A cluster randomised controlled trial comparing the effectiveness and cost-effectiveness of a school-based cognitive behavioural therapy programme (FRIENDS)in the reduction of anxiety and improvement in mood in children aged 9/10 years

Article history

The research reported in this issue of the journal was funded by the PHR programme as project number 09/3000/03. The contractual start date was in January 2011. The final report began editorial review in January 2015 and was accepted for publication in August 2015. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The PHR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Dr Stallard reports support from Pathways, the organisation that markets the FRIENDS programme, during project dissemination. Accommodation was paid for by Pathways whilst presenting the results of this study at the ‘train the trainer’ conference in Brisbane in August 2014.

Copyright statement

© Queen’s Printer and Controller of HMSO 2015. This work was produced by Stallard et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Nature, extent and course of emotional problems in children

Anxiety and depressive disorders in children are common. The American Great Smoky Mountain Study found that 2.4% of children aged 9–16 years fulfilled diagnostic criteria for an anxiety disorder and 2.2% for a depressive disorder over a 3-month period. 1 Similar rates were found in the British Mental Health Survey, where 3.7% of 5- to 15-year-olds had a current anxiety disorder and 1% had a depressive disorder. 2 Comorbidity between anxiety and depression is common,3,4 with cumulative rates suggesting that by age 16–17 years 15–18% of children will have experienced an impairing emotional disorder of anxiety or depression. 1,4

Emotional problems have a persistent and unremitting course, with longitudinal studies highlighting that child mental health disorders persist into adulthood. In the Dunedin birth cohort study approximately 52–55% of young adults with depression or anxiety met diagnostic criteria for a mental health disorder before the age of 15 years, with 75% having a first diagnosis before the age of 18 years. 5 Childhood anxiety increases the risk of anxiety, depression, substance misuse and educational underachievement in early adulthood. 6,7 Similarly, childhood depression increases the risk of suicide, subsequent depression and substance misuse. The associated health-related burden and economic and societal costs are considerable and the need to improve the mental health of children is being increasingly recognised as a national and global priority. 8–10

Limited reach of treatment services

Improving the emotional health of children is an important public health issue that has become a major tenet of UK governmental policy. 11–13 Effective psychological treatments are available for children with mental health disorders, although few children actually receive them. 14,15 Surveys in the UK and USA found that approximately one-third of children with anxiety disorders and under half with depressive disorders had sought or received specialist help over a 1- to 3-year period. 16,17 Those with emotional disorders were less likely than those with other mental health disorders to have contact with specialist services. The limited reach and availability of specialist treatment services alongside a policy shift towards early intervention has led to a growing interest in preventative approaches and a move from clinical to community settings.

School-based preventative approaches

In the UK, almost eight million children and young people attend primary and secondary schools. 18 As such, schools provide an important environment for public health initiatives, offering the potential for delivering both primary prevention (i.e. promoting well-being and reducing the occurrence of new problems) and secondary prevention (i.e. stopping mild or moderate problems from worsening). Schools provide a familiar and natural environment, reaching a high percentage of children. Their central role in promoting emotional well-being has been emphasised in the national Social and Emotional Aspects of Learning (SEAL) initiative. 13

The effectiveness of school-based emotional health prevention programmes for primary school children has been the subject of two National Institute for Health and Care Excellence (NICE) reviews. 18,19 In the first, 31 studies were identified that adopted a universal approach (i.e. interventions were provided to all children regardless of need), with only one having been undertaken in the UK. 18 The second focused on targeted/indicated approaches in which interventions were provided to children at high risk or already displaying mild or moderate problems. 19 Ten studies that focused on internalising problems (anxiety and mood) were identified, but none was undertaken in the UK.

Both reviews found evidence that universal and targeted/indicated mental health programmes could have an effect on mental health. In terms of content, multicomponent programmes (i.e. teaching different skills such as relaxation, problem-solving and cognitive awareness) based on a clear theoretical framework, particularly cognitive–behavioural therapy (CBT), and which included some parental input (e.g. training/information) had the strongest evidence. This conclusion was also endorsed in a review of 27 randomised controlled anxiety prevention trials. 20 The results indicate that most universal, selective and indicated prevention programmes were effective in reducing anxiety symptoms. Although not formally tested, the authors note that the effects of CBT programmes were marginally larger than those of non-CBT interventions, with the median effect size for CBT programmes of 0.57 indicating a moderate effect. However, there was considerable variation in effect size between studies, suggesting that, although the content is important, mediating variables such as adherence to programme fidelity, leader rapport, levels of participation and audience appeal are also important factors that will influence effectiveness.

The reviews also identified a number of important methodological limitations, including small sample sizes, use of non-standardised mental health outcome measures and an absence of follow-up assessments. In addition, the comparative effectiveness of teacher-delivered and mental health-delivered interventions is unclear and has been directly investigated in only one study. 21 Before school-based emotional health programmes can be endorsed, implementation trials in which efficacious interventions are provided under diverse everyday conditions are required. The replication of treatment effects under everyday conditions cannot be assumed, with a number of recent school-based prevention studies failing to find positive effects on depression,22–25 anxiety26,27 and general emotional well-being. 28

The FRIENDS cognitive–behavioural therapy prevention programme

Cognitive–behavioural therapy is concerned with the way that children think about their world and the meanings and interpretations they make about the events that occur. It provides a framework for helping children to understand the link between what they think, how they feel and what they do. During CBT they are helped to explore and become aware of their cognitions and how these are associated with their feelings and affect their behaviour. This process allows unhelpful cognitions to be identified, tested and re-evaluated in more helpful ways that allow the child to experience more pleasant feelings and to become more motivated and able to face challenges and problems. In addition to understanding and challenging cognitions, CBT helps children to develop better emotional awareness, so that they can become better at identifying their different emotions, and a range of emotional management skills. Finally, CBT includes many behaviour techniques to facilitate behaviour change such as positive reinforcement, contingency management, monitoring and graded exposure.

Expected outcomes of CBT relate to each of the three core domains of the model: first, cognitive changes such as reductions in worries or negative thoughts; second, improved emotional management resulting in reduced symptoms of anxiety and depression; and, third, changes in behaviour resulting in improvements in self-esteem as the child learns to successfully overcome and cope with challenges. Another outcome that has been observed from the small group process that is an integral part of child CBT programmes such as FRIENDS is improvements in relationships and reductions in bullying.

Of the school-based CBT preventative programmes that have been developed, the FRIENDS programme is one of the better evaluated and has more consistent evidence of effectiveness. 20,29 This was noted by the World Health Organization,30 who identified the FRIENDS programme as having strong evidence of being effective as a school-based intervention for anxiety. The programme addresses a number of the issues identified in the previous reviews. It has a clear theoretical model, sufficient sessions, age-appropriate materials, enjoyable and fun activities, a structured leader manual with detailed session plans, standardised leader training, ongoing supervision to ensure fidelity, a parent session and weekly parent contact sheets.

In an initial randomised controlled trial (RCT) involving 489 children aged 10–12 years, significant post-intervention reductions in anxiety were reported following the FRIENDS programme. 21 These results were replicated in a subsequent study involving 594 children aged 10–13 years and were found to be maintained at 12 months. 31,32 The FRIENDS programme also had a positive effect on mood in the high-anxiety group. In terms of those with more significant problems, 85% of those in the FRIENDS group who initially scored above the clinical cut-off for anxiety and low mood were diagnosis free at 12 months compared with 31% in the comparison group. In the most recent study involving 692 children, the FRIENDS group demonstrated significant reductions in anxiety 3 years after the FRIENDS programme. 33 In addition, comparison between children aged 9/10 years and those aged 14/16 years showed that, although both age groups benefited from the FRIENDS programme, the younger group demonstrated the greatest changes in anxiety symptoms. 34 Although these results are promising, recent trials in Canada failed to find additional benefits for the FRIENDS programme, as either a universal or a targeted intervention. 26,27 Finally, no RCTs of the FRIENDS programme have been undertaken in the UK and so it is unclear whether the programme is effective when delivered in UK educational settings.

The current study

The systematic reviews summarised above indicate that school-based programmes can have benefits in terms of both secondary (post-intervention reductions in symptoms) and primary (preventing the development of significant symptoms) prevention. Programmes with a clear theoretical model based on CBT appear the most effective for anxiety and mood disorders. In addition, multicomponent programmes that teach children skills in different areas and which involve parents (e.g. relationship building/skill enhancing) appear particularly promising.

Of the available programmes fulfilling these criteria, the FRIENDS programme has a strong evidence base. Small-scale cohort studies of the FRIENDS programme have been undertaken in the UK and demonstrate the feasibility of delivering the programme within the UK educational system. These studies have found encouraging post-intervention results, with gains being maintained 1 year after the programme. 35,36 Similarly, a recent small-scale evaluation has found preliminary evidence to suggest that the FRIENDS programme may also have a primary preventative effect. 37

In the UK studies, the FRIENDS programme was delivered by trained school nurses from outside the school. This method of delivery is consistent with findings from a systematic review of school-based anxiety prevention programmes which noted that programme leaders were more likely to be mental health professionals. 20 However, the review also noted that, in one-quarter of studies, programmes were led by trained teachers and delivery by school staff also resulted in significant reductions in anxiety. 20 In terms of the FRIENDS programme, only one study has directly compared leader effects and found school leaders to be as effective as health leaders. 21 However, this study was underpowered and recent implementation trials of the FRIENDS programme delivered by trained school staff failed to find a positive effect. 26,27 It is therefore unclear whether school leaders are as effective as health staff in delivering the FRIENDS programme.

This trial compared the relative effectiveness of the FRIENDS programme delivered by trained school and health staff compared with usual school lessons (personal, social and health education, PSHE). The study addressed the methodological concerns identified above, had adequate power to detect predicted differences between groups and included an assessment of treatment fidelity; included follow-up at 12 and 24 months; and included an analysis of primary and secondary preventative effects and an evaluation of cost-effectiveness and acceptability. If found to be effective, the FRIENDS programme could be made widely available in the UK; it could be integrated within the school PSHE curriculum and would complement and build on other school initiatives.

Objectives

1. Primary outcome: to evaluate the effectiveness of the FRIENDS programme in reducing symptoms of anxiety and low mood at 12 months.

2. Primary outcome: to evaluate the effectiveness of the FRIENDS programme for children with low and high anxiety at baseline in terms of symptoms of anxiety and low mood at 12 months.

3. Secondary outcomes: to examine the effectiveness of the FRIENDS programme in terms of self-esteem, worry, bullying and overall well-being at 12 months.

4. Medium term: to examine the medium-term effects of the FRIENDS programme on symptoms of anxiety and low mood at 24 months.

5. Medium term: to evaluate the effects of the FRIENDS programme for children with low and high anxiety at baseline in terms of symptoms of anxiety and low mood at 24 months.

6. Medium term: to examine the effects of the FRIENDS programme on secondary outcomes of self-esteem, worry, bullying and overall well-being at 24 months.

7. Cost-effectiveness: to assess the cost-effectiveness of the FRIENDS programme in terms of health-related quality of life (and cost–utility) at 6 months.

8. Acceptability: to assess the acceptability of the FRIENDS programme including participant perceptions of usefulness, examples of ongoing skill usage and satisfaction (6 months).

Design

Preventing Anxiety in Children through Education in Schools (PACES) was a pragmatic, cluster, three-arm RCT; the three arms consisted of the FRIENDS programme delivered by health staff or school staff and the usual school curriculum (Table 1). The key difference in the two FRIENDS conditions was the person leading the sessions. In the health-led FRIENDS programme the leaders were health professionals from outside the school whereas in the school-led FRIENDS programme the leader was the teacher or a member of the school staff with responsibility for delivering PSHE.

Ethical approval, consent and trial monitoring

The study was approved by the Department for Health Research Ethics Committee at the University of Bath.

Participants, recruitment and randomisation

Interventions

Demographics and context

Outcome measures

Assessments were completed at baseline and 6, 12 and 24 months and involved a combination of child-, parent- and teacher-completed questionnaires and semistructured interviews.

Statistical analysis

Statistical analysis was conducted using IBM Statistical Product and Service Solutions (SPSS) Statistics version 22 (IBM Corporation, Armonk, NY, USA) and was undertaken blinded to allocation. Analysis and presentation of data are in accordance with Consolidated Standards of Reporting Trials (CONSORT) guidelines and in particular the extension to cluster randomised trials. 67–69

The effects on the primary outcome (RCADS score at 12 months’ follow-up) were assessed by intention-to-treat analysis without imputation. To take account of the hierarchical nature of the data, we used multivariable mixed-effects models to compare the mean RCADS score at 12 months for health-led FRIENDS with that for school-led FRIENDS and usual school provision, with adjustment for baseline RCADS score, sex and school effects. These analyses were repeated for secondary outcomes. Group comparisons were Bonferroni corrected for multiple testing.

For RCADS score we undertook a further planned analysis. We used repeated-measures mixed-effects analysis of variance models to investigate convergence/divergence between trial arms over time. We carried out preplanned subgroup analyses using interaction terms in the regression models between randomised arm and the baseline variable [RCADS score 0–48 (low anxiety), ≥ 49 (high anxiety)].

We conducted sensitivity analyses to assess the potential effect of missing data. Completion rates for all groups at 12 months were high (91.8–92.7%) although non-completers tended to be more symptomatic on our primary outcome measure (RCADS) at baseline (data not shown). Using multiple imputation methods 20 data sets were created and showed that imputation for missing data made no material difference to the overall results (see Table 10).

Qualitative analysis

Qualitative interviews were undertaken with staff from all 28 schools who received the FRIENDS programme. Children from 19 of these schools participating in the first two terms of the programme volunteered to take part in focus groups. A total of 115 children participated, a sample that was sufficient to ensure that all themes had been identified. Interviews followed topic guides informed by previous research on the FRIENDS programme to assess programme acceptability.

Interviews and focus groups were digitally audio recorded and transcribed verbatim. The transcripts were thematically analysed following the guidelines of Braun and Clarke. 70 Recordings were transcribed using NVivo 10 software (QSR International, Warrington, UK). The data were then openly coded, that is, without trying to fit them into the pre-existing coding frame. The coding was conducted at a semantic level, that is, by explicit or surface meaning of the data, without looking beyond what a participant said, and was allowed to form as many codes as appropriate.

Coding reliability and validity were checked by two researchers independently coding three randomly selected transcripts. Inter-rater reliabilities were calculated using NVivo 10. The coding agreement was in the range of 79–100%, indicating satisfactory agreement and consistency.

Transcripts were then independently coded and analysed by four researchers. A final coding framework was generated by discussion and consensus. The final analysis identified six distinctive themes relating to programme overview, programme content and delivery, the FRIENDS workbook, positive aspects of the programme, programme benefits and continued use of skills. The themes were checked for internal consistency. Further analysis involved building detailed data maps and examining data prevalence.

Economic analysis

An analysis comparing the two versions of the FRIENDS programme and treatment as usual was undertaken. The cost-effectiveness analysis was based on the primary outcome measure (i.e. cost per extra point reduction per child on the primary outcome measure RCADS) and quality-adjusted life-years (QALYs) (i.e. cost–utility analysis). The analysis of costs and cost-effectiveness was conducted using prices from the year 2013. It was carried out according to current best practice methods for conducting economic evaluation alongside trials71,72 and alongside cluster RCTs. 73,74

Both the cost analysis and the cost-effectiveness analysis were from the joint perspective of the health sector (NHS) and the education/social services sector (e.g. capturing children’s within-trial contacts with mental health services as well as those opportunity costs incurred by schools in order to participate). They encompassed resources needed to provide the intervention (teacher and health professional time, training time and materials, recruitment of schools) and estimated resource impacts of altered outcomes (e.g. mental health service consultations and treatments). The study identified, measured and valued the resource consequences of each alternative (applying opportunity costs as the main principle for valuation), including the separate identification of those costs/resources associated with the provision and evaluation of the interventions within the context of a research trial (i.e. those costs that would probably not be incurred should the programme be more widely implemented). The opportunity cost of a consumed resource is the value of the benefits foregone by using the resource for the next best alternative; conventionally, in most circumstances, market rates (e.g. pay per hour) or prices paid are assumed to represent opportunity costs.

Service use data

Individual-level resource use data was collected during interviews with parents of a subsample of the trial participants, using an adapted version of the CSRI61 at baseline and 6 months’ follow-up. The range of services assessed is summarised in Table 4.

The unit costs applied to different types of health service use and for visits to different types of professionals or services because of anxiety or depression are provided in Table 5. The two main sources for the unit costs were the Department of Health’s NHS reference costs76 (for primary care trusts and NHS trusts combined) and the Personal Social Services Research Unit’s unit costs of health and social care75 (hourly costs of patient or client contact for various types of health or social care professional).

Costing the intervention

The resource use involved in delivering the FRIENDS programme was costed using project records of staff time and other expenditure, based on as detailed a breakdown as possible of different resources used (i.e. a microcosting approach). This included the paid time of facilitators or teachers delivering the programme, the cost of their training and ongoing supervision and management, travel costs, printing costs for course booklets and an apportionment of the cost of recruiting schools. The calculated intervention costs excluded the costs of developing or adapting the new materials (these were treated as ‘sunk costs’ as it was assumed that they would not be incurred again). We also excluded the proportion of the facilitators’ delivery time that was spent completing additional research measures. The costs did, however, include the initial training costs of the facilitators (time of trainers and facilitators, room hire and subsistence). Usual school provision involved no intervention costs.

All costs were calculated as either the amount of resource used multiplied by a unit cost or the total amount incurred over the trial period divided by the number of pupils in participating classes, the number of sessions delivered or the number of schools, depending on the level at which the cost was incurred. Teacher time costs were based on hourly average pay rates for mid-grade primary school teachers, whereas the cost of the health facilitators was based on hourly actual salary costs of those employed over the relevant period of intervention delivery (see Appendix 2).

Recruitment and participant flow

In total, 45 primary schools were enrolled in the study, with 41 providing signed consent by the specified deadline. Following randomisation, one school allocated to the usual school provision arm withdrew before the baseline assessment. The remaining 40 schools were retained throughout the study and the flow of participants is summarised in Figure 1.

FIGURE 1.

The PACES CONSORT flow diagram.

Of the 1448 eligible participants, 1362 (94%) consented to participate in the study with 1339 (98%) completing baseline assessments. Of these, 489 were allocated to health-led FRIENDS, 472 to school-led FRIENDS and 401 to usual school provision. Interventions were delivered to all participating schools in the academic year September 2011–July 2012.

At 12 months, 65 (4.8%) children had left school, nine (0.7%) withdrew consent and 31 (2.3%) were absent on the day(s) of assessment, resulting in 1257 children (92.3%) being assessed.

By the 24-month assessment children had transitioned to secondary school. For this assessment we had to initiate a new recruitment and consent process. We wrote to all of our initial cohort and asked parents to opt in to this assessment by returning a signed consent form and contact details. Data were obtained from 594 children, 43.6% of those who initially consented to participate. The 24-month assessments were completed by 221 (45.2%) in the health-led FRIENDS arm, 206 (43.6%) in the school-led FRIENDS arm and 167 (41.6%) in the usual school provision arm.

School demographics

Table 6 summarises the 40 participating schools by trial arm. On average, the schools were representative of the UK in terms of academic attainment (i.e the percentage of children achieving Key Stage 2 level 4 in maths and English). However, there were more children with special educational needs [23.2% vs. 17.1%; t = 4.180, degrees of freedom (df) = 39; p < 0.001] and lower rates of pupil absence (4.4% vs. 5.1%; t = –4.513, df = 38; p < 0.001) and eligibility for free school meals (12.4% vs. 18.2%; t = –3.540, df = 39; p < 0.001) in the study cohort than the national average.

There were no significant differences for any variable between trial arms.

Outcomes

Objective 1: to evaluate the effectiveness of the FRIENDS programme in reducing symptoms of anxiety and low mood (primary outcome) at 12 months

At the 6-month asssessment, data were available from 1317 children, 96.7% of those who completed the baseline assessments. Our analysis revealed no significant between-arm differences in total RCADS score (health-led FRIENDS: baseline 26.24 (SD 15.56) vs. 6 months 22.99 (SD 14.52); school-led FRIENDS: baseline 24.91 (SD 14.32) vs. 6 months 24.32 (SD 15.95); usual school provision: baseline 26.78 (SD 16.32) vs. 6 months 24.70 (SD 15.84).

Our primary outcome was the child-reported RCADS score at 12 months. Data at 12 months were available for 1257 children, 92.3% of those who completed the baseline assessments (health-led FRIENDS 91.8%; school-led FRIENDS 92.4%; usual school provision 92.7%). Tables 10 and 11 summarise the total and subscale scores by trial arm at baseline and 12 months.

TABLE 10 - Analysis of primary outcome (child-reported RCADS score) at 12 months for health-led FRIENDS vs. school-led FRIENDS and usual school provision (PSHE)

CI, confidence interval; OCD, obsessive–compulsive disorder.

Overall group effect adjusted for sex, school and baseline RCADS score.

RCADS scale	Health-led FRIENDS, mean (SD)		Adjusted difference (95% CI), health-led FRIENDS vs. usual school provision	Usual school provision, mean (SD)		Adjusted difference (95% CI), health-led FRIENDS vs. school-led FRIENDS	School-led FRIENDS, mean (SD)		p-valuea
	Baseline (n = 489)	12 months (n = 449)		Baseline (n = 401)	12 months (n = 372)		Baseline (n = 472)	12 months (n = 436)
Depression	4.04 (2.60)	3.15 (2.53)	–0.36 (–0.76 to 0.04)	3.85 (2.77)	3.47 (2.,72)	–0.34 (–0.82 to 0.14)	3.68 (2.34)	3.34 (2.51)	0.120
Social anxiety	5.28 (3.28)	4.39 (3.32)	–0.41 (–0.94 to 0.12)	5.13 (3.26)	4.68 (3.37)	–0.79 (–1.42 to-0.16)	5.05 (3.23)	5.04 (3.43)	0.013
Separation anxiety	3.81 (3.34)	2.48 (2.94)	–0.42 (–0.97 to 0.12)	4.23 (3.45)	3.07 (3.14)	–0.42 (–1.07 to-0.23)	3.68 (3.14)	2.89 (2.96)	0.190
Generalised anxiety	5.79 (3.80)	4.43 (3.56)	–0.77 (–1.40 to –0.14)	5.97 (3.94)	5.15 (3.70)	–0.89 (–1.64 to-0.14)	5.64 (3.54)	5.19 (3.64)	0.011
Panic	2.85 (2.87)	2.03 (2.56)	–0.34 (–0.78 to 0.09)	2.97 (3.11)	2.42 (3.00)	–0.37 (–0.89 to 0.15)	2.68 (2.79)	2.33 (2.74)	0.157
OCD	4.56 (3.31)	3.43 (3.10)	–1.99 (–0.71 to 0.31)	4.61 (3.22)	3.79 (3.20)	–0.51 (–1.12 to 0.10)	4.45 (3.12)	3.99 (3.20)	0.124
Total RCADS	26.24 (15.56)	19.49 (14.81)	–2.66 (–5.22 to –0.09)	26.78 (16.32)	22.48 (15.74)	–3.91 (–6.48 to-1.35)	24.91 (14.32)	22.86 (15.24)	0.009

TABLE 11 - Analysis of primary outcome (child-reported RCADS score) at 12 months for school-led FRIENDS vs. health-led FRIENDS and usual school provision (PSHE)

CI, confidence interval; OCD, obsessive–compulsive disorder.

Overall group effect adjusted for sex, school and baseline RCADS score.

RCADS scale	School-led FRIENDS, mean (SD)		Adjusted difference (95% CI), school-led FRIENDS vs. usual school provision	Usual school provision, mean (SD)		Adjusted difference (95% CI), school-led FRIENDS vs. health-led FRIENDS	Health-led FRIENDS, mean (SD)		p-valuea
	Baseline (n = 472)	12 months (n = 436)		Baseline (n = 401)	12 months (n = 372)		Baseline (n = 489)	12 months (n = 449)
Depression	3.68 (2.34)	3.34 (2.51)	–0.02 (–0.43 to 0.38)	3.85 (2.77)	3.47 (2.,72)	–0.34 (–0.82 to 0.14)	4.04 (2.60)	3.15 (2.53)	0.120
Social anxiety	5.05 (3.23)	5.04 (3.43)	0.38 (–0.16 to 0.91)	5.13 (3.26)	4.68 (3.37)	–0.79 (–1.42 to-0.16)	5.28 (3.28)	4.39 (3.32)	0.013
Separation anxiety	3.68 (3.14)	2.89 (2.96)	0.00 (–0.54 to 0.55)	4.23 (3.45)	3.07 (3.14)	–0.42 (–1.07 to-0.23)	3.81 (3.34)	2.48 (2.94)	0.190
Generalised anxiety	5.64 (3.54)	5.19 (3.64)	0.12 (–0.52 to 0.75)	5.97 (3.94)	5.15 (3.70)	–0.89 (–1.64 to-0.14)	5.79 (3.80)	4.43 (3.56)	0.011
Panic	2.68 (2.79)	2.33 (2.74)	0.03 (–0.41 to 0.47)	2.97 (3.11)	2.42 (3.00)	–0.37 (–0.89 to 0.15)	2.85 (2.87)	2.03 (2.56)	0.157
OCD	4.45 (3.12)	3.99 (3.20)	0.31 (–0.21 to 0.83)	4.61 (3.22)	3.79 (3.20)	–0.51 (–1.12 to 0.10)	4.56 (3.31)	3.43 (3.10)	0.124
Total RCADS score	24.91 (14.32)	22.86 (15.24)	1.28 (–1.30 to 3.87)	26.78 (16.32)	22.48 (15.74)	–3.91 (–6.48 to-1.35)	26.24 (15.56)	19.49 (14.81)	0.009

Our analysis was adjusted for school, baseline symptomatology (RCADS score) and sex. There was a significant difference in adjusted mean total RCADS score at 12 months between health-led FRIENDS and school-led FRIENDS [–3.91, 95% confidence interval (CI) –6.48 to –1.35; p = 0.0004] and usual school provision (–2.66, 95% CI –5.22 to –0.09; p = 0.043). The 95% CIs include our predefined clinically important difference of 3.6 points on the RCADS. Analysis of the RCADS subscales (see Table 11) showed a difference in generalised (p = 0.011) and social (p = 0.013) anxiety but not depression (p = 0.12).

Missing data analysis

Although data completion at 12 months was very high (93.9%), an analysis of missing data was undertaken to compare baseline RCADS scores of those who did and those who did not complete the 12-month assesments (Table 12). On our primary outcome measures (RCADS), child non-completers at 12 months had higher baseline scores (indicating more symptomatology) on the total RCADS and all subscales.

TABLE 12 - Comparison of baseline characteristics for those children with/without missing primary outcome data at 12 months

OCD, obsessive–compulsive disorder.

Characteristic	Health-led FRIENDS, mean (SD)	School-led FRIENDS, mean (SD)	Usual school provision, mean (SD)	p-value (interaction)
Completers, n (%)	449 (91.8)	436 (92.4)	372 (92.8)	0.867
Non-completers, n	40	36	29
Sex male, n (%)
Completers	234 (52.1)	220 (50.5)	159 (42.7)	0.649
Non-completers	21 (52.5)	17 (52.8)	11 (37.9)
Child-reported RCADS score
Depression
Completers	4.01 (2.58)	3.62 (2.31)	3.73 (2.68)	0.001 (0.140)
Non-completers	4.41 (2.89)	4.46 (3.04)	5.43 (3.45)
Separation anxiety
Completers	3.84 (3.43)	3.63 (3.15)	4.07 (3.36)	0.015 (0.012)
Non-completers	3.54 (2.18)	4.24 (3.04)	6.29 (4.03)
Social anxiety
Completers	5.24 (3.28)	4.97 (3.14)	5.00 (3.20)	0.001 (0.298)
Non-completers	5.72 (3.24)	5.97 (4.13)	6.79 (3.65)
General anxiety
Completers	5.71 (3.76)	5.65 (3.56)	5.76 (3.79)	0.001 (0.010)
Non-completers	6.64 (4.13)	5.54 (3.35)	8.64 (4.75)
Panic
Completers	2.85 (2.89)	2.65 (2.72)	2.84 (3.01)	0.020 (0.045)
Non-completers	2.84 (2.70)	3.00 (3.54)	4.64 (3.88)
OCD
Completers	4.49 (3.26)	4.42 (3.10)	4.48 (3.17)	0.002 (0.361)
Non-completers	5.38 (3.84)	4.91 (3.38)	6.18 (3.47)
Total RCADS score
Completers	26.10 (15.66)	24.63 (14.13)	25.88 (15.78)	0.001 (0.025)
Non-completers	27.87 (13.53)	28.21 (16.32)	37.96 (18.99)

Sensitivity analyses were conducted to assess the potential effect of missing data. Twenty imputed data sets were created using imputations based on RCADS total and subscale scores (Table 13).

TABLE 13 - Primary outcome (child-reported RCADS score) at 12 months based on the pooled results of 20 data sets with imputed missing values

OCD, obsessive–compulsive disorder.

Adjusted for baseline, sex and school-level effects.

RCADS scale	Health led FRIENDS (n = 489), mean (SD)	School led FRIENDS (n = 472), mean (SD)	Usual school provision (n = 401), mean (SD)	p-valuea
Depression	3.10 (2.85)	3.39 (2.87)	3.44 (2.84)	0.163
Separation anxiety	2.55 (3.78)	2.952 (3.82)	2.900 (3.77)	0.209
Social anxiety	4.34 (3.74)	5.04 (3.69)	4.62 (3.69)	0.021
General anxiety	4.36 (4.27)	5.20 (4.28)	4.94 (4.23)	0.013
Panic	2.03 (2.99)	2.38 (3.00)	2.32 (2.96)	0.170
OCD	3.47 (3.43)	4.03 (3.45)	3.69 (3.42)	0.052
Total RCADS score	19.79 (17.18)	23.01 (17.23)	21.89 (17.00)	0.020

Imputation for missing data made no material difference to the overall results. There continued to be a between-group difference in total RCADS score and on the generalised and social anxiety subscales.

Objective 2: to evaluate the effectiveness of the FRIENDS programme for children with low and high anxiety at baseline in terms of symptoms of anxiety and low mood at 12 months

We were interested to explore the effects of the programme on those children with elevated symptoms of anxiety. Within community surveys 3–4% of children will be suffering with an anxiety disorder. In addition, a further group of children will have significant symptoms but may not fulfil all diagnostic criteria. We therefore chose to identify the 10% with the highest RCADS scores to cover both of these groups.

The distribution of child-reported total RCADS scores at baseline was examined. A total RCADS score of ≥ 49 identified 10.1% of children and was used as a cut-off to categorise children as having either high anxiety (n = 130) or low anxiety (RCADS score of ≤ 48, n = 1151). Using this cut-off, 99 high-anxiety and 1029 low-anxiety children completed the RCADS at 12 months. Table 14 summarises the total RCADS score at baseline and 12 months by trial arm for the high- and low-anxiety subgroups.

TABLE 14 - Subgroup analysis of primary outcome (child-reported RCADS score) at 12 months for high- and low-anxiety children

Adjusted for sex, school and baseline RCADS score.

Subgroup	Health-led FRIENDS, mean (SD)	School-led FRIENDS, mean (SD)	Usual school provision, mean (SD)	p-value overall group effecta
High anxiety (RCADS score of ≥ 49)	n = 36	n = 31	n = 32
Baseline	57.59 (8.18)	55.66 (7.16)	57.57 (7.90)	0.288
12 months	35.31 (19.24)	40.65 (21.40)	33.97 (21.15)	0.368
Low anxiety (RCADS score of ≤ 48)	n = 374	n = 360	n = 295
Baseline	22.78 (11.86)	22.01 (11.05)	22.51 (12.03)	0.623
12 months	17.68 (13.40)	21.06 (13.42)	20.74 (14.12)	0.006

There were significant within-group reductions for the high-risk group at 12 months but no between-group effects. For the low-risk group, there were significant within-group reductions and between-group differences in mean RCADS scores at 12 months (p = 0.006). Adjusted mean differences showed an effect for health-led FRIENDS compared with school-led FRIENDS (–3.78, 95% CI –6.16 to –1.40; p = 0.003) and for health-led FRIENDS compared with usual school provision (–3.13, 95% CI –5.61 to –0.65; p = 0.015). Post hoc analysis indicated that this related to a reduction on the social (p = 0.013) and generalised anxiety (p = 0.006) subscales in the health-led FRIENDS group.

Objective 3: to examine the effectiveness of the FRIENDS programme in terms of self-esteem, worry, bullying and overall well-being (secondary outcomes) at 12 months

At the 6-month assessment, data were available from 1317 children and 479 parents. Analysis of our secondary outcomes revealed no significant between-group differences (data not reported here).

Our primary assessment point was 12 months post baseline assessment. Table 15 summarises the secondary outcomes (child-reported outcomes and parent and teacher assessments) at baseline and 12 months by trial arm.

TABLE 15 - Analysis of secondary outcomes for all children at 12 months

Between-group differences at 12 months adjusted for baseline, sex and school-level effects.

Outcome	Health-led FRIENDS		School-led FRIENDS		Usual school provision		p-value overall group effecta
	Baseline	12 months	Baseline	12 months	Baseline	12 months
Child reported
n	489	449	472	436	401	372
Penn State Worry Questionnaire for Children score, mean (SD)	10.63 (8.14)	8.19 (7.93)	10.99 (8.24)	9.62 (8.30)	10.46 (8.35)	9.03 (8.52)	0.0136
Rosenberg Self-Esteem Scale score, mean (SD)	18.94 (5.34)	20.90 (6.22)	19.43 (5.39)	20.77 (5.82)	19.57 (5.98)	20.87 (5.95)	0.639
Total life satisfaction, mean (SD)	14.21 (6.77)	13.87 (6.00)	13.32 (5.71)	13.73 (6.08)	13.76 (6.82)	13.95 (5.88)	0.770
Bullied more than two or three times a month, n (%)	142 (29.3)	74 (16.5)	124 (26.8)	98 (22.5)	112 (28.6)	86 (23.2)	0.156
Not bullied or bullied once or twice, n (%)	343 (70.7)	374 (83.5)	338 (73.2)	337 (77.5)	279 (71.4)	285 (76.8)
Parent reported
n	217	173	198	159	152	119
Total RCADS score, mean (SD)	12.55 (8.81)	10.76 (8.90)	10.99 (8.60)	9.82 (7.13)	12.52 (9.34)	10.03 (7.31)	0.816
Total SDQ score, mean (SD)	9.09 (6.32)	7.06 (6.00)	8.31 (6.28)	6.67 (5.62)	9.00 (6.24)	7.32 (9.95)	0.767
Total SDQ threshold ≥ 17, n (%)	22 (10.5)	15 (9.3)	25 (13.0)	11 (7.3)	21 (14.4)	9 (7.9)	0.333
Teacher reported
n (%)	487 (99.6)	454 (92.8)	466 (98.7)	445 (94.3)	396 (98.8)	375 (93.5)
Teacher SDQ impact, mean (SD)	119 (24.4)	131 (28.9)	125 (26.8)	143 (32.1)	109 (27.5)	117 (31.2)	0.538

There were no between-group differences on any measure after adjusting for baseline, sex and school-level effects.

Objective 4: to examine the medium-term effects of the FRIENDS programme on symptoms of anxiety and low mood (primary outcome) at 24 months

A total of 594 children completed assessments at 24 months. A comparison of the baseline characteristics of those who did and those who did not complete the 24-month assessment is shown in Table 16.

TABLE 16 - Comparison of baseline characteristics of 24-month completers and non-completers

OCD, obsessive–compulsive disorder.

p < 0.05.

Characteristic	Completers (n = 594)	Non-completers (n = 768)	p-value
Study arm, n
Health-led FRIENDS	221	268	0.568
School-led FRIENDS	206	266
Usual school provision	167	234
Sex, n
Male	279	383	0.288
Female	315	385
Family affluence, mean (SD)	6.18 (1.46)	5.96 (1.53)	0.009a
Child baseline RCADS score, mean (SD)
Depression	3.98 (2.61)	3.77 (2.54)	0.139
Separation anxiety	3.92 (3.32)	3.86 (3.31)	0.758
Social anxiety	5.21 (3.12)	5.12 (3.37)	0.61
Generalised anxiety	5.91 (3.69)	5.69 (3.80)	0.286
Panic	2.98 (3.06)	2.71 (2.80)	0.106
OCD	4.66 (3.26)	4.45 (3.19)	0.25
Total anxiety RCADS score	22.57 (13.58)	21.76 (13.68)	0.288
Total RCADS score	26.50 (15.29)	25.52 (15.47)	0.255
Child baseline worries, mean (SD)	11.20 (8.33)	10.32 (8.14)	0.056
Child baseline self esteem, mean (SD)	19.25 (5.30)	19.33 (5.74)	0.794
Child baseline CHU-9D score, mean (SD)	8.12 (6.12)	7.76 (6.23)	0.299
Child baseline total happiness, mean (SD)	14.12 (6.58)	13.50 (6.32)	0.085
	n = 383	n = 183
Parent baseline RCADS score, mean (SD)
Depression	1.55 (1.47)	1.40 (1.74)	0.289
Separation anxiety	2.08 (2.34)	2.11 (2.69)	0.907
Social anxiety	4.38 (2.77)	3.84 (2.68)	0.029a
Generalised anxiety	2.87 (2.36)	2.70 (2.54)	0.458
Panic	0.52 (1.04)	0.69 (1.42)	0.111
OCD	1.06 (1.41)	0.93 (1.52)	0.301
Total anxiety RCADS score	10.68 (7.58)	10.16 (8.50)	0.504
Total RCADS score	12.19 (8.43)	11.59 (9.86)	0.499
Parent baseline SDQ score, mean (SD)
Prosocial	8.22 (1.83)	8.34 (1.83)	0.471
Hyperactivity	3.51 (2.56)	3.46 (2.50)	0.808
Emotional symptoms	2.21(2.27)	2.24 (2.50)	0.885
Conduct problems	1.56 (1.74)	1.57 (1.80)	0.919
Peer problems	1.48 (1.77)	1.49 (1.93)	0.978
Total SDQ score	8.75 (6.15)	8.89 (6.59)	0.81

The only differences were in family affluence (higher in completers) and parent-rated social anxiety (higher in completers).

Table 17 summarises the total and subscale mean scores for the primary outcome measure (child-reported RCADS) at baseline and 12 and 24 months. Paired t-tests were undertaken to compare within-group change over time (baseline to 24 months). Anxiety symptomatology in all groups on the RCADS total scale and each subscale, including depression, had reduced by 24 months.

TABLE 17 - Revised Child Anxiety and Depression Scale total and subscale scores by trial arm at baseline and 12 and 24 months

OCD, obsessive–compulsive disorder.

RCADS scale	Baseline, mean (SD)	12 months, mean (SD)	24 months, mean (SD)	Paired t-test baseline to 24 months p-value
Health-led FRIENDS				n = 204–19
Depression	4.04 (2.60)	3.15 (2.53)	2.63 (2.03)	< 0.001
Separation anxiety	3.81 (3.34)	2.48 (2.94)	1.82 (2.26)	< 0.001
Social anxiety	5.28 (3.28)	4.39 (3.32)	4.01 (2.59)	< 0.001
General anxiety	5.79 (3.80)	4.32 (3.56)	3.68 (2.63)	< 0.001
Panic	2.85 (2.87)	2.03 (2.56)	1.39 (2.18)	< 0.001
OCD	4.56 (3.31)	3.43 (3.10)	2.62 (2.61)	< 0.001
Total RCADS score	26.24 (15.56)	19.49 (14.81)	15.87 (11.26)	< 0.001
School-led FRIENDS				n = 180–201
Depression	3.68 (2.34)	3.34 (2.51)	2.12 (1.90)	< 0.001
Separation anxiety	3.68 (3.14)	2.89 (2.96)	1.83 (1.85)	< 0.001
Social anxiety	5.05 (3.22)	5.04 (3.43)	4.23 (3.01)	< 0.001
Generalised anxiety	5.64 (3.54)	5.19 (3.64)	3.81 (2.80)	< 0.001
Panic	2.68 (2.79)	2.33 (2.74)	1.21 (1.68)	< 0.001
OCD	4.45 (3.12)	3.99 (3.20)	2.56 (2.41)	< 0.001
Total RCADS score	24.91 (14.32)	22.86 (15.24)	15.58 (10.40)	< 0.001
Usual school provision				n = 151–64
Depression	3.85 (2.77)	3.47 (2.72)	2.56 (1.95)	< 0.001
Separation anxiety	4.23 (3.45)	3.07 (3.14)	2.36 (2.41)	< 0.001
Social anxiety	5.13 (3.26)	4.68 (3.37)	4.48 (2.70)	< 0.001
Generalised anxiety	5.97 (3.94)	5.15 (3.70)	3.89 (2.58)	< 0.001
Panic	2.97 (3.11)	2.42 (3.00)	1.40 (1.90)	< 0.001
OCD	4.61 (3.22)	3.78 (3.21)	2.75 (2.49)	< 0.001
Total RCADS score	26.78 (16.32)	22.48 (15.73)	17.40 (10.89)	< 0.001

At 24 months, there were no between-group effects on the primary outcome after adjusting for baseline, sex and school effects. Children in each condition achieved similar reductions in anxiety symptoms and depression at 24 months (Table 18).

TABLE 18 - Distribution of primary outcome (child report RCADS) at 24 months

OCD, obsessive–compulsive disorder.

Adjusted for baseline, sex and school-level effects.

RCADS scale	Health-led FRIENDS (n = 218–20), mean (SD)	School-led FRIENDS (n = 201–6), mean (SD)	Usual School Provision (n = 164–7), mean (SD)	p-valuea
Depression	2.63 (2.03)	2.12 (1.90)	2.56 (1.95)	0.069
Separation anxiety	1.82 (2.26)	1.83 (1.85)	2.36 (2.41)	0.224
Social anxiety	4.01 (2.59)	4.23 (3.01)	4.48 (2.70)	0.183
Generalised anxiety	3.68 (2.63)	3.81 (2.80)	3.89 (2.58)	0.683
Panic	1.39 (2.18)	1.21 (1.68)	1.40 (1.90)	0.697
OCD	2.62 (2.61)	2.56 (2.41)	2.75 (2.49)	0.831
Total RCADS score	15.87 (11.26)	15.58 (10.40)	17.40 (10.89)	0.182

Objective 5: to evaluate the medium-term effects of the FRIENDS programme for children with low and high anxiety at baseline on symptoms of anxiety and low mood (primary outcome) at 24 months

A total of 535 children had baseline and 24-month data available, of whom 48 had high anxiety and 487 had low anxiety at baseline. Data on child-reported total RCADS scores are summarised in Table 19.

TABLE 19 - Subgroup analysis of primary outcome (child-reported RCADS score) at 24 months for high- and low-anxiety children

Adjusted for baseline, sex and school-level effects.

Subgroup	Health-led FRIENDS, mean (SD)	School-led FRIENDS, mean (SD)	Usual school provision, mean (SD)	p-value overall group effecta
High anxiety	n = 20	n = 15	n = 13
Baseline	57.59 (8.18)	55.66 (7.16)	57.57 (7.90)
12 months	35.31 (19.24)	40.65 (21.40)	33.97 (21.15)	0.763
24 months	25.20 (14.99)	23.60 (13.14)	27.00 (15.93)	0.773
Low anxiety	n = 184	n = 165	n = 138
Baseline	22.78 (11.86)	22.01 (11.05)	22.51 (12.03)
12 months	17.68 (13.40)	21.06 (13.42)	20.74 (14.11)	0.020
24 months	14.68 (10.10)	14.62 (9.82)	16.36 (9.76)	0.184

The number of high-anxiety children with 24-month follow-up data available was low [health-led FRIENDS 20/36 (55.6%); school-led FRIENDS 15/31 (48.4%); usual school provision 13/32 (40.6%)] and appropriate caution needs to be exercised in interpreting these data. Paired t-tests revealed a significant within-group reduction in total RCADS score from baseline to 24 months for all high-risk children (health-led FRIENDS: t = 8.52, df = 19, p < 0.001; school-led FRIENDS: t = 8.96, df = 14, p < 0.001; usual PSHE: t = 6.99, df = 12, p < 0.001). However, there were no between-group effects.

Approximately half of the children identified with low anxiety at baseline completed the 24-month assessment [health-led FRIENDS 184/374 (49.2%); school-led FRIENDS 165/360 (45.8%); usual school provision 138/295 (46.8%)]. Paired t-tests revealed a significant within-group reduction in total RCADS scores from baseline to 24 months for all children (health-led FRIENDS: t = 9.10, df = 183, p < 0.001; school-led FRIENDS: t = 8.85, df = 164, p < 0.001); usual school provision: t = 7.209, df = 137, p < 0.001). However, there were no between-group effects when baseline to 24-month total RCADS scores were examined.

Objective 6: to examine the effects of the FRIENDS programme on secondary outcomes of self-esteem, worry and overall well-being at 24 months

Table 20 summarises baseline and 24-month scores for the secondary child outcomes of worry, self-esteem and happiness (life satisfaction) and parent-reported anxiety and general behaviour. The results indicate no between-group effects for any outcome.

TABLE 20 - Analysis of secondary outcomes for all children at 24 months

Between-group differences at 24 months, adjusted for baseline, sex and school-level effects.

Outcome	Health-led FRIENDS, mean (SD)		School-led FRIENDS, mean (SD)		Usual school provision, mean (SD)		p-value overall group effecta
	Baseline	24 months	Baseline	24 months	Baseline	24 months
Child reported
n	489	212–20	472	201–6	401	164–7
Penn State Worry Questionnaire for Children score	10.63 (8.14)	7.08 (6.40)	10.99 (8.24)	7.78 (7.01)	10.46 (8.35)	7.51 (5.99)	0.685
Rosenberg Self-Esteem Scale score	18.94 (5.34)	21.58 (5.71)	19.43 (5.39)	22.75 (5.37)	19.57 (5.98)	21.80 (5.35)	0.218
Total life satisfaction	14.21 (6.77)	12.05 (4.50)	13.32 (5.71)	11.69 (4.03)	13.76 (6.82)	12.26 (4.21)	0.770
Parent reported
n	217	181–8	198	184–93	152	145–7
Total RCADS score	12.55 (8.81)	13.06 (8.60)	10.99 (8.60)	11.64 (7.16)	12.52 (9.34)	11.28 (7.17)	0.719
Total SDQ score	9.09 (6.32)	8.79 (6.55)	8.31 (6.28)	7.42 (5.16)	9.00 (6.24)	7.86 (6.20)	0.161

Similarly, there were no between-group differences on the SCQ (health-led FRIENDS 43.67 (SD 25.87); school-led FRIENDS 43.49 (SD 23.94); usual school provision 45.87 (SD 25.12). Although 67 (11.2%) children had not yet transitioned to secondary school, the mean ratings for this group did not differ from those of the group who had transitioned.

Objective 7: to assess the cost-effectiveness of the FRIENDS programme in terms of health-related quality of life (and cost–utility) at 6 months

A total of 308 parents volunteered to be interviewed, with 284 providing data about service usage at both baseline and 6 months. Within this subgroup, 268 had valid child-reported RCADS data at 6 months and 273 had valid CHU-9D-derived QALY data (for 0–6 months). The differences in the baseline characteristics of those supplying service use data (interviewed) and those not interviewed are shown in Table 21.

TABLE 21 - Comparison of baseline variables between those who were and those who were not interviewed for the cost-effectiveness analysis

OCD, obsessive–compulsive disorder.

Variable	Interviewed (n = 308)	Not interviewed (n = 1054)
Study arm, n
1	122	367
2	96	376
3	90	311
Sex, n (%)
Male	151 (49.0)	511 (48.5)
Female	157 (51.0)	543 (51.5)
Family affluence, mean (SD)	6.12 (1.50)	6.04 (1.50)
Child baseline RCADS score, mean (SD)
Depression	4.16 (2.62)	3.77 (2.55)
Separation anxiety	3.96 (3.34)	3.87 (3.30)
Social anxiety	5.50 (3.07)	5.05 (3.31)
Generalised anxiety	6.19 (3.77)	5.67 (3.74)
Panic	3.29 (3.16)	2.69 (2.83)
OCD	5.11 (3.37)	4.37 (3.16)
Total anxiety RCADS score	24.18 (13.73)	21.49 (13.55)
Total RCADS score	28.35 (15.73)	25.22 (15.22)
Child baseline worries, mean (SD)	11.90 (8.76)	10.35 (8.04)
Child baseline self-esteem, mean (SD)	18.96 (5.30)	19.39 (5.62)
Child baseline CHU-9D, mean (SD)	8.95 (6.60)	7.61 (6.08)
Child baseline total happiness, mean (SD)	15.02 (7.11)	13.40 (6.19)
	n = 306	n = 254
Parent baseline RCADS score, mean (SD)
Depression	1.85 (1.78)	1.08 (1.12)
Separation anxiety	2.44 (2.74)	1.66 (1.97)
Social anxiety	4.70 (2.84)	3.63 (2.52)
Generalised anxiety	3.20 (2.65)	2.35 (2.01)
Panic	0.71 (1.35)	0.41 (0.90)
OCD	1.25 (1.55)	0.74 (1.26)
Total anxiety RCADS score	12.06 (8.63)	8.62 (6.33)
Total RCADS score	13.87 (9.82)	9.71 (6.95)
Parent baseline SDQ score, mean (SD)
Prosocial	8.06 (1.92)	8.51 (1.68)
Hyperactivity	3.74 (2.69)	3.20 (2.31)
Emotional symptoms	2.62 (2.50)	1.74 (2.06)
Conduct problems	1.77 (1.91)	1.31 (1.52)
Peer problems	1.75 (1.98)	1.17 (1.56)
Total SDQ score	9.91 (6.70)	7.46 (5.47)

Table 21 highlights that those in the group interviewed for the cost-effectiveness analysis were more symptomatic on child- and parent-reported measures on almost every scale and subscale, including the primary outcome measure (child-reported RCADS: t = 3.07, df = 1279; p = 0.002) and the health-related quality of life measure for deriving utility (child-reported CHU-9D: t = 3.36, df = 1331; p = 0.001).

Intervention costs

The health-led and school-led FRIENDS groups had similar intervention costs of £52.25 and £55.92 per student respectively (Table 22).

TABLE 22 - Per-student cost of delivering the FRIENDS programme

Resource	Health-led FRIENDS (£)	School-led FRIENDS (£)
Leader training (leaders’ time)	1197	11,572
Leader training (trainers’ time)	1299	1299
Leader manuals (printing)	159	625
Supervision of delivery (receivers’ time)	2432	7232
Supervision of delivery (supervisors’ time)	1918	1918
Supervision travel costs	0	2250
Delivery (teachers’ time)	0	3255
Delivery (facilitators’ time)	13,468	5693
Travel	9720	0
Booklet/printed materials (£4.25 per child)	2673	2784
Total	32,866	36,628
No. of classes in trial arm	24	25
Mean no. of children per participating class (data from 33 schools in the study)	26.2	26.2
Total no. of children	629	655
Mean intervention cost per child	52.25	55.92

Although the overall costs are similar, this conceals a different mix of cost components. Having fewer and externally based staff delivering the intervention in the health-led FRIENDS group meant higher travel costs and higher facilitator costs for the actual time delivering the programme (two people per class per session). In contrast, the school-led FRIENDS group needed to train and supervise more people delivering the intervention, at least one per school, but only had to pay for the extra preparation time of teachers, not their classroom intervention delivery time (as this would have already been used as part of PSHE lessons).

Health-care usage

Table 23 shows the very similar health and social care service usage and care costs for the trial arms in the 6 months before and during and after the intervention.

TABLE 23 - Parent-reported health service use for worry, anxiety or unhappiness

SE, standard error.

Resource	Health-led FRIENDS		School-led FRIENDS		Usual school provision
	No. of children (% of 122)	Mean (SE) cost (£)	No. of children (% of 96)	Mean (SE) cost (£)	No. of children (% of 90)	Mean (SE) cost (£)
Over the 6 months previous to baseline
GP visits	2 (1.6)	0.61 (0.43)	6 (6.2)	2.70 (1.13)	4 (4.4)	2.06 (1.07)
Seeing other professionals	10 (8.2)	7.02 (4.02)	15 (15.6)	6.54 (3.01)	17 (18.9)	14.73 (5.22)
Total cost of service use		7.63 (4.04)		9.24 (3.26)		16.79 (5.86)
Resource	No. of children (% of 112)	Mean (SE) cost (£)	No. of children (% of 89)	Mean (SE) cost (£)	No. of children (% of 82)	Mean (SE) cost (£)
From baseline to 6 months
GP visits	2 (1.8)	1.65 (1.36)	5 (5.6)	2.49 (1.15)	5 (6.1)	3.61 (1.99)
Seeing other professionals	10 (8.9)	9.27 (5.36)	13 (14.6)	5.88 (3.42)	14 (17.1)	12.62 (6.54)
Total cost of service use		10.92 (5.56)		8.37 (3.65)		16.23 (7.16)
Change in cost of service use between the two time periods		+3.29		–0.87		–0.56

Tables 24 and 25 show the incremental costs, RCADS scores and QALYs and incremental cost-effectiveness of the interventions. Incremental analysis compares the gain or loss in effectiveness with the additional costs, or cost savings, for one intervention compared with another.

TABLE 24 - Incremental per-student cost-effectiveness of health-led FRIENDS vs. usual school provision from baseline to 6 months

Means and SDs calculated based on non-missing participants only; between-arm comparisons based on analysis of complete data at 6 months for costs and QALYs/RCADS.

Outcome	Health-led FRIENDS, mean (SD)a	Usual school provision, mean (SD)a	Mean difference (95% CI),a adjusted	p-value	ICER (95% CI), adjusted
Cost (£)	63.68 (60.2)	11.19 (44.15)	52.50 (36.70 to 68.30)	< 0.001
RCADS score	25.61 (16.0)	27.70 (16.7)	–2.56 (–6.20 to 1.08)	0.49	18 (7 to infinity)
QALYs	0.388 (0.057)	0.390 (0.056)	–0.004 (–0.021 to 0.014)	0.69	–14,617 (3407 to –2243)

TABLE 25 - Incremental per-student cost-effectiveness of health-led FRIENDS vs. school-led FRIENDS from baseline to 6 months

Means and SDs calculated based on non-missing participants only; between-arm comparisons based on analysis of complete data at 6 months for costs and QALYs/RCADS.

Outcome	Health-led FRIENDS, mean (SD)a	School-led FRIENDS, mean (SD)a	Mean difference (95% CI),a adjusted	p-value	ICER (95% CI), adjusted
Cost (£)	63.68 (60.2)	64.37 (34.82)	0.041 (–14.01 to 14.09)	0.995
RCADS score	25.61 (16.0)	23.98 (14.0)	–1.68 (–5.80 to 2.44)	0.43	0 (undefined)
QALYs	0.388 (0.057)	0.401 (0.051)	–0.015 (–0.031 to 0.002)	0.08	–3 (undefined)

The only statistically significant difference between groups in either costs or effects at 6 months was the cost difference between health-led FRIENDS and usual school provision. Correspondingly, the ICERs have extremely wide uncertainty limits (when they can be calculated). This is clearly depicted by the scatterplots on the cost-effectiveness plane and cost-effectiveness acceptability curves for the two analyses based on QALYs (Figures 2 and 3). Compared with usual school provision, health-led FRIENDS never reaches more than a 35% probability of being cost-effective at any willingness to pay for a QALY.

FIGURE 2.

Cost-effectiveness acceptability curve and scatterplot of health-led FRIENDS vs. usual school provision, QALY outcome, adjusted analysis. (a) Cost-effectiveness plane: health-led FRIENDS vs. usual school provision; and(b) probability that health-led FRIENDS is cost-effective vs. usual school provision. Note that cost differences in GBP and effect differences in QALYs.

FIGURE 3.

Cost-effectiveness acceptability curve and scatterplot of health-led vs. school-led FRIENDS, QALY outcome, adjusted analysis. (a) Cost-effectiveness plane: health-led FRIENDS vs. school-led FRIENDS; and (b) probability that health-led FRIENDS is cost-effective vs. school-led FRIENDS. Note that cost differences in GBP and effect differences in QALYs.

Overall, whether considering the impact on quality of life or on anxiety, the FRIENDS programme is unlikely to be cost-effective, at least over the short term. However, this conclusion needs to be treated with caution as the subgroup used for the economic analysis was different in a number of ways from the total cohort (see Table 21).

We did not conduct an economic evaluation at the 24-month time point. This was because (1) there were no statistically significant between-group effects at 24 months and (2) the interview subsample of parents and their children who supplied resource use data was substantially different from the group of non-interviewed parents/children (see Table 21) and was also smaller again at the 24-month follow-up time point (only 252 parents were interviewed at this follow-up point compared with 308 at baseline).

Principal findings

This is the first large pragmatic randomised trial undertaken in the UK comparing a universally provided classroom-based CBT anxiety prevention programme (FRIENDS) led by health and school staff with usual school provision. At 12 months we assessed 92.3% of those who completed our baseline assessments. There were no baseline differences in anxiety and only one difference between our groups (sex), which was controlled for in all subsequent analyses. In view of this we feel confident in our findings, which show that, at 12 months, when transported to everyday settings, health-led FRIENDS was effective in reducing child-reported symptoms of anxiety compared with school-led FRIENDS or usual school provision. Children with high anxiety levels in each group showed reductions in anxiety symptoms over time although the effect for health-led FRIENDS was particularly evident in the low-anxiety group.

The overall completion rate for the 24-month assessment was low (43.6%) although there were few differences in baseline characteristics between completers and non-completers. By 24 months, child-reported anxiety in all three groups had reduced. There were no longer any between-group effects on total anxiety for the whole sample or in the low- or high-anxiety subgroups. Similarly, there were no between-group effects on any of the child- or parent-reported secondary outcomes.

The FRIENDS programme was accommodated within primary school timetables with all nine sessions being delivered to all participating schools. Intervention fidelity was good and the qualitative evaluation indicated that the FRIENDS programme was acceptable to teachers and children, who found the programme enjoyable and useful. The qualitative evaluation also found evidence of ongoing skill usage and a ripple effect whereby those who were not the direct target of the intervention benefited from the programme.

The cost of delivering the 9-week FRIENDS programme was £52–56 per child. Determining cost-effectiveness was difficult as the subgroup who volunteered for the economic analysis was not representative of the total cohort on key baseline assessments and health service use over a 12-month period was low. Although our results do not provide evidence of the cost-effectiveness of the FRIENDS programme over the short term, this conclusion should be treated with considerable caution.

Intervention leader

Our results indicate that outcomes depended on who delivered the programme. Despite intervention leaders in the school- and health-led FRIENDS groups receiving the same initial training and following a manualised programme, school leaders achieved smaller reductions than health leaders in child-reported anxiety at 12 months. Only one previous study has directly compared the FRIENDS programme led by trained psychologists with the FRIENDS programme led by school staff;21 although both leaders were effective the study lacked statistical power. Our study was appropriately powered and is consistent with a review in which the intervention leader was found to moderate programme effectiveness. 29 Our findings are also consistent with other implementation trials in which teacher-led FRIENDS was not found to be effective. 26,27 Although training teachers to deliver mental health prevention programmes offers a way of increasing anxiety awareness and embedding anxiety management skills within schools, our results suggest that this approach is not as effective, in the short term, as programme delivery by health professionals. Further exploration identified three potentially important differences between health and school leaders.

First, although programme fidelity was high, the home assignment was not undertaken in 40% of the school-led sessions that were assessed. Ongoing practice of newly acquired skills is an integral part of CBT programmes and the absence of this may have compromised the effectiveness of session leaders who were new to this approach. 79 However, this possibility needs to be balanced against the increased opportunities for class teachers to integrate and consolidate programme skills more intensively through their daily contact with children.

Second, although both groups had the same initial training, comparatively few teachers attended ongoing supervision, a key aspect of effective training in CBT. 80 Although teachers will be competent in some of the areas addressed in supervision, such as classroom management and engaging children of differing abilities, they will be less familiar with the underlying cognitive model. Limited ongoing opportunities to consolidate the theoretical model, share good practice, discuss challenging issues and refine skills may therefore have compromised programme delivery in the school-led group.

A further possible difference, which we did not assess in this study, was the manner in which the programme was delivered. The majority of school leaders delivered the programme only once and so were less familiar with the materials. Health leaders delivered more sessions and may have developed a more thorough understanding of the materials and become more confident and enthusiastic in their delivery. These factors have been noted as potentially important in systematic reviews and need to be assessed in future studies. 20

Although our results suggest that, in the short term, the FRIENDS programme should be led by health providers, it is possible that school staff could become more effective with additional training, practice and supervision. Further training in the cognitive model may help to develop a greater understanding of the theoretical underpinning of the programme and the subtleties of the techniques and skills taught. The importance of and rationale for home-based practice could be highlighted and inducements to attend supervision sessions might improve short-term outcomes for the school-led condition. However, this will increase delivery costs and so the relative effectiveness and costs associated with health- and school-led models of delivery need to be carefully detailed and assessed. This is important as it is often assumed that it will be cheaper and more sustainable to train and support school staff to deliver emotional health prevention programmes. Our findings suggest that this may not be the case and raises a key question about how these programmes would be funded if they are rolled out across the UK.

Universal delivery

The FRIENDS programme was delivered as a universal intervention to all 9- to 10-year-old children in participating classes. Children classified as having high and low anxiety on the basis of baseline assessments showed significant reductions in anxiety symptoms over time. There were no group effects for the high-anxiety children, although numbers were small and our study was not powered to compare between-arm differences within this subgroup. The absence of between-group effects may therefore be a power issue or alternatively may reflect regression to the mean. This issue needs to be clarified in further suitably powered studies with highly anxious children.

Our study was, however, sufficiently powered to detect differences within the low-anxiety group. We found a marked between-group reduction in favour of the health-led FRIENDS group at 12 months. Although this reduction in symptomatology was maintained at 24 months, all groups improved over time and there were no longer any between-group effects. This result is promising and suggests that universally delivered anxiety prevention programmes may result in a faster reduction in anxiety symptoms, which is maintained over time. However, our study was not able to demonstrate a primary preventative effect as we were not able to assess changes in rates of emerging anxiety disorders. This would have been very time-consuming and expensive, requiring diagnostic interviews, which could potentially be stigmatising and unacceptable within a community population. 3 Nonetheless, our study does support the use of universal anxiety prevention programmes and highlights that both high- and low-anxiety children show reductions in anxiety symptomatology at 12 months, which are maintained at 24 months. Given that fears, anxiety and stress are common in children, anxiety prevention programmes may be particularly suited to universal delivery. 29 Furthermore, universal approaches are less stigmatising and can be better accommodated within school timetables.

Programme effects

Our study found a significant effect on our primary outcome of anxiety in the health-led FRIENDS groups at 12 months. There were no between-group differences for any secondary outcome suggesting that the intervention effects were specific to anxiety. The FRIENDS programme is designed as an anxiety prevention programme and specifically develops skills known to reduce anxiety, that is, anxiety awareness and management, replacing anxiety-increasing cognitions and reducing avoidance. The specific effect that we found is therefore consistent with the underlying theoretical model and programme focus. However, given the comorbidity between anxiety and depressive disorders and the shared elements of many CBT programmes, this absence of a positive effect on depressive symptoms, worries, self-esteem and reported bullying is disappointing. 81 Although positive effects had been noted in our previous uncontrolled studies,35–37 these effects do not appear robust compared with treatment as usual. The results of this trial suggest that, although anxiety prevention programmes may have a positive effect on anxiety symptoms, they cannot be assumed to enhance the general emotional well-being of children.

The effect was also specific to child report as although parent and teacher ratings of symptomatology reduced over time there were no between-group differences. These findings may suggest a greater change in internal, anxious distress rather than in observable, anxiety-related behaviours. Changes such as these would be less likely to be identified by the SDQ, which is better at detecting behavioural rather than emotional disorders. Our results therefore highlight the difficulty of assessing changes in internal emotional symptoms and cognitions that are not directly observable by parents or teachers.

Other possible explanations include the low parental reponse rate, the lack of sensitivity of our measures and/or floor effects. The response rate for parent-completed baseline questionnaires was only 42% and as such our findings may not be representative of the total cohort. The teacher assessment involved a global rating and as such may be insensitive to specific changes in anxiety symptoms.

Further evidence that the quantitative measures did not capture changes in symptoms and behaviour is provided by the qualitative analysis. Although our qualitative data were obtained from a small, self-selected sample, all parents interviewed noted improvements in their child’s confidence, a variable that was not directly assessed by our structured questionnaires. However, parents attributed this change to increased maturity and there is some evidence from community surveys to suggest that anxiety does reduce in early adolescence. For example, using our main outcome measure (RCADS), a study of a community cohort in the Netherlands found a natural reduction in anxiety between the ages of 10 and 12 years. 82 The possibility of a naturally occurring decrease in anxiety symptoms in our cohort needs to be acknowledged, although the average total size of our reduction (9–10 points) is double that found in a general population over a similar time frame (4 points). 82

Finally, baseline symptom levels in this predominantly subclinical sample were relatively low. This will result in floor effects, which will make it difficult to detect any small but important changes in symptomatology.

Cognitive–behavioural therapy in the classroom

The FRIENDS programme is based on CBT, a model and psychotherapeutic approach that is typically used to treat mental health disorders. When providing school-based universal emotional health programmes the majority of children will be healthy and will not require treatment. It would be inappropriate to provide individual treatment in a classroom context where personal information would be publicly shared. It is therefore important to emphasise that the aim of the FRIENDS programme in this study was not to provide treatment but to use the CBT framework to help children develop emotional, cognitive and behavioural skills that they can then apply to everyday life. Developing ‘skills for life’ within the school context fits with the recognition that schools are not just concerned with developing academic skills but also have an important role in enhancing the emotional development of children. 13,83

Our qualitative analysis highlighted how teachers valued the underlying CBT model, commenting that the logical and sequential structure of the programme facilitated the gradual introduction and learning of new skills. Skills from all three core elements of the CBT model (e.g. cognitive, emotional and behavioural) were highlighted as valuable. Children particularly commented on the behavioural (coping step plan and problem-solving) and emotional (relaxation techniques) elements whereas teachers were particularly positive about the cognitive (‘red and green thoughts’) and emotional (relaxation techniques) elements. Although many teachers felt that the skills being taught in the FRIENDS programme were similar to those that children would be learning through their usual classes, a number felt that the CBT model provided a more focused approach that complemented the usual classes well.

Fit with schools

The FRIENDS programme was successfully accommodated within the timetables of our primary schools, with all nine sessions being delivered to participating classes. In primary schools there is less competing pressure on academic activities and so the delivery of emotional health prevention programmes is easier than in secondary schools. For example, in a previous study evaluating a universal depression prevention programme in secondary schools, we experienced regular requests to reduce the length of the nine-session intervention and in some schools had to shorten the programme. 84 The amount of time required to deliver the programme was nonetheless identified as an issue, particularly by teachers. Children wanted more or longer sessions whereas teachers were divided in their views. Some felt that nine sessions were not enough whereas others felt that the programme content could be compacted and delivered in fewer sessions.

In addition to securing dedicated time to focus on pupil well-being and emotional health, primary schools are facing significant financial challenges, which will impact on their ability to maintain this focus. Although the health-led FRIENDS programme resulted in quicker reductions in anxiety, the staff resources required were higher than those typically available for usual PSHE in schools. The delivery method that we evaluated involved having three people in the classroom for the FRIENDS programme compared with the single teacher who typically delivers PSHE. The education sector has been subject to significant cuts, with central funding for some well-being initiatives such as the National Healthy Schools Programme and SEAL ceasing and PSHE being removed from the statutory national curriculum. 85,86 Although the costs of delivering the nine-session FRIENDS programme are comparatively reasonable (i.e. £52–56 per child), this nonetheless represents a significant financial commitment for schools with limited funds. Thus, although schools offer a convenient location for delivering mental health programmes, primary schools will continue to have many competing demands for their limited time and resources, which will limit their ability to take up and implement prevention programmes such as this.

The above pressures also highlight a key challenge for school-based emotional health prevention, namely the need to ensure flexibility to adapt the programme to the local situation, balanced against the need to maintain programme fidelity. This is a particular issue if school staff are trained to lead the programme. They will be more directly and readily influenced by competing demands and pressures from within the school, which may result in the programme being more prone to adaptation and fidelity being compromised. With external (health) leaders there is a degree of external scrutiny and accountability. The programme would need to be contracted with the school and the number of sessions agreed so that session cancellation would therefore be harder. The health leaders would inevitably be delivering the programme in a number of schools, potentially increasing programme familiarity and fidelity.

A further consideration, likely to influence uptake and fit within schools, is the effect of emotional health prevention programmes on academic outcomes. Emotional health prevention programmes typically focus on mental and psychological outcomes and little attention has been paid to the effect on educational attainment. The primary goal of schools is to improve learning and as such they will be more interested in adopting emotional health programmes if they benefit educational attainment as well as psychological well-being.

Cost-effectiveness

Our study raises a number of questions about economic evaluations of universal mental health prevention programmes for children. Although anxiety disorders are relatively common, they tend to be poorly identified and comparatively few children receive treatment. 16,17 With poor identification, service usage amongst those with anxiety disorders will be low. Furthermore, in universal trials the majority of children will be healthy and therefore will not need to use health services. Therefore, potential savings from trials of universal interventions, where the target condition is of low frequency and service use limited, will be hard to identify. With indicated anxiety prevention programmes, which are delivered only to those identified with the condition, the benefits are easier to demonstrate, with a recent evaluation demonstrating that a targeted approach is cost-effective. 87

In view of the low service usage and the fact that the intervention was primarily aimed to prevent rather than treat existing low mood and anxiety, a time frame longer than the 6 months that we used to capture pre and post intervention service use may be required to accurately capture potential changes in outcomes and related health service use. Similarly, cost-effectiveness analyses of universal approaches need to include a wider range of potential resources other than the typical focus on health and social care resources. These include education service usage, parental productivity loss and loss of child’s leisure time, factors that we did not assess in our current study.

Although service use data were not collected at 12 months’ follow-up, the main trial had estimated a statistically significant effect on the trial’s primary outcome, RCADS, at this time point. In a speculative analysis it is possible to compare this favourable effect (–2.66, 95% CI –5.22 to –0.09; p = 0.009) with the cost of delivering the health-led FRIENDS programme compared with usual school provision. This implies an ICER of £20 per point reduction in RCADS score. As with all such cost-effectiveness estimates in relation to clinical outcome scores, it is impossible to know whether this represents good value for money.

Although the estimated mean QALY gain at 12 months’ follow-up was very small and imprecise (0.0093, 95% CI –0.007 to 0.026; p = 0.265), this would imply an ICER of £5600 per QALY gained. If the mean additional cost of the intervention is £52, then the lowest mean QALY gain required for the ICER to be < £30,000 per QALY (the current policy-making threshold of affordability for the NHS in England) is 0.0017. If such small mean gains could be demonstrated with greater precision then relatively cheap group-delivered programmes such as the FRIENDS programme are more likely to be judged as cost-effective without assuming downstream savings from any improvements in mental health.

Strengths and limitations

Our study has many strengths. We used a manualised anxiety prevention programme that has proved efficacious and has been piloted in UK schools. 29,35–37 We had a large and appropriately powered cohort, recruitment and retention at 12 months were high, absenteeism was low and programme fidelity was good. However, our study does have a number of limitations.

First, children were not blind to the intervention that they received. We relied on child-completed self-report measures and did not undertake any diagnostic interviews. Although anxiety symptoms were reduced in our health-led FRIENDS group at 12 months, it is unclear whether this reflected changes in diagnostic status or impairment. Similarly, it is unclear whether this statistically significant change in symptom score reflects clincally important changes in everyday functioning.

Second, although our study included a representative sample of UK schools, our cohort was less disadvantaged and had more white British participants than the average UK state school. It is therefore unclear whether similar results would be obtained with a more disadvantaged or ethnically diverse population. Although our schools were well matched on a number of objective dimensions, there could be differences in more subjective factors such as school culture and ethos, which we did not comprehensively assess.

Third, although the health- and school-led FRIENDS facilitators had the same initial training, the health-led group had lower levels of anxiety at 12 months. This suggests that a manualised programme may result in different outcomes depending on who delivers it. Exploring reasons for this are important and, although our study assessed content fidelity, that is, the number of core tasks delivered, we did not assess how they were delivered. Health leaders may therefore have been more familiar with the materials and more enthusiastic in their delivery than school leaders. Additionally, although our qualitative analysis included the views of children, parents and school staff, we did not directly assess the views of the health leaders. It is therefore unclear why health leaders were more effective at 12 months or whether, with additional input, teachers could become more effective at delivering the FRIENDS programme.

Fourth, although recruitment and retention were very high at 12 months, we were able to assess only 43.6% of our cohort at 24 months. This is disappointing but understandable given that this assessment occurred after our cohort transitioned to secondary school and we had to initiate a new recruitment and consent process. All parent contact was made through schools as we did not have direct access to the home details of our study participants. We are therefore unclear whether our poor uptake was because information was not being passed by the schools to parents or because parents were actively refusing to consent. Although there were few differences between 24-month completers and 24-month non-completers on baseline variables, with such a low response rate our 24-month data will lack statistical power and need to be treated with appropriate caution.

Our qualitative evaluation provides additional data to complement our quantitative results and is the first to assess the views of children, teachers and parents. We used a robust qualitative methodology for data collection, analysis and reporting and data were obtained from a large sample, allowing saturation of dominant themes. However, the children who participated in our qualitative interviews were self-selected and, as such, may be more vocal, confident and positively disposed to the FRIENDS programme. Similarly, the parents who we interviewed were selected from a subgroup of more engaged parents who volunteered for additional interviews. These parents may therefore have been more motivated and engaged with our project and their views may be different from those of the wider population.

Finally, our cost-effectiveness analysis was carried out according to current best practice for conducting economic evaluation alongside randomised trials. 71–74 However, data for the economic analysis were obtained from a subgroup of our cohort by retrospective recall rather than using real-time diaries or service/practitioner records. Our subsequent analysis identified that, although we had a large sample, at baseline this group was not representative of our whole cohort on a number of factors, including our primary anxiety (RCADS) and utility (CHU-9D) outcomes. We are therefore not confident that the cost-effectiveness and service use data are representative of our full cohort.

Main findings

1. Our study suggests that a school-based anxiety prevention programme informed by CBT (FRIENDS) is effective in the short term when transported to everyday schools. The programme had good acceptability and was delivered with fidelity although teachers were unsure about the amount of time (9 hours) needed for the programme.

2. Effectiveness depends on who delivers the programme, with school leaders being less effective than health leaders.

3. At 24 months’ follow-up anxiety had reduced across all trial arms although significant reductions were achieved more quickly in the health-led FRIENDS group. In this group anxiety symptoms had reduced by 12 months and were maintained at 24 months.

4. The effects were specific to anxiety and were observed only for child-completed measures. Although parents and teachers reported a number of more general improvements during the qualitative interviews, these were not reflected in the standardised assessments that we used. Anxiety prevention programmes cannot therefore be assumed to enhance the general emotional well-being of children.

5. Fears, anxiety and stress are common in children and our study suggests that anxiety prevention programmes may be particularly suited to universal delivery. At 12 months there was evidence that the FRIENDS programme particularly benefited children with low levels of anxiety.

6. We were unable to assess the cost-effectiveness of the FRIENDS programme. Although our sample for the economic analysis was large, it was more symptomatic and differed from our full cohort on almost every measure. However, health service usage within this predominantly healthy group was low. It is therefore not possible to conclude whether the level and certainty of the additional effectiveness gains in terms of RCADs score would warrant the additional costs of providing the intervention in schools.

Implications

The FRIENDS anxiety prevention programme is acceptable to children and school staff and can be implemented with good fidelity in primary schools. Short-term effectiveness depended on who delivered the programme, with quicker reductions in anxiety occurring when the FRIENDS programme was delivered by health leaders. Children with low levels of anxiety benefited from the programme, suggesting that a universal delivery approach may be beneficial in terms of primary as well as secondary prevention. Although the cost-effectiveness and longer-term benefits of the programme need to be demonstrated, our results suggest that universal delivery will result in significant costs that may be beyond the finances available to most schools.

Research recommendations

Our study raises a number of questions for future research.

• Although using the same programme, school leaders were less effective than health leaders. Further research should explore potential programme moderators such as delivery variables (e.g. leader confidence, understanding of CBT and enthusiasm), school factors (e.g. school ethos and commitment to emotional health) and student variables (e.g. sex, motivation and disruption).

• Given the pressures on school time and the concerns of teachers about the length of the programme, identifying the core ‘active ingredients’ and exploring how these can be delivered to maximise engagement and use time allocated effectively would be helpful.

• Although children with high levels of anxiety showed reduced symptoms our sample was not sufficiently large to reliably determine whether there were between-group differences in this subgroup. The effectiveness of the FRIENDS programme delivered as a universal intervention for high-anxiety children needs to be investigated.

• Our cohort consisted of predominantly white British children and was less disadvantaged than the UK population as a whole. The effectiveness of anxiety prevention programmes based on CBT needs to be demonstrated with more ethnically diverse and disadvantaged children and with those who have additional leaning needs.

• The cost-effectiveness of universal anxiety prevention programmes needs to be determined. Given the low service usage that we found, cost-effectiveness needs to be assessed over a longer time frame and should capture a wide range of resources including health, social care and educational resources and impacts on parental productivity.

• The effect of emotional health prevention programmes on academic outcomes needs to be investigated.

• Future studies should define more clearly the content of interventions received by comparison groups. This will determine any differences and overlaps between the content of usual school classes and the content of active interventions.

Contribution of authors

All authors had access to all study data and participated in interpretation of the findings, contributed core ideas and were involved in critically revising the paper for important intellectual content. All authors read and agreed the final report.

Paul Stallard (Professor of Child and Family Mental Health) was the chief investigator on the study and took overall responsibility for the study and the writing of the report.

Elena Skryabina (Research Fellow) was the trial manager and contributed to the design of the 24-month follow-up and led the qualitative analysis.

Gordon Taylor (Reader in Medical Statistics) contributed to the design of the study and undertook the statistical analysis.

Rob Anderson (Associate Professor of Health Economics and Evaluation) contributed to the design of the study and led the cost-effectiveness analysis.

Obioha C Ukoumunne (Senior Lecturer in Medical Statistics) undertook the analysis of costs and cost-effectiveness.

Harry Daniels (Professor of Education) contributed to the design of the study and advised on educational issues.

Rhiannon Phillips (Research Fellow) contributed to the design of the study and advised on the qualitative analysis.

Neil Simpson (Consultant Paediatrician) contributed to the design of the study and advised on intervention delivery.

Publications

Stallard P, Taylor G, Anderson R, Daniels H, Simpson N, Phillips R, et al. School based intervention to reduce anxiety in children: study protocol for a randomised controlled trial (PACES). Trials 2012;13:227.

Stallard P, Skryabina E, Taylor G, Phillips R, Daniels H, Anderson R, et al. Classroom-based cognitive behaviour therapy (FRIENDS): a cluster randomised controlled trial to Prevent Anxiety in Children through Education in Schools (PACES). Lancet Psychiatry 2014;1:185–92.

Stallard P, Taylor G, Anderson R, Daniels H, Simpson N, Phillips R, et al. The prevention of anxiety in children through school based interventions: study protocol for a 24 month follow-up of the PACES project. Trials 2014;15:77.

Data sharing statement

The project data archive can be obtained from the corresponding author.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the PHR programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the PHR programme or the Department of Health.