Introduction
In spite of decades of research, the prevalence of attention-deficit/hyperactivity disorder (ADHD) has been difficult to estimate and it is still a matter of controversy how frequent this phenotype is in a general population setting [
6]. Some of the discrepancies may be caused by cultural and social differences, acting on both the prevalence directly and on the reporting style. This may be the reason for the somewhat lower prevalence rates of ADHD found in the Scandinavian countries [
12]. There is also a wide variation as regards measures and sample characteristics [
6]. Optimally, the population prevalence should reflect the total population, but in practice it has been difficult to establish a level of study participation that makes the sample representative. Those who participate do not represent a random sample and this differentiated attrition biases the prevalence estimates of child psychiatric disorders such as ADHD. Parents of children rated as deviant by teachers have been found to be less likely to consent to research on child psychiatric disorders compared to parents of children rated within the normal range [
16]. In a previous publication from the Bergen Child Study (BCS), the impact of non-responder bias on the prevalence of several different child mental health problems was explored and an important finding was that teachers rated non-responders higher on all symptom scales, except tics, and as more impaired than responders [
18]. Teacher high scores (75, 90 and 95th percentiles) on inattention and/or hyperactivity had significantly increased relative risk for parental non-response. Yet we know little about the quantitative effect this would have on the estimation of ADHD prevalence. Another important issue of non-response is whether high scorers in the non-participating group might be qualitatively different from high scorers in the participating group with respect to symptom constellation and/or severity. Such bias could lead to important misinterpretation of results in the further stages of the study where clinical measures are applied and one seeks knowledge about clinical conditions in a representative sample from the general population. Few previous studies have had access to data for non-participants, and if such data have been available, it has included only demographics such as living area, ethnicity, age and gender.
Other important factors that influence the prevalence estimate in ADHD include the definition applied, symptom count, use of impairment, cross-situational criteria and choice of informant. As there is a wide variety of definitions, measures, informants and samples [
6], a better understanding of the factors that influence prevalence estimates is important when interpreting differences between studies.
The aims of the present study were (1) to estimate the prevalence of the ADHD phenotype in a general child population, based on parent and teacher reports, and (2) to analyze the effect of parental attrition, informant and gender on ADHD prevalence.
We report the prevalence of the ADHD phenotype based on reported symptoms from questionnaires and making no correction for level of impairment, while acknowledging that a clinical diagnosis cannot be based on questionnaire data only. For clinical purposes, the impairment of the symptoms is crucial, but for epidemiological purposes and comparison with other studies we rely on this readily reproducible method to measure the ADHD phenotype in the community.
Discussion
We found twice as many children with the ADHD phenotype among children in the Anonymous Data group versus the Full Data group, which demonstrated that attrition in studies with a typical attrition rate underestimated the ADHD phenotype prevalence. We estimated the prevalence of the ADHD phenotype to 5.2% (parent and teacher reports) among children whose parents consented to participate in the study, but 6.6% in the total population. Both parents and teachers reported more ADHD symptoms in boys than in girls, but the gender difference was greater according to teacher reports. The excess proportion of boys with hyperactivity/impulsivity and the combined symptom constellation high score was higher according to teacher reports than parent reports. Informant agreement was low to fair.
The estimated ADHD phenotype prevalence of 5.2% in our study was considerably higher than the DAWBA-based ADHD prevalence of 1.3% from a second study phase in the same population based on the Development and Well-Being Assessment (DAWBA) [
13]. This is not unexpected given that a DAWBA diagnosis requires the impairment criteria to be fulfilled and is therefore more comparable to a clinical diagnosis. Interestingly, our prevalence estimate for the ADHD phenotype was in the range of that reported from similar studies, while the DAWBA ADHD prevalence rate in the BCS was considerably lower than in a comparable British survey in a head-to-head comparison of the two samples with similar age groups and informants [
12].
Our prevalence estimate of 5.2% for the ADHD phenotype is comparable to that found in a recent German study [
5] reporting a prevalence of 6.4% in the same age group. Our prevalence estimate relied on two informants (which led to a decrease in prevalence), whereas the German study only included parent reports. On the other hand, the study included a 4-point response scale and the two most deviant responses were regarded as indicating the presence of “symptom”. This is probably a more conservative symptom definition than ours, given that we had only three response categories and defined the two most deviant as indicating symptom. Observing the behavior in different settings diminishes the likeliness of mixing it up with other behavioral disorders. The German study included no adjustment for non-responders, meaning that their prevalence rate was probably also an underestimate. Given these important methodological differences, it is somewhat surprising that the prevalence estimates are in the same range. This is not to be taken as support for a more solid evidence basis—that at the end of the day the reported prevalence rates were very similar. This may rather reflect that the choices made in a study may be influenced by previously reported results. This also demonstrates the liability of prevalence estimates to definition and the importance of thorough characterization of the methodology applied when referring to any reported prevalence of ADHD.
The access to anonymous teacher questionnaires for most of the non-participants was a special asset of our study. Comparing participants to non-participants, a much higher level of ADHD symptoms was found in the latter group and this finding is relevant for all population-based epidemiological studies independent of their definition of ADHD. Similar trends have been reported for autistic symptoms in the same cohort [
15]. Teacher reports showing a prevalence of 19.9% ADHD high scorers in the Anonymous Data group compared to 10.4% in the Full Data group (an OR of 2.1) clearly illustrate the very important effect of non-participation in population studies of ADHD symptomatology. The non-participant ADHD high scorers did not significantly differ from the participant ADHD high scorers in boy:girl ratio, age or on the impact measure. Thus, we did not get any support for the hypothesis that teacher-rated non-participant children with ADHD symptoms would be more impaired than participants. Teachers completed the questionnaires without any knowledge of who would later belong to the non-participant group. However, one could suspect that they might have had a pre-conceived idea of who was going to participate or not. Interestingly, the non-participant ADHD high scorers had higher inattention scores and less hyperactivity/impulsivity than the participant ADHD high scorers. The explanation for this finding is speculative as we lack comparable reports from other studies. The finding underscores the importance of trying to assess non-responder bias in epidemiology in general and in psychiatric research specifically. Though generally assumed that the non-participants are at higher risk for mental disorders and less privileged socially, few studies have explored the non-participation in sufficient detail to characterize the possible heterogeneity of non-participation. Investigating selective participation in the British Child and Adolescent Mental Health Surveys, Goodman and Gatward reported important heterogeneity in the effect of deprivation on parental non-participation [
9]. Thus, it is important to note that the process of non-participation is probably complicated with a heterogeneous set of reasons, which give rise to diverse effects on the non-participating group.
We reported an estimate of the influence of attrition on the ADHD phenotype prevalence estimate by assuming that the hypothetical parent reports of the children in the Anonymous group would have related to teacher reports at the same high scorer ratio as in the Full Data group. More sophisticated methods taking account of the differential parent–teacher agreement across number of symptoms for the high scorers or bootstrap methods might have been used to estimate the effect of attrition on the total population prevalence. However, as discussed above, there are several different uncertainties and limitations attached to the prevalence estimate (such as the differential use of impact, etc.) that in the end we opted for illustrating the non-response effect by this simple method as the interpretation of this estimate is straightforward. We underline the importance of evaluating each aspect of the various methodological influences rather than taking any one prevalence estimate as reflective of the “true” rate.
Our reported boy:girl ratios for ADHD high scorers on DSM-IV symptoms are in the range of earlier studies in community samples [
4,
8,
10,
11,
14,
20,
21]. Parents identified more girls than teachers, a finding that has been reported for the hyperactive/impulsive and for the combined subtypes in a previous study [
8], but it is not clear whether the higher number of girls identified by parents represent an underidentification by teachers or an overidentification by parents. Boys with ADHD are reported to engage in more rule breaking and externalizing behavior than girls with ADHD [
2], and this has been found to affect teacher ratings of ADHD [
1]. Some authors have found support for the hypothesis that the difference in symptom ratings across informants could be due to real situational differences [
7]. Although the cause of the difference in parent and teacher reports on ADHD symptoms in girls remains unresolved, it is important to bear this in mind and to explore the issue further in future studies.
The BCS is unique in that teacher questionnaires cover 97% of the total population. The current study focused on symptoms of ADHD as reported on questionnaires. The validity of such reports may be questioned, since informants may misunderstand items, and may also have reasons for over- and underreporting problems in the child. Also, the DSM-IV diagnostic criteria require an early onset of the disorder (before age 7) and pervasive impairment from the symptoms. Thus, the phenotype and subtypes referred to here only indicate symptom constellations as specified in the diagnostic criteria and are not comparable to a clinical diagnosis. However, the symptom count approach may be more readily reproducible than clinical diagnoses in epidemiological research.
The use of only three response categories represented a problem in the current study. It is not clear whether the middle category should be regarded as having the symptom or not. Many DSM-IV ADHD rating scales have used 4-point scales, where the two highest scores have been interpreted as indicating a symptom [
4,
8,
21]. However, our prevalence of the ADHD phenotype according to teacher reports is comparable to figures reported in previous studies of teacher-reported DSM-IV ADHD [
4,
8,
15,
21]. Similarly, the frequency of parent-reported ADHD symptom subtypes was comparable to that found by other studies using parent information [
10]. Our use of strict cross informant criteria compensated for a somewhat less restrictive individual symptom definition in estimating the prevalence based on both informants. A more conservative definition of symptom presence would have been to count only “Certainly true” answers as symptom present. We considered that the somewhat more inclusive symptom criteria were suitable for the epidemiological considerations in this general child population study.
Acknowledgments
We are grateful for the participation of children, parents and teachers in the study. We thank Stein Atle Lie and Tore Wentzel-Larsen for statistical advice and to Jim Stevenson and Astri J. Lundervold for important comments on the manuscript. This study was supported by the Norwegian Research Council, the Norwegian Directorate of Health, the Western Regional Health Authority and the Centre for Child and Adolescent Mental Health, Uni Health, Bergen. Christopher Gillberg was funded by the Swedish Medical Research Council. The work of Carsten Obel was funded by the Nordic Council of Ministers research program ‘Longitudinal Epidemiology’ (020056).