potential.

Classification analysis

A classification analysis (obeying Bayes’ rule) was conducted based on the results of the original discriminant analysis using the go/no go SD and the errors in divided attention as predictors. In the original count, 14 of the participants diagnosed with ADHD-C and 12 of the comparison group could be successfully classified based on their results in both KITAP subtests. Based on the 1000 bootstrap discriminant analyses, the mean correct classification rate was 85.4% (SD 9.9%) for children with ADHD-C and 85.7% (SD 10.4%) for comparison children. However, these rates reflect the classification based on different sizes of predictor sets (1–9). Restricting predictors to only the go/no go SD and divided attention errors and subsequently computing 1000 bootstrap discriminant analyses leads to classification rates ranging from 44.1% to 100% (mean 78.3%, SD 8.0%). A restriction to the go/no go SD, divided attention errors, and alertness SD yields classification rates from 54.6 to 97.4% (mean 80.2%, SD 7.3%). Thus, the increase in the correct classification rate resulting from adding the alertness SD as a predictor seems to be negligible. This can be attributed to the rather high Pearson’s correlation of 0.559 (p<0.01; n=34) between the go/no go and alertness SDs, presumably representing a common attentional intensity aspect inherent in both variables.

The performance of correctly and incorrectly classified individuals in the two predictor variables (go/no go SD, divided attention errors) is depicted in Fig. 1. Children correctly classified as having ADHD-C show increased and children correctly classified as not having ADHD-C show decreased values in both parameters compared with the other group. The three ADHD-CT children misclassified as not having ADHD-C exhibited a small SD in go/no go and an average divided attention error rate, whereas in the case of the five control children misclassified as having ADHD-C, values of both parameters were around the mean.

Figure 1.

 Mean z-standardized predictor scores (go/no go SD and divided attention errors) of correctly and incorrectly classified participants. The error bars represent standard errors of the mean.

Discussion

The main objective of the present study was to assess which attentional components – as measured by a commercially available test series21– might be of diagnostic utility in discriminating children diagnosed with ADHD-C from comparison children without ADHD. The decision to utilize a commercially available test series was a pragmatic one and based on the observation that most similar studies have employed experimental tasks but that assessment tools designed for research purposes are generally not available to clinicians.

Taken together, our results are only partially consistent with our initial assumptions proposing that (1) tasks tapping orienting (i.e. visuospatial attention tasks) and executive attention are most likely to discriminate between children with and without ADHD-C; and (2) in addition to response accuracy, response latencies may also prove useful in identifying children with ADHD-C.

The results of a stepwise discriminant analysis revealed that only two variables out of 12 contributed significantly to group discrimination: response time variability in the go/no go task and errors committed in the divided attention task. At first glance, these results (i.e. high discriminatory power of tasks thought to tap executive attention such as go/no go and divided attention) seem to be compatible with a large body of literature suggesting that deficient executive attention is a defining feature of ADHD.14,17,28 However, on closer examination our findings are less stringent: if a generalized deficit in executive attention is present in ADHD, one would expect deficient performance in all tasks and variables thought to tap executive function. However, this was not the case here, as none of the variables pertaining to the flexibility task (which is also thought to measure executive attention) contributed significantly to group discrimination. Likewise, with the exception of the alertness task (i.e. response time variability), which was found to be of potential discriminatory utility, variables pertaining to non-executive attention aspects (i.e. visuospatial/orienting attention – visual scanning task) did not contribute to group discrimination. The finding that orienting attention tasks lack discriminatory power is not compatible with our assumption16 but is in line with the results of a recent meta-analysis reporting that there is no evidence for significant deficiencies in orienting attention in ADHD.29

A second important finding of our study is the discriminatory power of response time variability in go/no go and, to a lesser extent, alertness tasks (as shown by bootstrapping), which were also found to be highly intercorrelated. Thus, our results extend previous findings19,20 by showing that response time variability in itself might not be useful to distinguish between children with and without ADHD-C. Rather, our findings are suggestive of task-dependent effects of response variability and, in particular, provide evidence that children with ADHD-C tend to exhibit increased SDs in response latencies predominantly in processing tasks requiring inhibitory control. Interestingly, as revealed by bootstrap analysis, response time variability in processing tasks tapping general response readiness (i.e. alertness) may also be of potential discriminatory utility. Nonetheless, owing to the rather high correlation between response time variability in the alertness and go/no go tasks, the results of a subsequent stepwise discriminant analysis revealed that, in addition to the variable go/no go SD, the added value of the variable alertness SD was small and, thus, did not contribute significantly to discriminatory power. A likely explanation for the high correlation between the two variables (i.e. go/no go SD and alertness SD) is a common aspect of attentional intensity. Notably, response time variability reflects an individual’s ability (in case of low SD) or inability (in case of high SD) to maintain a stable response style and, thus, response time variability is considered to tap the intensity aspect of attention. In the case of unstable performance (i.e. high SD), an individual’s median response time may be rather misleading and, therefore, is considered to be less informative as regards attentional intensity.

Third, among children with ADHD-C, the significantly increased error rates in the divided attention task (which were accompanied by quicker response latencies, indicating the presence of a speed–accuracy trade-off; see Table III) mirror the high task complexity, which seriously hampered information processing in our group of children with ADHD. A plausible, but thus far speculative, assumption is that in the case of very complex tasks that are likely to increase error susceptibility, accuracy rates – tapping the selectivity aspect of attention – may be sufficient to determine group membership.

A novel finding is that the discriminatory power of executive attention seems to be task dependent and, moreover, dependent on processing demands. Upon attempting to delineate processing demands inherent in specific attention tasks, the distinction between intensity and selectivity aspects of attention may prove useful (see Table I).13 According to this terminology, the significantly increased response time variability in the go/no go task (and to a somewhat lesser extent in the alertness task) displayed by children with ADHD-C is most likely to reflect their difficulty in recruiting intensity aspects of attention, whereas the significantly higher error rates they displayed in the divided attention task are suggestive of deficient recruitment of attentional selectivity. Our findings of a combined deficit of intensity and selectivity aspects of attention (go/no go response time variability and divided attention errors, respectively) are compatible with the cognitive–energetic model of ADHD proposing a multilevel attentional deficiency17,30,31 and, furthermore, are in line with the notion that ADHD is characterized by a specific attentional regulation disorder.18,28

An alternative explanation for our finding that children diagnosed with ADHD-C perform poorly in some but not all executive function tasks speaks against a generalized executive attention deficit, thus rendering Posner’s attention model too simple to fully explain the present results. Rather, our results corroborate the existence of ‘executive fingerprints’32 in neurodevelopmental disorders such as ADHD-C, and suggest that the search for executive fingerprints in ADHD might be a useful diagnostic tool for clinical practice. Importantly, in addition to executive aspects of attention, specific aspects of non-executive attention also seem to be useful to discriminate children with and without ADHD-C. In particular, our findings highlight the diagnostic utility of attention tasks tapping inhibition, divided attention, and alertness (but less so flexibility and visual scanning).

These results adequately fit the observation that in everyday life the attentional control of children with ADHD varies considerably (depending on factors such as time of the day, level of fatigue, whether the task/activity at hand is extrinsically stimulating). Moreover, children with ADHD frequently fail to stay on track in situations involving dual-task processing that require them to divide their attentional resources. Using the KITAP, attentional subtests and variables that are expected to discriminate best between children with and without ADHD-C are the go/no go task (RT variability), the divided attention task (error rate), and, to a lesser extent, the alertness task (RT variability).

Limitations of the study

Potential limitations of the study are the rather small sample sizes (n=17 per group) and the unequal sex distribution in the two groups. As we are not aware of any study reporting differential effects of sex on attention processing, it is very unlikely that group differences reported here are attributable to sex effects. Another potential limitation of the study is that only children with a diagnosis of ADHD-C were included in the analyses. It is plausible that the different ADHD subtypes may exhibit differential attentional profiles. Thus, the variables found to be of comparatively high predictive validity for differentiating children with ADHD-C from their peers without ADHD may be of limited diagnostic utility for the predominantly inattentive or the predominantly hyperactive ADHD types.

Conclusion

Compared with children without ADHD, children with ADHD-C exhibited significantly larger response time variability in a go/no go task tapping executive attention and, moreover, committed significantly more errors in a divided attention task. Importantly, in 1000 bootstrapped discriminant analyses, the mean rate of correct classification of children with and without ADHD-C employing only these two tasks/variables was 78.3%. Adding alertness response time variability as a predictor (which showed a relatively high correlation with go/no go response variability) led to only a slight increase in correct classification. Of further interest is the small proportion of children with ADHD-C who were misclassified (n=3, corresponding to 16.6% of our sample). Although sample sizes are too small to derive any firm conclusions from this finding, our results suggest that cognitive subgroups of ADHD exist that warrant further investigation. Importantly, the discriminative power of executive attention seems to be task dependent and, moreover, seems to affect multiple processing demands (i.e. intensity and selectivity aspects of attention alike). Our results are new in so far as the assessment of attentional functions by a commercially available test series21 proved to be an economical and powerful instrument for differentiating children diagnosed with ADHD-C from their peers without ADHD.

Acknowledgements

We thank the anonymous reviewers and Klaus Willmes for valuable statistical advice. Moreover, we thank all participating children and their parents. L Kaufmann was supported by grants T286-B05 (Austrian Science Fund) and UNI-0404/523 (Tyrolean Science Fund).

Ancillary