Epidemiology of psychiatric disorder in childhood and adolescence
E. Jane Costello
Adrian Angold
Epidemiology is the study of patterns of disease in human populations.(1) Patterns are non-random distributions, and patterns of disease distribution occur in both time and space. Whenever we observe a non-random distribution, we have the opportunity to identify causal factors that influence who gets a disease and who does not. For example, we observe that depression rises rapidly after puberty in girls, but not to the same extent in boys.(2) This non-random distribution in time suggests that there may be something about puberty in girls that is causally related to depression.(3) An example of disease distribution in space can be seen in the Methods for the Epidemiology of Child and Adolescent Mental Disorders (MECA) study of five sites in the United States and Puerto Rico.(4) Although the prevalence of psychiatric disorders was fairly similar across sites, the likelihood that a psychiatric diagnosis was accompanied by significant functional impairment was much higher in children at the mainland sites than in Puerto Rico. This offers the opportunity to study between-site differences that might result in differences in the level of impairment caused by psychiatric disorders. The task of epidemiology is to understand these observed patterns in time and space, and to use this understanding as a basis for the prevention and control of disease.
Epidemiological medicine has both similarities to and differences from clinical medicine. Like clinical medicine, epidemiology is an action-oriented discipline, whose goal is intervention to prevent and control disease. Scientific knowledge about the cause and course of disease is another common goal. Epidemiology also reflects clinical medicine in using two methods of attack on disease: tactical methods, concerned with the practical and administrative problems of disease control at the day-to-day level, and strategic methods, concerned with finding out what causes disease so that new weapons of prevention and control can be engineered.(5,6) Thus, for example, in their tactical or public health role epidemiologists can be found reporting on the prevalence of adolescent drug abuse, the social burden (including cost) that drug abuse creates, and the best ways to control its spread, while others working at the strategic level might be exploring the science underlying environmental constraints on gene expression.
Epidemiology diverges from clinical medicine to the extent that it concentrates on understanding and controlling disease processes in the context of the population at risk, whereas the primary focus of clinical medicine is the individual patient. This does not mean that epidemiology is not concerned with the individual; on the contrary, it is very much concerned with understanding the individual’s illness and the causes of that illness. The difference lies in the frame of reference. Put crudely, clinical medicine asks: ‘What is wrong with this person and how should I treat him or her?’ Epidemiology asks: ‘What is wrong with this person and what is it about him or her that has resulted in this illness?’ Why is this child depressed, but not her brother? If her mother is also depressed, is the child’s depression a cause, a consequence, or an unrelated, chance co-occurrence? Such questions immediately set the individual child within a frame of reference of other children, or other family members, or other people of the same sex or race or social class.
Sampling, or selecting the population within which to count cases, is of central importance in epidemiology. Counting cases is an important first step towards measuring the social burden caused by a disease, and the effectiveness of prevention. For most diseases, however, simply counting the number of individuals presenting for treatment will produce estimates that are seriously biased by referral practices, ability to pay, and other factors. This is a big problem in child psychiatry because parents, teachers, and pediatricians all serve as ‘gatekeepers’ to treatment.(7) Community-based data are needed to measure the extent of need, and the unmet need, for prevention or treatment. Methods for assessing psychiatric disorders in the general population are discussed in another chapter. However, it is worth noting that methods for assessing disorder, whether they take the form of interviews, questionnaires, or neuropsychiatric tests, can only be as good as the taxonomy they are designed to operationalize. Current instruments mainly use scoring algorithms that turn the responses into diagnoses based on the DSM-IV or ICD-10 taxonomies. If these taxonomies do not mirror the ‘reality’ of psychiatric disorder then the results of using interviews or questionnaires based on them will in turn be faulty.
Estimating the burden of child and adolescent psychiatric disorders
In a world of scarce health care resources, it is important to understand the size of the burden to the community caused by these disorders. Burden, in terms of numbers affected, impact on the individual, and cost to the community, is a crucial factor in the battle for resources for treatment and prevention.
Attempts to reduce the burden of mental illness must, of necessity, pay attention to the early years. It is becoming increasingly clear that most psychiatric disorders have their onset before adulthood, and that many should be regarded as chronic or relapsing disorders. For example, the National Comorbidity Survey Replication, a representative population sample of over 9000 adults aged 18 and over in the United States,(8) found that, of the 46.4 per cent of all participants reporting one or more psychiatric disorders during there lifetime, half reported onset by age 12, and three-quarters by age 24.(9) Since we can expect a lot of forgetting of early episodes by older participants,(10) it is likely that onset in childhood is even more common than this.
If the burden of mental illness begins to be felt in childhood, it is important to know the extent of the problem so that we can begin to plan for treatment and prevention. Unfortunately, the data on which to build such estimates are very sparse. We have to rely on a national prevalence study of psychiatric disorders in the United Kingdom, and another of a large area of Brazil, together with a few national or large community surveys using symptoms scales, and a handful of diagnosis-based studies in smaller community samples, some of them longitudinal. Questionnaire-based surveys are not very useful for measuring prevalence, because they tend to define ‘caseness’ in terms of a certain percentage of the sample with high scores; a method that predefines prevalence.
In the past decade the United Kingdom has carried out a national prevalence study,(11,12) conducted by the Office for National Statistics, with funding from the Department of Education and other agencies. The primary purpose was to produce prevalence estimates of conduct, emotional, and hyperkinetic disorders, as well as pervasive developmental disorder, eating disorders, and tic disorders, using both ICD-10 and DSM-IV criteria. The second aim was ‘to determine the impact or burden of children’s mental health. Impact covers the consequences for the child; burden reflects the consequences for others’.(13) (p. 185). Third, the study measured service use. A stratified random sampling plan for England, Scotland, and Wales produced a sample of 10438 children aged 5 to 15. Parent and child were interviewed using the Development and Well-Being Assessment (DAWBA),(14) a computer-assisted lay interview that uses a ‘best-estimate’ approach to diagnosis, in which responses recorded by lay interviewers are evaluated by clinicians. The first interview wave, conducted in 1999,(13) was followed by a questionnaire mailed 18 months later to all ‘cases’ with a diagnosis at Time 1, and a one-in-three random sample of non-cases. A second interview of all those completing questionnaires at Time 2, and all others who were cases at Time 1, was completed in 2002.(15) By weighting the responses to account for the various selection factors and for non-response, Meltzer and colleagues developed estimates of prevalence (i.e. the presence of a disorder at the Time 1 interview), of incidence (new cases between the two interviews), and of persistence.
The UK study found that almost one child in 10 (9.5 per cent) aged 5 to 15 had a psychiatric disorder based on the ICD-10 classification system. Prevalence was higher in adolescents (11.2 per cent at 11 to 15) than in children (8.2 per cent at 5 to 15), and in boys (11.4 per cent than girls 7.6 per cent). Conduct disorders were the most common (5.3 per cent), followed by anxiety disorders (3.8 per cent). Depression was rare in both sexes and all age groups
(0.9 per cent over all), as were hyperkinetic disorders (1.4 per cent). Seven per cent of previously unaffected children developed a psychiatric disorder in the 3 years between the interviews. Four per cent developed a new emotional disorder (anxiety and/or depression), and 5 per cent a behavioural and/or hyperkinetic disorder. More girls developed emotional disorders, and more boys developed behavioural disorders. Persistence, measured as the presence of the same diagnosis the years apart, was higher for behavioural disorders (43 per cent) than for emotional disorders (about one in four).
(0.9 per cent over all), as were hyperkinetic disorders (1.4 per cent). Seven per cent of previously unaffected children developed a psychiatric disorder in the 3 years between the interviews. Four per cent developed a new emotional disorder (anxiety and/or depression), and 5 per cent a behavioural and/or hyperkinetic disorder. More girls developed emotional disorders, and more boys developed behavioural disorders. Persistence, measured as the presence of the same diagnosis the years apart, was higher for behavioural disorders (43 per cent) than for emotional disorders (about one in four).
Factors affecting prevalence estimates
It is not a simple matter to compare the British prevalence rates with those from other countries, because there are few large studies, and the age ranges do not overlap. A study of youth age 7 to 14 in south-eastern Brazil, which used the same diagnostic interview but the DSM-IV taxonomy, found an overall prevalence of 12.7 per cent. Although prevalence estimates were slightly different from those reported by the UK study, the relative ordering was the same. Behavioural disorders were again the most common (7 per cent), followed by anxiety disorders (5.2 per cent) and ADHD (1.8 per cent). Once again, depression was rare (1.0 per cent). Other studies from around the world(16) usually generate prevalence rates of around 20 per cent. This puts the British and Brazilian studies at the low end of the range. However, there are many factors other than the ‘true’ rate of psychiatric disorder (if there is any such thing) that affect a published prevalence rate. The most important of these are:
1 The time frame of the diagnostic measure. Questions can be asked about symptoms occurring ‘now’, ‘in the past month’, ‘in the past 3, 6, or 12 months’, or ‘ever’. Clearly, if recall is accurate the latter questions will elicit more symptoms than the former. Unfortunately, recall is not always accurate. Prevalence rates are higher from interviews with longer time frames, but not as much higher as would be consistent with accurate recall. For example, The National Comorbidity Study Replication, based on a nationally representative sample of adults in the United States, found that the lifetime prevalence of any disorder was 46.4 per cent, while the 12-month prevalence was 26.2 per cent. This means that 26.2 per cent /46.4 per cent = 56.5 per cent of all cases across the lifespan were present in the past 12 months. This could be explained in several ways: (i) there was an epidemic of psychiatric disorders in the 12 months before the survey; (ii) over half of all psychiatric diseases are chronic; once they occur they remain active for the rest of life; (iii) many early episodes are forgotten, and people report the onset of the most recent episode as the first occurrence of the disorder. In the absence of any evidence for (i), some combination of (ii) and (iii) seems the most likely explanation. We have evidence that the reliability with which children and adults recall the first occurrence of a symptom falls dramatically after 3 months,(17) and recommend concentrating on symptoms occurring in the past 3 months if a fairly reliable estimate is sought.
In general, when comparing prevalence rates from different reports it is important to bear in mind the time frame. In a comparison of reported rates of child and adolescent depression published since the 1970s, we found that the time frame of the interview accounted for most of the variance, compared with taxonomy (DSM-III, DSM-IIIR, DSM-IV, ICD-9, ICD-10), diagnostic interview, or birth cohort.(2)
2 The number and nature of the informants. For several decades now clinicians and epidemiologists alike have recommended collecting information about a child from a range of informants: the parents, siblings, teachers, and peers, as well as the child. Most diagnostic instruments, whether questionnaires or interviews, exist in forms for diverse informants, with scoring algorithms that allow a diagnosis to be made on the basis of one informant or more. In the latter case, most follow the rule that clinicians generally observe, of counting a symptom as present if reported by any informant, rather than expecting agreement among informants, which rarely occurs.(18) Rates of psychiatric disorder will vary with the number of informants, and also depending on which informants report on which diagnoses. For example, across repeated assessments of 1420 children and adolescents, only 26 per cent of those with a diagnosis from the child interview also had one from the parent interview, and only 22 per cent of those with a diagnosis based on the parent interview had one from the child interview. This was statistically a highly significant level of agreement (OR 5.4, 95 per cent; CI 3.6, 7.9; p <.0001), but nevertheless only 13.5 per cent of cases were reported by both informants. Readers of epidemiological studies need to decide for themselves how much the number and type of informant matters in judging the accuracy of a prevalence estimate of a specific disorder. For example, parents often do not know much about their children’s drug use, while young children themselves generally have little insight into their own hyperactivity, and teachers seldom notice children’s depression. Prevalence rates based solely on these informants would be likely to be quite low.Related posts:
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
