Participants

The study sample was derived from a multicentre controlled trial of parenting groups for childhood antisocial behaviour (Reference Scott, Spender and DoolanScott et al, 2001b ). This study investigated the last 80 cases recruited in the original trial, which had a total sample of 141. The children were referred to four child and adolescent mental health services (CAMHS) in London (Brixton, Belgrave, Camberwell and St George's Hospital) or to the CAMHS in Chichester, in the south of England. Children eligible for inclusion in the study were aged 3–8 years, with severe antisocial behaviour as the presenting problem. On average, they were in the worst 1% of the population with the most disruptive behaviour, as measured by semi-structured interview. Children were excluded if there were clinically apparent developmental delays, hyperkinetic syndrome or any other condition requiring separate treatment.

Measures

Statistical methods

From the measures, a set of possible predictors was selected on the basis of previous research and discussions with the clinical expert on the team. We used multivariate statistical methods, taking the total cost of care as the dependent variable. The explanatory variables were age, gender of the child, gender of the main carer, ethnicity of the main carer, marital status of the main carer, education of the main carer, severity of symptoms in each of three domains (antisocial behaviour, hyperactivity and emotional problems), the subjective difficulty the main carer experienced with the child, and whether there was special educational needs provision made for the child.

Because cost data are typically skewed in their distribution, we planned to use both parametric and non-parametric methods of analysis. Initially, bivariate analysis was used to investigate associations between each of the child and family measures and the total costs of care, using simple linear regression. Associations between costs and continuous variables were conducted, but when we present the descriptive results below we separate two groups distinguished by each variable's median value. Both ordinary least squares and generalised linear modelling were then used to examine the associations between baseline characteristics, total costs and component costs such as NHS costs, education service costs, voluntary sector costs and indirect costs. Multiple regression using ordinary least squares can be misleading when the error term is non-normally distributed and has a non-constant variance. If the dependent variable is non-normally distributed, there is a good chance that the residuals will be too, which will invalidate tests of significance associated with fitting a model with ordinary least squares and so produce imprecise estimates of costs (Reference Dunn, Mirandola and AmaddeoDunn et al, 2003). An alternative is to use generalised linear modelling to address these concerns about the distribution of the data, as employed previously in the mental health field (Reference Knapp, Chisholm and LeeseKnapp et al, 2002a ). In this study, the Park test on the raw scale residuals was used to select one of the relevant family distributions (Gaussian, Poisson, gamma or inverse Gaussian) for the generalised linear model. If the values were consistently near 0, Gaussian was chosen; for values near 1, the Poisson was used; for values nearer to 2, gamma was used; for values nearer to 3, inverse Gaussian was used. For values far below 0 or much bigger than 3 (say 5 and above), it would be necessary to select another link function (Reference Manning and MullahyManning & Mullahy, 2001). Once the appropriate family distribution had been selected, generalised linear modelling was repeated with this choice of family using a robust option.

When multiple regression using standard ordinary least squares was conducted, all explanatory variables that had a bivariate association with costs were initially included in the model. Variables that did not have bivariate associations with costs were then included one at a time and were kept if they added significantly to the model. The decision to retain or discard a variable was based on significance at the 10% level. Analyses were conducted using the continuous variables.

Service use and impact on family

Service use patterns are shown in Table 3.

Table 3

Service use and costs and non-service impact of antisocial behaviour by children over the previous 12 months

Resource use	Participants using the resource n (%)	Use1 mean (s.d.)	Mean duration1 in minutes (s.d.)	Cost, £		Proportion of total cost %
				Unit cost2	Mean (s.d.)
Child
National Health Service
In-patient care (days)	32 (40)	8 (13.7)		300-916	287 (2000)	5
Accident and emergency (attendance)	20 (25)	2 (1.9)	180 (174)	52	29 (80)	0
Out-patient care (attendance)	19 (24)	3 (3.2)	60 (36)	151	77 (219)	1
Day hospital (attendance)	5 (6)	1 (0.9)	300 (252)	353	11 (45)	0
General practitioner (attendance)	57 (71)	5 (4)		19/h	13 (34)	0
Child guidance clinics (attendance)	17 (21)	4 (4)	60 (3)	55-87	16 (70)	0
Child development centre (attendance)	20 (25)	6 (10)	30 (60)	33-43	39 (245)	1
Health visitor (attendance)	28 (35)	4 (5)	36 (24)	21/h	19 (47)	0
Social care services
Social worker (attendance)	5 (6)	7 (11)	36 (24)	25-35/h	8 (61)	0
Voluntary services3
Voluntary services (attendance)	16 (20)	7 (19)	240 (60)	26-28	166 (1270)	3
Education services
Nursery school (attendance)4	52 (65)				410 (763)	7
Educational psychologist (visits)	23 (29)	2 (4)		32/h	12 (46)	0
Special needs support (months)	17 (21)	3 (1)	40 (18)	19/h	13 (85)	0
Remedial/classroom assistance (months)	19 (24)	5 (8)	120 (189)	19/h	103 (383)	2
Statement	2 (2)			2560 per child	74 (464)	1
Total service-based costs incurred by child					1277 (2309)
Family
National Health Service
Services5	36 (45)			19-114	45 (187)	1
Non-service costs incurred by family
Additional home repairs (number)	40 (50)	4 (6)		6.40/week	32 (73)	1
Extra time spent on household tasks6	76 (95)		462 (407)	5.57-17/h	4526 (4022)	76
Time carer spent off work (days)	10 (12)	16 (15)		4.17-11.61/h	79 (337)	1
Total non-service costs incurred by family					4637 (4432)
Total service and non-service costs, per child					5960 (13553)	100
State benefits paid to family					4307 (4189)

Table 4

Bivariate analysis of total costs for the child and family over 12 months

Predictors	n	Costs, mean (s.d.)	P
Child's age
< 5 years	35	5971 (5110)
			0.281
≥ 5 years	45	4810 (3705)
Child's gender
Male	60	6517 (5024)
			0.05
Female	20	4080 (3406)
Child's PACS score
Antisocial behaviour
Score ≤ 1.5	38	3711 (2554)
			0.031
Score > 1.5	42	5318 (4385)
Emotional symptoms
Score ≤ 0.8	48	5224 (4727)
			0.241
Score > 0.8	32	5447 (3941)
Hyperactivity
Score ≤ 1.1	42	4247 (3655)
			0.41
Score > 1.1	38	6752 (5624)
Special educational needs
Child has special education needs provision	19	5954 (3190)
			0.96
Child has no special education needs provision	61	5893 (5188)
Mother's ethnicity
White	56	5789 (4297)
			0.34
Other ethnic group	24	6186 (5824)
Mother's marital status
Single/lone parent	19	5946 (6064)
Cohabiting	49	5502 (3963)	0.96
Has visiting partner	12	7508 (5591)
Mother's education2
Left school at age 16 years	42	5125 (4086)
Left school aged 17-18 years	15	6032 (6156)	0.78
Left school after age 18 years	22	5360 (3667)
Mother's difficulty with child
Difficulty coping with child in first year	31	6136 (4784)
			0.79
No difficulty coping with child in first year	49	5764 (4807)

Cost of care

Annualised costs indicate the intensity of service use weighted by their unit costs. Table 3 provides a summary of the costs of services and other resource impacts over the previous 12 months. The results show that the mean cost of NHS, voluntary services and education services used directly by the child was £1277, which is 21% of the total costs (median £509, range 0–17 446). Additionally, service use by other family members because of the child's behaviour cost £45 – mainly owing to contact with primary care and community health services, such as seeing the general practitioner or a counsellor, or receiving therapy for anxiety and depression. However, the mean non-service costs of the child's behaviour to the family were high, at £4637 (median £3217; range 0–19 533), because of the burden of extra time spent on household tasks, the need for repairs, and time off work looking after the child.

The overall mean total cost per child was £5960 per annum (median £4957, range 48–19 940). Many children used few or no services; consequently many children had zero or very small costs. However, a smaller number of children made heavy use of relatively expensive services such as in-patient care. The family impact costs represent the highest element, contributing over three-quarters of the total. Indeed, for the majority of the families it was estimated that the time spent on additional household tasks had an opportunity cost of £4526 over the past year. This is equivalent to paying a local authority home care worker £87 per week to provide extra support around the home. The overall balance of costs is shown in Fig. 1.

Fig. 1

Service and non-service annual costs for child and family (local authority social services costs not represented as statistically 0%). NHS, National Health Service.

Cost associations

The only variables associated with total costs in the bivariate analyses were being male and increased severity of antisocial behaviour (Table 4).

The final ordinary least squares regression model is shown in Table 5. When all variables that had a significant bivariate relationship with costs were included, the only variable that was significantly associated with costs was increased severity of antisocial behaviour (P=0.05). Being male was of borderline significance (P=0.09). However, these variables did not explain much of the total cost variance (only 7%). Because of potential distributional problems with the ordinary least squares regression we re-estimated the equation by fitting the generalised linear model (Table 5). Only variables that were significant (P<0.1) in the bivariate testing were included. In the generalised linear model, both increased severity of antisocial behaviour score (P=0.01) and being male (P=0.02) were significant (Table 6). Thus, the relationships between costs and the predictors identified included in the final ordinary least squares and generalised linear models were broadly similar, although being male failed to reach significance using ordinary least squares. Further analyses were conducted with various cost components and demographic and clinical variables. Results using the generalised linear model are shown in Table 6. There were associations between non-service based costs and hyperactivity, antisocial behaviour score and being male.

Table 5

Ordinary least squares (OLS) and generalised linear modelling (GLM) regression analyses for total cost

Variables	OLS1			Log gamma GLM2
	β	(95% CI)	P	β	(95% CI)	P
Constant	660	(-2980 to 430 1)	0.72	7.4	(6.7 to 8.1)	<0.001
Gender: male (=1)	1880	(-317 to 4079)	0.09	0.4	(0.1 to 0.8)	0.02
Child PACS antisocial behaviour score	2192	(-25 to 4409)	0.05	0.5	(0.1 to 0.9)	0.01
Child PACS hyperactivity score	1533	(-812 to 3879)	0.20	0.1	(-0.8 to 0.3)	0.23

Table 6

Ordinary least squares (OLS) and generalised linear modelling (GLM) regression analyses for non-service costs incurred by families

Variables	OLS			Log gamma GLM
	β	(95% CI)	P	β	(95% CI)	P
Constant	2537	(830 to 4244)	0.04	7.1	(6.5 to 7.9)	<0.001
Gender: male (=1)	875	(-3568 to 2764)	0.35	0.4	(0.1 to 0.8)	0.02
Child PACS antisocial behaviour score	1694	(-165 to 3552)	0.07	0.5	(0.2 to 0.9)	0.01
Child PACS hyperactivity score	1404	(126 to 2672)	0.02	0.3	(0.0 to 0.5)	0.04

Implementation of effective interventions

Given the high burden and cost of antisocial behaviour in the short term, and the even higher costs in the long term, the case for using effective interventions seems overwhelming. There have been several hundred controlled trials proving the effectiveness of interventions (Reference Kazdin, Hill and MaughanKazdin, 2001), with parenting programmes having the greatest impact (Reference Scott, Rutter and TaylorScott, 2002); yet in England at present only a quarter of children with mental health problems receive specialist treatment (Reference Ford, Goodman and MeltzerFord et al, 2003). With the increasing pressure on child and adolescent mental health services to focus their limited resources on life-threatening disorders such as depression, psychosis, self-harm, anorexia and major drug misuse, some services are refusing to see cases of conduct disorder on the grounds that they are a ‘social problem’. This stance is understandable, given the very small amounts spent on child and adolescent mental health (about £10 per head per year; Audit Commission, 1999), but the thinking is fallacious. First, the problem is by no means caused only by faulty socialisation – indeed, recent studies show a high genetic contribution. Thus, for example, Scourfield et al (Reference Scourfield, Van den Bree and Neilson2004) found that the heritability of pervasive antisocial behaviour was 100%, with no environmental contribution at all. Second, other agencies are often less well placed to assess and treat antisocial behaviour effectively. It requires a careful assessment of established contributory factors, many of which require mental health expertise, such as child hyperactivity and maternal depression (Reference Rutter, Giller and HagellRutter et al, 1998), and then the treatments that are effective mostly require mental health expertise, such as parent training and medication for hyperactivity. Although considerable efforts are being made to train non-mental-health staff to deliver some of these interventions, the evidence suggests that treatments provided without high levels of skill and fidelity to the manual are much less effective (Reference Henggeler, Schoenwald and LiaoHenggeler et al, 2002).

In addition to treatment programmes for established cases of severe antisocial behaviour of the kind studied here, there is a strong case for prevention and early intervention strategies. This is because ‘lifetime persistent’ antisocial behaviour first becomes evident very early, typically when the child is 2–3 years old (Reference Broidy, Nagin and TremblayBroidy et al, 2003), and because if left untreated until adolescence, it is then very hard to correct. In the present study the younger children did not incur costs significantly less than those incurred by the older ones, so on financial grounds the sooner effective intervention is started, the greater the cost saving. By late adolescence the total cost of untreated antisocial behaviour is far higher, as it includes criminality and lost earning potential (Reference Greenwood, Model and RydellGreenwood et al, 1996; Reference Walsh, Loeber and FarringtonWalsh, 2001; Reference AosAos, 2002). High-quality early prevention can be notably cost-effective (Reference Schweinhart, Weikart and CraneSchweinhart & Weikart, 1998). In the USA the cost of crimes committed by a juvenile delinquent (under 10 years old) was estimated at $80 000 to $350 000 (£56 000–232 000), whereas rescuing a high-risk youth from this life path was estimated to save $1.7–2.3 million (Reference CohenCohen, 1998). In England, government initiatives such as SureStart are beginning to tackle this issue, although preliminary results are not encouraging (Reference Melhuish, Belsky and LeylandMelhuish et al, 2004), possibly because of the failure to adopt evidence-based interventions widely.

Methodological issues

Our study sample was reasonably large, and drawn from a range of clinics across London and in south-east England. The families showed characteristics typical of those who present children with antisocial behaviour across the UK. We used a careful, semi-structured interviewing approach (Reference Knapp, Scott and DaviesKnapp et al, 1999) which required the investigator to continue questioning the informant until the former was sure a reliable answer had been obtained, otherwise no cost was applied for the service in question. Although the economic section of the interview was not formally tested for interrater reliability, the immediately following section on psychopathology was found to have intraclass correlations of 0.71–0.82 (Reference Scott, Spender and DoolanScott et al, 2001b ). It remains possible that some parents did not remember some of the services they used, which might have led to underestimation of the costs. Also, we did not determine whether the parent was not seeking work because of the child's behaviour. Anecdotally, this happens with some frequency, in which case it results in parents failing to be economically productive and contribute taxes, and leads them to claim benefits. However, this could be a result of a lack of employable skills (52% of mothers in the sample left school before the age of 16 years).

Costs were only applied for services used by particular individuals, and were not apportioned for universal service provision. For example, schools have to spend time training teachers and playground supervisors how to deal with the increasing levels of children's antisocial behaviour and implement anti-bullying programmes; shops and institutions have to install closed-circuit television cameras and employ security staff to deter theft and destruction to property; car drivers have to pay higher insurance premiums to cover vandalism and thefts from vehicles, and so on. If these expenses were included, estimates of the annual cost of antisocial behaviour would be higher.

More generally, the study has attached costs to the services that were used and not to those that were needed. Differences between these could arise because of the limited availability of established services, and according to whether or not any service exists for the problem in question. For established services, such as (say) educational psychology, there may not be sufficient provision to meet the need, so that some children who should have received the service do not do so, and no cost is applied. This applies in large degree to mental health services, where, as noted above, only about a quarter of cases receive specialist help in England at present (Reference Ford, Goodman and MeltzerFord et al, 2003). More generally, a service has to exist to be counted as a cost, and for many aspects of antisocial behaviour there is no specific service. Thus, for example, in future someone might set up a service for siblings of antisocial children, because their lives were being made a misery by constant attacks; as it became used, this would add to the cost calculated for antisocial behaviour. For other consequences of antisocial behaviour there may never be a service, but none the less there may be an economic impact. For example, having several disruptive pupils in a class may take up much of the teacher's time and energy, thus impairing the quality of education and learning that the remainder receive; these children may then get poorer examination grades and less well-paid jobs. Again, these costs of antisocial behaviour remain uncounted.

CLINICAL IMPLICATIONS

1. ▪ Clinical and demographic characteristics that predict costs for children with severe antisocial behaviour are severity of antisocial behaviour and hyperactivity (both of which are treatable), and being male.

2. ▪ The costs and wider burden of persistent antisocial behaviour are high, and should encourage the implementation of evidence-based treatments.

3. ▪ Families carry a heavy economic burden.

LIMITATIONS

1. ▪ The study was of referred children, who represent about a quarter of all cases of conduct disorder; costs for the remaining three-quarters may be different.

2. ▪ The proportion of cost variation that could be explained by child characteristics was low; either there were other factors not measured in the study, or service responses to children's behavioural and other needs are inherently inconsistent.

3. ▪ More information is needed on the link between costs of early intervention services for children with severe antisocial behaviour and the cost savings to agencies and society of a decrease in crime, violence and possible drug misuse.