Cognitive behavioural therapy and short-term psychoanalytic psychotherapy versus brief psychosocial intervention in adolescents with unipolar major depression (IMPACT): a multicentre, pragmatic, observer-blind, randomised controlled trial

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 06/05/01. The contractual start date was in December 2009. The draft report began editorial review in December 2015 and was accepted for publication in September 2016. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Bernadka Dubicka is a member of the Health Technology Assessment Mental, Psychological and Occupational Health Panel and has received personal fees as a consultant to Lundbeck. She also has a licensed patent: there is a licence to Lundbeck to use brief psychosocial intervention in their current trial (future payment anticipated). Peter Fonagy is in receipt of a National Institute for Heath Research Senior Investigator Fellowship. Paul Wilkinson has received personal fees as a consultant to Lundbeck, a consultant to Takeda and a supervisor in interpersonal psychotherapy. He has also had non-financial support from the interpersonal psychotherapy UK Training Committee. Ian M Goodyer has received personal fees as a consultant to Lundbeck, is supported by a strategic award from the Wellcome Trust, research support from the Friends of Peterhouse and is senior scientific advisor to and chairperson of the Peter Cundill centre for research into mood disorders in young people, University of Toronto, ON, Canada. Raphael Kelvin has received personal fees as a consultant to Lundbeck.

Copyright statement

© Queen’s Printer and Controller of HMSO 2017. This work was produced by Goodyer 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.

Unipolar major depression in adolescents

Unipolar major depression is a significant mental illness affecting a substantial proportion of the adolescent population worldwide. 2 The disorder presents in episodes and the estimated 12-month period prevalence of major depression episodes in teenagers is 7.5%. Approximately twice as many girls as boys are affected and an estimated one in four cases experience a severe, impairing and clinically referable condition. 3,4 There is a growing concern based on longitudinal evidence that the consequences of some adolescent emergent major depression episodes include suicide, persistent and chronic mental health disorders, substance misuse, and failure to achieve both educationally and in the work place. 5 Furthermore, in adults there is evidence that a history of depression may interfere with treatment compliance and self-care in patients with type 2 diabetes and cardiovascular disease,6 which suggests that reducing incident risk for depression early in life may have wider physical health-care benefits for later in the life course. These medium- and long-term negative outcomes also come at great economic cost to the UK and other nations, including in the developing world where depression has been noted to be at least as disabling as any other chronic illness in adult life as urbanisation increases. 7,8 Therefore, clinical methods for the treatment of depressed adolescents must go beyond short-term remission of a single episode; objectives should include reducing the risks for diagnostic relapse and recurrence risk by lowering depressive symptoms before independent adult life.

The combined effect of a high level of emerging mental illness with long-term consequences for health, together with the increasing demands for treatment from adolescents and their parents/guardians, makes it imperative to provide effective interventions that can be implemented by developing the current mental health workforce. 9,10 From this policy perspective it is also essential to consider the extent to which any effective treatment is deliverable and affordable. Currently, the cost-effectiveness of treatments aimed at reducing recurrence risk by lowering depressive symptom rate and avoiding diagnostic relapse up to 18 months after entering treatment are not known.

Are there effective treatments for depressed adolescents?

Over the past 20 years there have been a series of important randomised controlled trials (RCTs) determining both the efficacy and clinical effectiveness of psychological and pharmacological treatments for depression in adolescents that result in remission in the short term, that is by 28 weeks. 11,12 Original guidance compiled in 2005 by the National Institute for Health and Care Excellence (NICE) for the treatment of a moderate to severe depression episode, referred to and treated in NHS Child and Adolescent Mental Health Services (CAMHS), advised the use of evidence-based psychological therapies, such as cognitive–behavioural therapy (CBT), as the first-line treatment, with selective serotonin reuptake inhibitor (SSRI) antidepressants constituting the pharmacological treatment of choice only if there was no satisfactory response. 13 The 2015 revisions now recommend that a SSRI may be used as a first-line treatment in combination with psychological therapy (individual CBT, interpersonal therapy, family therapy or psychodynamic psychotherapy) that runs for at least 3 months for a major depression episode, defined as five or more symptoms (of which one must be a depressed or irritable mood state of at least 2 weeks, most days of the week and most hours of the day) and associated with observable personal impairment. 14 However, the revised 2015 guidelines continue to warn against the use of SSRIs on their own. Furthermore, this revision applies only to patients with moderate to severe depression, defined as five or more symptoms associated with concurrent impairments in personal, social or educational life domains for longer than 2 weeks.

Research studies in adults have provided evidence for the clinical effectiveness of a number of psychological therapies for inducing clinical remission for adult patients suffering with a moderate to severe depression episode. These include CBT,15 interpersonal therapy, short-term psychoanalytic psychotherapy (STPP)16 and non-directive brief psychosocial interventions (BPIs). 17 Overall, psychological therapies appear to be effective in a broadly equivalent manner, although results vary with methods and measures. 18 Current evidence indicates that findings from adult patients cannot be assumed to reflect comparable efficacy or effectiveness for depressed adolescents. 19

Evidence to date from RCTs on adolescents suggest that CBT is not rapidly therapeutic in the acute phase of treatment. 11 Furthermore, in the short term, CBT may not provide added clinical value when patients are already receiving fluoxetine plus active specialist clinical care in a UK CAMHS setting. 12 There is relatively little evidence on the use of STPP with children or adolescents, although the one clinical trial with this population had encouraging outcomes. 20

Interpersonal psychotherapy is effective for adolescents with mild to moderate depression in the short term, but there is no evidence for efficacy in patients with moderate to severe depression episodes. However, interpersonal psychotherapy for adolescents (IPT-A) is not widely available on the NHS in the UK to treat depression in adolescents, although this problem is being addressed as part of the Improved Access to Psychological Therapies initiative of NHS England. 21 Relatively brief, problem-solving, approaches with a focus on promoting good interpersonal relationships may be of value in more severe forms of these illnesses but this remains to be fully evaluated. 22,23

Brief psychosocial intervention incorporates general principles from psychological therapies (e.g. agenda setting, problem-solving and facilitating relationships with peers, school and family). BPI has recently been formalised into a manual for systematic delivery within CAMHS clinics for the treatment of major depression aimed at inducing short-term remission. 24 A previous RCT reported that BPI combined with fluoxetine is as effective as BPI combined with fluoxetine and CBT in producing remission at 28 weeks of treatment. 12 It is not known if BPI alone is efficacious and clinically effective in the short term for patients suffering from a moderate to severe depression episode.

Only one study has tested the efficacy of psychological treatments against placebo and showed that time to remission is significantly quicker for SSRIs and SSRIs plus CBT against pill placebo. 11 All RCTs to 2016 have focused on establishing clinical effectiveness in reducing immediate symptoms and restoring personal functioning. 11,12,25 Although there is evidence for clinical effects, there is no evidence that any of the aforementioned treatments, individually or in combination, is efficacious in reducing recurrence risk by lowering depressive symptom rate or diminishing diagnostic relapse in the medium term (i.e. ≥ 1 year following intervention).

This lack of understanding about therapeutic effects on recurrence risk and clinical relapse is compounded by a remaining concern regarding the extent of the effectiveness of existing treatments and the natural history of these disorders. 4,5,26 RCT data to date have shown that, even when treatment is successfully delivered, a substantial number of depressed adolescents do not recover or they relapse following recovery. Thus, the proportion of depressed patients who meet the criteria for clinical remission did not rise above 70% by 28 weeks from the start of treatment. Indeed, in studies of moderate to severely depressed adolescents, one RCT in the UK reported that only 42% were very much or much improved by 12 weeks, rising to 53–61% by 28 weeks. 27 Of the remaining patients, a further 30% described themselves as no better or worse at 12 weeks, which fell to 18–24% by 28 weeks. 27 A recent meta-analysis noted that, compared with CBT alone, the combination of fluoxetine and CBT may produce greater improvement in psychosocial functions but no greater reduction in residual symptoms. 28 Even for successfully treated cases, there is a high relapse rate in the next 5–10 years. Overall, approximately 50–70% of patients attending a NHS CAMHS clinic may relapse in the 10 years between mid-adolescence and young adulthood, coinciding with some of the largest educational milestones and social changes they may face over their lifetime. 4,5,26

To date, there have been two naturalistic follow-up investigations of psychological treatment effects in the medium term (i.e. > 52 weeks) in depressed adolescents entered into RCTs. Both of these studies show that the likelihood of recurrence and relapse of diagnosis following successful treatment is substantial, occurring in 50–75% of treated patients, beginning within 1 year of clinical remission and being significantly higher in patients with recurrent versus single episodes. 29–31 These studies were, however, conducted with < 150 patients each and did not plan to investigate the relative effects of treatment in maintaining reduced depressive symptoms in the medium term.

There are no studies in the NHS that have investigated clinical effectiveness of specialist or general psychological treatments in reducing symptomatic recurrence risk or diminishing clinical diagnostic relapse of depression in the medium term up to 18 months following treatment. Therefore, it is unclear if there are superiority effects of one psychological treatment over another in reducing and maintaining lower depressive symptoms over time and, therefore, diminishing symptomatic recurrence risk.

Theoretically, psychodynamic practitioners suggest that potentially more enduring changes will be associated with this form of psychological treatment, as it aims to address and repair underlying mental models of interpersonal relationships that may be associated with the depressive symptoms. In a similar way, cognitive–behavioural therapists theoretically aim to teach new, more adaptive, methods of behaving and thinking, which should continue after therapy ends and, thus, reduce long-term relapse of disorder. These forms of therapy are therefore hypothesised to be more likely, if successful, to predict more enduring recovery than therapies simply addressing current symptom reduction, alleviating the impact of provoking life events and difficulties, providing psychoeducation and offering problem-solving advice.

Rationale for the current study

The current study was devised knowing that there is now a clear evidence base for implementing psychological treatments that are clinically effective for inducing short-term remission but whose efficacy for (1) reducing recurrence risk indexed by rising depressive symptoms in the medium term 18 months after treatment began or (2) preventing clinical diagnostic relapse after treatment is completed, remains unknown. The literature implicates a number of candidate specialist psychological treatments for putative effects on reducing recurrence risk and relapse rates, among which are CBT and STPP. Both treatments aim to reduce symptoms and future risk of relapse. Given clear-cut evidence that, in depressed adolescents, active psychological treatments are clinically more effective than no treatment, a pragmatic effectiveness superiority trial was conceived as the best design. The standard treatment chosen as the reference therapy was BPI, a relatively brief (i.e. maximum of 12 sessions) psychosocial approach to problem-solving, mental hygiene and well-being management with education about depressive illnesses. BPI was designed to be shorter than specialist psychological therapies and is aimed, theoretically, at gaining remission as quickly as possible from the depression episode. Therefore, the therapeutic protocol focuses on practical advice giving, psychoeducation about depression and how to manage daily life challenges.

Therefore, we designed a RCT to test the risk reduction effect of two specialist psychological treatments against a BPI primarily focused on short-term clinical remission. In each of the three arms of the RCT, SSRIs were available as a combination treatment option following the 2005 NICE guidelines. 13

The study must allow for the prescribing of SSRIs because short-term clinical effectiveness is also achievable with fluoxetine without the addition of a protocol driven psychological therapy. 11 However, as with psychological treatments, the contribution of SSRIs to reduce recurrence risk and relapse rate remains unknown. There is also evidence that cases which have been resistant to other antidepressant medication may show significant clinical improvement with a change to a different SSRI, if prescribed in conjunction with CBT. 25 This is the only published study25 of treatment resistance in depressed adolescents and suggests that medium-term treatment goals may be best achieved by combination therapies. This remains to be evaluated in a systematic RCT in which reduction in recurrence risk and relapse rates is the clinical outcome objective. Nevertheless, in the current study we judged it essential to abide by current NICE guidelines14 for the NHS and allow SSRI prescribing within each arm based on clinical judgement of psychological treatment progress.

Given the importance of cost-effectiveness and deliverability within the NHS, we also included an economic evaluation component to the trial. Results from a previous RCT conducted in the USA suggest that CBT alone is relatively expensive and cost-ineffective compared with a SSRI alone. 32 The same RCT demonstrated that fluoxetine alone was much more cost-effective than combined fluoxetine and CBT over 12 weeks. 32 In the UK, a RCT reported that adding CBT to an existing combination treatment of fluoxetine plus active specialist clinical care (the non-manualised forerunner of BPI) in a CAMHS setting was not cost-effective over 28 weeks. 33 Whether or not more expensive psychological treatments may become cost-effective in the medium term by diminishing the subsequent use of health, education and social services more than the use of a ‘treatment as usual’ or BPI protocol is not yet known.

For this study, clinic-referred depressed adolescents only were considered for recruitment. To make the results applicable to NHS CAMHS services, we included those with suicidal thoughts, psychotic behaviours and non-depression comorbid disorders. This is comparable to the Adolescent Depression Antidepressant and Psychotherapy Trial (ADAPT)12 but distinguishes both of these UK studies from the major study of adolescent depression [Treatment of Adolescent Study (TADS)11] in the USA, as suicidal and psychotic cases were excluded from TADS and participants were recruited by advertisement. 11,27,34

Participants were recruited from patients referred to routine NHS clinics. All trial participants were treated in standard clinical settings using NHS staff to deliver treatments under supervision. This maximised ecological validity and generalisability to NHS settings and was intended to assure commissioners and providers that the study results could inform routine NHS service design, delivery and implementation.

Unlike the UK ADAPT RCT,12 those who responded to an initial phase of BPI (2–3 weeks) and who therefore may have been close to remission or were especially responsive, were not excluded. This is because all participants were prescribed fluoxetine in ADAPT and it was in keeping with the clinical practice at the time of offering a BPI before patients were committed to a 6-month course of fluoxetine. At the time of starting Improving Mood with Psychoanalytic and Cognitive Behaviour Therapy (IMPACT), it was standard practice (in keeping with NICE guidelines) to start treatment with specific psychological therapy. In addition, unlike prior UK RCT studies,12,35 we added potential moderators of recurrence risk and relapse rates. These were individual differences in (1) self-reported rumination (persistently brooding or dwelling, often to the exclusion of other themes in the patient’s life)36,37 and (2) depressive thinking style (the extent to which patients with clinical depression may be characterised in terms of immaturity of the cognitive styles of self-criticism and perfectionism). 38–40

Aims and objectives

This superiority powered pragmatic effectiveness RCT was designed to determine whether or not psychological treatment delivered to adolescents with unipolar major depression would reduce the risk of recurrence through lowering depressive symptoms by 86 weeks post randomisation. As this was designed as a superiority effects trial, we tested whether or not, compared with a standard BPI, two specialist psychological treatments were more likely to result in lowering the rate of depressive symptoms from randomisation to 86 weeks after beginning treatment.

The primary objective contained two related questions. The first was to determine whether or not, in a specialist CAMHS setting, CBT and STPP were superior to a standardised BPI in reducing self-reported depressive symptoms over time. The second was to test if there were non-inferiority effects of CBT compared with STPP for lowering depressive symptoms by 86 weeks. The secondary hypothesis tested whether or not self-reported anxiety symptoms and research interviewer-evaluated psychosocial function were also significantly more likely to remain improved in those receiving STPP or CBT as opposed to BPI. An additional objective was to evaluate whether or not these specialist individual psychological treatments were more effective than standard BPI at reducing the clinical diagnostic rate for a major depression episode by 18 months after randomisation. However, the study was not powered to consider this outcome as a formal hypothesis.

Specific hypotheses

The hypotheses of the trial relate to the treatment effect in the post-treatment period (≥ 36 weeks).

The study had four primary hypotheses.

When comparing CBT with STPP:

1. CBT will show non-inferiority effects compared with STPP at 52 weeks

2. STPP will show superiority effects compared with CBT at 86 weeks

and when comparing CBT and STPP with BPI:

1. the specialist intensive interventions (CBT/STPP) will show superiority effects compared with BPI at 52 weeks

2. the specialist intensive interventions (CBT/STPP) will show superiority effects compared with BPI at 86 weeks.

This RCT tested a primary superiority clinical effectiveness hypothesis that, when compared with the reference BPI treatment:

• STPP and CBT are both more clinically effective at maintaining a reduction in depressive symptoms at 52 and 86 weeks reassessment after randomisation.

The secondary hypothesis tested was:

• STPP is superior in maintaining a reduction in depressive symptoms over the follow-up assessments (52 and 86 weeks) compared with CBT.

The RCT also tested an economic hypothesis to determine:

• whether or not any additional costs of specialised treatments accrued by the end of treatment are justified by decreased use of resources (health, education and social care services, voluntary agencies) by 86 weeks of follow-up.

Participants and patient involvement

Patients from a prior treatment study provided advice on formulating the research question and evolving the BPI manual described in Chapter 3. Patients did not contribute to the recruitment procedure but those recruited provided ongoing feedback on the burden of assessment over the course of the study. This resulted in a 30–50% time reduction in the length of reassessment in the post-treatment follow-up phase. All participants and their families will receive a summary document of the main findings. Families will be informed that the full details of the study will also be posted on a weblink41 and be available through mandatory open access arrangements for published work under the auspices of the University of Cambridge, Cambridge, UK.

The research teams in each of the three regions contacted their regional clinics on a weekly basis to ascertain if there were any patients referred. The first point of contact was with clinical NHS staff who determined if their patient was in scope for the RCT. An initial screening checklist for major depression episode provided to the clinics by the research team and used to inform the research assessors of potential cases aided this clinical task (Figure 1).

FIGURE 1.

Screening to assess suitability for full assessment and recruitment.

Recruitment

Participants were recruited if they met criteria for an episode of DSM-IV major depression and were aged between 11 and 17 years. This diagnosis is achieved by the presence of at least five symptoms, one of which must be a mood symptom present nearly every day and most of the day for at least 2 weeks, together with four others and accompanied by observable personal and/or social impairment. The criteria are shown in Figure 2. Mood change plus four other symptoms are required for the diagnosis.

FIGURE 2.

The DSM-IV criteria for major depression disorder.

Participants were recruited when clinical staff considered a patient to be depressed and requested that they completed the research checklist and asked them and their parents/guardians if they would consider taking part in a RCT investigating the extent to which treatment was able to lower recurrence risk and relapse rate. They were informed that the trial only included and compared treatments already known to contribute to producing clinical remission. If they expressed interest in the study, their contact details were passed to the research group and research staff contacted the patient with an expression of interest letter and a reply-paid envelope.

Inclusion criteria

• Aged 11–17 years.

• Current diagnostic episode of DSM-IV unipolar major depressive disorder (MDD).

Patients with suicidal intent, past or recent suicidal behaviour, psychotic symptoms or any comorbidity, other than those specifically defined in the exclusion criteria below, were included.

Patients who met the inclusion criteria but had started a SSRI within 1 month were included.

As part of the screening process prior to enrolment, individuals were asked if their current depressive illness was a first episode or a relapse.

Exclusion criteria

• Generalised learning difficulties.

• Pervasive developmental disorder.

• Pregnancy.

• Currently taking another medication that may interact with a SSRI and unable to stop this medication.

• Substance abuse.

• A primary diagnosis of bipolar type I, schizophrenia or eating disorders.

Individuals who had received a psychological therapy consistent with the trial protocol for CBT, STPP or BPI were excluded.

Overall, 470 individuals were recruited and provided written informed consent, as did their parents/guardians. Ethics approval was by the Cambridgeshire 2 Research Ethics Committee, Addenbrooke’s Hospital Cambridge, UK. Follow-up was undertaken with repeated reassessments at nominal points time periods set at 12, 36, 52 and 86 weeks after randomisation to evaluate recurrence of self-reported depressive symptoms and enable re-evaluation of clinical diagnosis of an episode of major depression.

Psychopathology

Health economic measures

Planned interventions

IMPACT was a pragmatic superiority trial comparing the relative clinical effectiveness of three psychological treatments each with evidence of clinical efficacy being associated with clinical remission in the short term (i.e. 3–6 months). These treatments are available in CAMHS NHS practice, although distribution around the UK is not standardised. The three treatment approaches tested in this study were all manualised.

A duty of care by clinical staff to patients was observed in all clinical arms. This included parent/guardian support and engagement, explanation of treatment principles, maintenance and support of family during individual treatment, individual risk management strategies and contact with other agencies, if appropriate.

Comprehensive treatment protocols were developed for the trial and designed for delivery by practitioners working in routine NHS CAMHS settings. The rationale for using treatment manuals as guides to therapy is that manuals:

• aid dissemination of treatment methods into clinical practice

• help to standardise the intervention between therapists and across sites

• form the basis for audio tape ratings of treatment adherence and differentiation; thus, ensuring that the interventions have been given properly in keeping with the trial protocol.

The three treatments differed in the total number of sessions they offered over the study period. The number of sessions offered for each treatment were as follows.

• BPI: up to 12 sessions, consisting of up to eight individual and four family/parent/guardian sessions, to be delivered over 20 weeks.

• STPP: up to 28 individual sessions plus up to seven parent/guardian sessions to be delivered over 30 weeks.

• CBT: up to 20 individual sessions plus up to four family/parent/guardian sessions to be delivered over 30 weeks.

The treatments are described below.

Prescribing of fluoxetine during the trial

For all three arms, fluoxetine or another SSRI could be added if clinicians judged that combination therapy may accelerate the time to remission, following NICE guidelines for a major depression episode in adolescents. 13 A test dose of 10 mg was given for 48 hours, followed by 20 mg as a single dose. If there was no improvement within 2–4 weeks then the dose could be adjusted upwards to a maximum of 60 mg.

Design

Two independent raters, blind to treatment allocation, rated each treatment session from the three treatment modalities using the Comparative Psychotherapy Process Scale (CPPS), which is a widely used measure of therapeutic techniques in psychodynamic therapies and CBTs. 82 These ratings were used to assess treatment fidelity for the CBT and STPP arms of the study and treatment differentiation between all three arms of the study. In addition, sessions from the BPI arm of the study were each rated by two raters, using a newly devised BPI fidelity measure in order to assess treatment fidelity to the BPI manual. Double ratings were used to check the reliability of each measure and improve the precision of the estimate for each tape.

Sample size

All therapists and young people in IMPACT agreed to their sessions being tape recorded for the purposes of the fidelity and differentiation analysis. Recorded sessions were categorised as either ‘early’ or ‘medium/late’ in therapy. A random sample of 232 tapes (76 CBT tapes, 81 STPP tapes and 75 BPI tapes) were selected and stratified by modality and timing (‘early’, i.e. the first third of therapy, or ‘medium/late’, i.e. the middle or last third of therapy), and were then rated on the measure of comparative (psychodynamic and cognitive–behavioural) techniques. The slight difference in the number of sessions rated by arms arose owing to the number of tapes available by treatment arm and site. As the comparative measure did not include the active features of BPI, the 75 BPI tapes were additionally rated on a treatment-specific measure.

Instruments

Ruminative response style

Rumination is the compulsively focused attention on the symptoms of one’s distress and on its possible causes and consequences as opposed to its solutions. 85 Rumination is similar to worry with the exception that rumination focuses on bad feelings and experiences from the past, whereas worry is concerned with potential bad events in the future. 86 Both rumination and worry are associated with clinical anxiety and depression. 86

Rumination has been widely studied as a cognitive vulnerability factor for depression but its measures have not been unified. 86 In the Response Styles Theory proposed by Nolen-Hoeksema,87 rumination is defined as ‘compulsively focused attention on the symptoms of one’s distress, and on its possible causes and consequences, as opposed to its solutions’. Because the Response Styles Theory has been empirically supported, this conceptual model of rumination is the most widely used.

Extensive research on the effects of rumination, or the tendency to self-reflect, shows that the negative form of rumination interferes with people’s ability to focus on problem-solving and results in dwelling on negative thoughts about past failures. 88 Evidence further suggests that the negative implications of rumination are due to cognitive biases, such as memory and attentional biases, which predispose ruminators to selectively devote attention to negative stimuli. 89 Such negative biases results in critical self-devaluing thinking and can be found in dysphoric adolescents with no history of depression but with a childhood temperamental style characterised by being easily distressed and fearful but likely to return to calm mood relatively rapidly. 90 Depressed adolescents who have high rumination scores are more likely to show persistent depression and demonstrate impairments in autobiographical memory retrieval. 91–93 Inducing ruminations in adolescents also results in increased depressive symptoms as there is a bias to ruminate on prior negative life events. 43,91,92

In this trial, self-reported rumination scores were measured by the ruminative responses styles questionnaire developed by Susan Nolen-Hoeksema and colleagues and validated independently. 94 The scale was completed prior to randomisation and the baseline raw sum score was planned (prior to analysis) to be tested as a potential moderator of treatment effects.

Hypothesis

Elevated RRS scores at baseline will be associated with the following:

1. lower treatment response in all arms and higher MFQ scores over the follow-up period of 36, 52 and 86 weeks

2. better treatment response in the CBT than the BPI and STPP arms.

Depressive experiences style

Both theoretical assumptions and empirical findings suggest that adult patients with clinical depression may be characterised in terms of immaturity of cognitive styles39,40,95,96 which manifest as excessive preoccupation with relatedness (principally focused on disappointment with relationships) and self-definition or identity (principally focused on self-criticism). Research has provided empirical evidence for the assumption that individuals with depression may be predominantly troubled by one of the following issues.

1. High concerns about the quality of interpersonal relatedness with feelings of emptiness and loneliness, and intense fears of being abandoned and left unprotected.

2. Possessing an extremely self-critical attitude together with feelings of worthlessness, guilt, failure and self-blame.

Two psychometrically relatively robust factors have been shown to emerge across a number of studies. 97 One factor, which may be termed dependent/relatedness is elevated by disruptions to rewarding affiliative interpersonal relationships. This is expressed primarily in dysphoric feelings following experiences characterised by personal loss, abandonment and being alone. The other is termed self-critical/identity and is elevated due to perceived personal failure and worthlessness. This is expressed in dysphoric feelings emerging in individuals who have a bias towards perfectionism but are vulnerable to criticism both from others and themselves.

Clinical evidence has accumulated on the difference in the responsiveness to different kinds of psychological treatment and their capacity to achieve therapeutic gain. Clinical research with depressed adult patients has indicated that elevated levels of the self-criticism factor at baseline are associated with poorer therapeutic outcome at termination of therapy and at the 3-month follow-up in brief treatments for depression. 96,98 In contrast, the dependent factor interfered with therapeutic progress primarily in the second half of the treatment process (in the last 8 weeks) by disrupting patients’ interpersonal relationships both within and external to the treatment process. 40,95

In this study, the short version of the DES-A (self-report) was used to measure relatedness and identity. 58 The scale was completed prior to randomisation and baseline subscale scores for relatedness and differentiation were tested as potential moderators of treatment effects.

We tested two hypotheses for the Depressive Experiences Questionnaire (DEQ) at baseline.

1. Elevated relatedness/dependent scores will be associated with a relatively better response in the STPP group than the BPI or CBT groups.

2. Elevated self-critical/identity scores will be associated with a relatively better response in the CBT group than the BPI or STPP groups.

The data analytic moderator strategy is described in Chapter 8 and the results are presented in Chapter 9.

Data analytic strategy

Statistical analysis of the three randomised treatment groups was based on the intention-to-treat principle, subject to the availability of data. Statistical analyses were carried out using Stata^® Release 13 (StataCorp LP, College Station, TX, USA).

Data cleaning of outcome and baseline data was conducted without the treatment group allocations in view. Many of the outcome measures were patient-completed or interviewer-rated psychometric instruments. When there were missing item-level data, these were imputed by replacing the missing item with the mean of the other available items for that occasion, provided that at least 50% of items had been completed (pro-rating).

To identify data errors prior to revealing treatment allocation, the trial research team reviewed the summary statistics of outcome data.

Analysis of the primary outcome measure and continuous secondary outcome measures

Characteristics of the study sample are reported using standard frequency measures and summary statistics. Preliminary inferential analysis investigated the pattern of missing outcome data comparing baseline characteristics of subjects with and without follow-up data using a logistic regression model.

Although time of assessment was scheduled at 6, 12, 36, 52 and 86 weeks, there was substantial variation in the timing of assessment compared with randomisation, leading to some overlap between assessment intervals of consecutive assessments. To prevent bias attributable to assessments being delayed, we have used time since randomisation as a continuous variable in a longitudinal mixed model rather than the notional assessment point. Random effects were included for between-subject variation in the intercept and time gradient of the subject. As there may be variations in patient outcomes, a between-therapists random-effect term was added to the models. When a participant’s therapist was not known or they received no therapy, the participant’s identifier code was used instead for the therapist code. Fixed covariates were included to model systematic differences due to treatment, the time with treatment interaction and prespecified participant characteristics at baseline (Table 1).

Trial hypotheses (see Chapter 9, Assessment of clinical outcome) were related to the immediate post-treatment follow-up period and the longer-term follow-up. The model was therefore fitted to the post-treatment data (≥ 36 weeks) and the marginal effect of treatment was estimated at 52 and 86 weeks post randomisation. At each time point, two comparisons were made: STPP versus CBT, and CBT and STPP versus BPI. The sample size calculation used a significance level of 2.5% to allow for this multiplicity. A Bonferroni correction has not been applied to the p-values, but readers should use a 2.5% significance level to maintain the familywise 5% level at a particular point of assessment. As well as the marginal effects, the treatment effect and time with treatment interaction are also reported with accompanying inference based on a likelihood ratio test. For all models, time was centred by subtracting the overall (grand) mean of assessment times based on the available data for the particular analysis being undertaken. This makes the intercept interpretable when there is a treatment by time interaction.

When baseline scale covariate data were not obtained, simple imputation, which is based on multiple regression, was used as suggested by White and Thompson. 102 The following covariates were used: region, comorbid behaviour disorder (conduct disorder and/or oppositional defiant disorder), all anxiety disorders combined, SSRI prescription before trial entry (if missing assumed not to be prescribed), age at randomisation, sex and baseline severity MFQ score.

The proportion of the total variance caused by the therapist, which can be called the ITCC, varies with time owing to the random gradient term in the model. For comparative purposes, the ITCC was calculated as:

where σT2 is the between-therapist variance, σP2 is the patient-level random intercept variance and σE2 is the residual error variance. This estimates the ITCC at the grand mean centred time point.

A secondary analysis estimated the treatment effect over the treatment period based on data gathered before 36 weeks. This model did not include a time with treatment interaction as there was only the notional week 6 and 12 assessments to use as outcome data.

Analysis of the binary secondary outcome measures

The analysis of binary secondary end points, which included the Antisocial Behaviour Questionnaire (ABQ), MDD and MFQ sum scores of ≥ 26 were analysed using a longitudinal generalised estimating equation (GEE) model with robust standard errors. This model was fitted to post-treatment data (≥ 36 weeks) and the marginal differences in proportions were estimated for STPP versus CBT and for CBT plus STPP versus BPI at weeks 36, 52 and 86.

Planned subgroup analyses

This trial included an investigation of potential moderator effects on treatment response for the primary outcome before and after 36 weeks. The effect of moderators was tested by adding a moderator with treatment interaction into the main effects models for treatment. The moderator measures and their associated testable hypotheses are described in Chapter 7.

Economic evaluation method

Calculation of quality-adjusted life-years

The QALYs were calculated using the area under the curve approach after the health states from the EQ-5D (see Chapter 5 for details of method and evaluation) were converted into utility scores using responses from a representative sample of adults in the UK. 64 It was assumed that changes in utility score over time followed a linear path. 65 QALYs in the second year were discounted at a rate of 3.5%, as recommended by NICE,66 and all analyses were adjusted for baseline utility scores to take into consideration the impact any baseline differences will have on the area under the curve. 106

Data analysis

For base-case calculations, complete case analysis (excluding subjects with missing data) was used, with the impact of missing data explored in sensitivity analyses. All analyses were carried out on an intention to treat basis using Stata 11.1.

Characteristics of young people entering the trial

Between 29 June 2010 and 17 January 2013, a total of 557 participants had baseline assessments. Of these, 87 were excluded from the study (see Figure 3 for reasons for exclusion). The remaining 470 participants were randomised, of which five later withdrew consent (three BPI, one CBT and one STPP). Among the remaining 465 participants, 155, 154 and 156 were randomised to BPI, CBT and STPP, respectively. The East Anglia regional centre recruited the largest number of participants (40%, n = 185) followed by the north-west England centre (33%, n = 153) and the North London centre recruited the smallest number (27%, n = 127). Recruitment was from five CAMHS clinics each in East Anglia and North London, and from six clinics in north-west England.

Table 3 summarises the demographic and clinical characteristics of the three randomised groups at entry into the trial. The mean age of the sample was 15.6 years (SD 1.4 years). A total of 348 (75%) participants were female and 85% (382/450) were white. Based on those with SSRI prescription information, 20% were prescribed a SSRI prior to entry into the trial. The baseline characteristics of each treatment group are also presented in Table 3 and there were no marked differences between treatment groups. Note that baseline values of outcome measures are tabulated with the summaries of follow-up data in Table 12.

Table 4 gives the prevalence of concurrent depressive symptoms from the K-SADS-PL. The most prevalent symptom was sleep disturbance (92%), followed by depressed mood (84%). The mean number of symptoms was 8.4 for the BPI group; 8.7 for CBT and 8.3 for STPP (see Table 3). Recent suicide attempts refer to current major depression episode. Lifetime suicide attempts refer to all lifetime except current episode.

Table 5 gives a detailed breakdown of comorbid psychiatric diagnoses recorded in the baseline K-SADS-PL by treatment group. A total of 225 (48%) participants were concurrently comorbid for at least one other psychiatric disorder. Of these, 134 (29%) participants had one comorbidity and 60 (13%) participants had two comorbidities. The maximum number of comorbidities was five in the BPI group and four in the other two groups. Overall, the most frequent comorbid diagnoses were generalised anxiety disorder and social phobia. There were no marked differences between the three treatment groups in these characteristics.

Non-suicidal self-injury

Recent non-suicidal self-injury during the current depression episode was reported by 85 (18.3%) patients and lifetime non-suicidal self-injury by 246 (52.9%) patients. The frequency of recent non-suicidal self-injury in the treatment groups was as follows: BPI (26 patients, 16.8%), CBT (25 patients, 16.2%) and STPP (34 patients, 21.8%). The frequency of lifetime non-suicidal self-injury in the treatment groups was as follows: BPI (87 patients, 56.1%), CBT (75 patients, 48.7%) and STPP (84 patients, 53.9%).

Trial treatments and medication

Uptake and duration of trial therapies

The number of patients recorded as having started trial therapy was 138 (89%) for BPI, 133 (86%) for CBT and 133 (85%) for STPP. Owing to differences in organisation and type of service, the time from randomisation to start of therapy could vary between treatments and regional centres. Figure 4 displays the time from randomisation until the start of therapy for the three trial interventions as a Kaplan–Meier plot. The longest time until the start of therapy was 36 weeks, which was for CBT. Forty-seven young people did not start therapy. There were no age differences between this subgroup and the remaining participants (no-therapy group: median age 15.58 years, range 11.30–17.90 years vs. therapy group: median age 15.86 years, range 12.74–17.99 years; Kruskal–Wallis rank test = 0.224). This no-therapy subgroup are censored at the longest recorded start time. When a Cox proportional hazards model was fitted to the time until start of therapy, with covariates including trial therapy and region, there was evidence of an interaction between region and trial therapy. Table 6 gives the median time to start of therapy by treatment and region, from which it is apparent that time until the start of BPI was rather shorter in North London than in other sites, and that time until the start of CBT was rather longer in East Anglia than in other sites. Table 6 also displays the median and mean number of sessions attended by participants in each treatment arm.

FIGURE 4.

Time from randomisation to start of trial therapy by group.

TABLE 6 - Estimated median time (95% CI) in weeks to start of therapy by trial therapy group and region

CI, confidence interval.

Region	Treatment group, median (95% CI)
	BPI	CBT	STPP
East Anglia	4.3 (3.3 to 5.6)	7.3 (5.1 to 10.1)	4.7 (3.9 to 5.7)
North London	2.9 (2.0 to 3.7)	4.0 (3.0 to 4.7)	3.9 (2.7 to 4.4)
North-west England	4.0 (3.1 to 4.6)	4.0 (2.9 to 4.9)	4.4 (3.1 to 6.1)

Each of the three trial therapies recommended a number of treatment sessions (see Chapter 5). The number specified for each of the three trial interventions was 12 for BPI, 20 for CBT and 28 for STPP. Table 7 gives the number of clinical sessions attended by young people. Of those patients randomised to BPI, 17% (24/138) underwent more than the suggested 12 sessions. Of those randomised to CBT, 3% (5/133) underwent more than the recommended 20 sessions, and for those randomised to STPP, 2% (3/133) underwent more than the recommended 28 sessions. Table 7 also gives summary statistics for numbers of sessions received. Among patients who received therapy, those randomised to BPI had fewer sessions than those randomised to CBT or STPP (Kruskal–Wallis p < 0.001) and the median number of treatment sessions attended was six for BPI, eight for CBT and 11 for STPP. Fewer than half of all patients who received their randomised treatment attended more than half of the recommended number of sessions for that treatment modality.

TABLE 7 - Number (%) of clinical sessions attended by treatment allocation (the recommended number of treatment sessions was 12 for BPI, 20 for CBT, and 28 for STPP)

IQR, interquartile range.

Calculated for young people receiving one or more session.

Number of sessions	Treatment group
	BPI			CBT			STPP
	Frequency	%	Cumulative %	Frequency	%	Cumulative %	Frequency	%	Cumulative %
Missing	6	3.9	–	6	3.9	–	2	1.3	–
0	11	7.1	100.0	15	9.7	100.0	21	13.5	100.0
1	12	7.7	92.6	11	7.1	89.9	8	5.1	86.4
2	13	8.4	84.6	8	5.2	82.4	9	5.8	81.2
3	9	5.8	75.8	4	2.6	77.0	4	2.6	75.3
4	15	9.7	69.8	6	3.9	74.3	6	3.8	72.7
5	9	5.8	59.7	10	6.5	70.3	8	5.1	68.8
6	12	7.7	53.7	11	7.1	63.5	5	3.2	63.6
7	10	6.5	45.6	6	3.9	56.1	4	2.6	60.4
8	4	2.6	38.9	8	5.2	52.0	13	8.3	57.8
9	8	5.2	36.2	5	3.2	46.6	1	0.6	49.4
10	6	3.9	30.9	6	3.9	43.2	3	1.9	48.7
11	7	4.5	26.8	7	4.5	39.2	6	3.8	46.8
12	9	5.8	22.1	5	3.2	34.5	2	1.3	42.9
13	3	1.9	16.1	7	4.5	31.1	6	3.8	41.6
14	5	3.2	14.1	8	5.2	26.4	2	1.3	37.7
15	3	1.9	10.7	2	1.3	20.9	3	1.9	36.4
16	2	1.3	8.7	4	2.6	19.6	2	1.3	34.4
17	1	0.6	7.4	7	4.5	16.9	1	0.6	33.1
18	4	2.6	6.7	2	1.3	12.2	–	–	32.5
19	1	0.6	4.0	3	1.9	10.8	5	3.2	32.5
20	1	0.6	3.4	8	5.2	8.8	3	1.9	29.2
21	1	0.6	2.7	3	1.9	3.4	4	2.6	27.3
22	–	–	2.0	1	0.6	1.4	4	2.6	24.7
23	1	0.6	2.0	–	–	0.7	5	3.2	22.1
24	–	–	1.3	1	0.6	0.7	5	3.2	18.8
25	–	–	1.3	–	–	0.0	7	4.5	15.6
26	–	–	1.3	–	–	0.0	4	2.6	11.0
27	--	–	1.3	–	–	0.0	4	2.6	8.4
28	–	–	1.3	–	–	0.0	6	3.8	5.8
29	–	–	1.3	–	–	0.0	1	0.6	1.9
33	1	0.6	1.3	–	–	0.0	–	–	1.3
39	–	–	0.7	–	–	0.0	1	0.6	1.3
42	–	–	0.7	–	–	0.0	1	0.6	0.6
43	1	0.6	0.7	–	–	0.0	–	–	0.0
Total	155	100		154	100		156	100.0
Median (IQR)a	6	4,11		9	5,14		11	5,23

A clinical estimate of adherence to therapy for each therapeutic modality was made by the lead clinical specialists for each trial treatment. As there is no prior scientific evidence, these operational definitions of minimum sufficient attendance were ‘best practice’ based and, therefore, cannot index a formal required therapeutic dose. They are given here for descriptive purposes only but may be hypothesis-forming for future reference.

Clinical adherence that might result in some therapeutic gain was defined by the lead therapeutic specialists in consultation with therapists in each treatment arm. The investigators agreed to set two sessions for BPI and five sessions for CBT and STPP as the minimum number of sessions likely to result in therapeutic gain. The proportion of patients in each treatment group attending what is estimated as inadequate or adequate number of sessions for putative clinical adherence is shown in Table 6. Conditional on attending at least one session, 17% (24/138) of patients exceeded the proscribed maximum number of sessions on BPI, compared with 4% (5/133) for CBT and 2% (3/133) for STPP. The number and per cent of sessions attended by treatment allocation is shown in Table 7.

Figure 5 displays the duration of trial therapy for those young people with a recorded start and completion date by treatment group. It was assumed that the duration of therapy was 1 day for young people attending just one therapy session. Summary statistics for the same data are given in Table 8. Average duration of therapy between treatment groups was quite similar and not significantly different (Kruskal–Wallis p = 0.238). However, the median duration of therapy was greater for STPP (30.1 weeks) than for either BPI (22.1 weeks) or CBT (23.1 weeks), which is illustrated in Figure 5. There is no noticeably greater variation in the length of therapy for BPI (SD 21.5 weeks) than for CBT (SD 17.7 weeks) and STPP (SD 16.8 weeks), which can be explained by rather more patients exceeding the suggested number of sessions for BPI (see Table 8).

FIGURE 5.

Duration of trial therapy by treatment group.

TABLE 8 - Summary statistics for duration of therapy (weeks from first to last session)

Note

The minimum duration was 1 day.

Treatment group	Median	Maximum	Mean	SD	n
BPI	22.1	111.9	27.5	21.5	130
CBT	23.1	99.6	24.9	17.7	130
STPP	30.1	97.0	27.9	16.8	131

All therapists delivering a trial therapy were given a trial therapist identifier. The therapist identifier was missing for 18 (12%) BPI trial treatments, 13 (9%) CBT trial treatments and 2 (1%) STPP trial treatments. A total of 63 therapists delivered BPI, 44 delivered CBT and 38 STPP. For all three modalities, the young person received their trial therapy from a single trial therapist. Figure 6 gives the distribution of the number of young people treated by each therapist for each treatment arm. The number of trial participants seen by a particular therapist ranged from 1 to 15. Forty BPI therapists treated only one young person in the trial, whereas the corresponding figures for CBT and STPP were 19 and 18, respectively. This difference in number of therapists per treatment group is due, in part, to the rather larger number of available BPI compared with CBT or STPP therapists within the 15 NHS CAMHS clinics.

FIGURE 6.

Frequency distribution of number of trial participants for each therapy modality seen by a trial therapist. (a) BPI; (b) CBT; and (c) STPP.

Fidelity and differentiation of trial therapies

Treatment differentiation

Table 9 gives the mean ratings for each of the three treatment arms on the CPPS subscales (psychodynamic interpersonal and cognitive–behavioural), showing the proportion of techniques associated with the other treatment arm that was used by BPI, CBT and STPP.

Figure 7 gives a box plot for the treatment differentiation score for each domain of the CPPS for the three treatments, and corresponding summary statistics are given in Table 12. This shows that all three treatment arms were significantly differentiated, based on blind double ratings of the CPPS. As predicted, the mean cognitive–behavioural subscale score was 1.91 higher for CBT than for STPP [95% confidence interval (CI) 1.73 to 2.09; p < 0.0001]. The mean psychodynamic interpersonal subscale score was 1.18 higher for STPP than for CBT (95% CI 1.01 to 1.3; p < 0.0001). BPI had a significantly lower cognitive–behavioural subscale mean than CBT (mean difference = –0.93, 95% CI –1.12 to – 0.75; p < 0.0001) and a significantly lower psychodynamic interpersonal subscale mean than STPP (mean difference = –1.30, 95% CI –1.48 to –1.11; p < 0.0001). In conclusion, ratings of the two CPPS subscales suggested that all three treatments were well differentiated.

FIGURE 7.

Treatment differentiation based on the CPPS. CB, cognitive–behavioural; PI, psychodynamic interpersonal.

Conclusion of the treatment fidelity and differentiation analysis

This analysis, based on the blind double ratings of 232 tapes, stratified by modality and timing, aimed to evaluate treatment fidelity and differentiation within the study.

Fidelity ratings were conducted with good (for CPPS) to moderate (for BPI-S) intraclass correlations scores. Overall, there was a relatively high level of protocol fidelity by the therapists in each arm of the study, with 81.3% of sessions meeting the threshold of fidelity for BPI, compared with 80% for STPP and 74% for CBT. Previous studies (e.g. ADAPT) have reported mean scores for therapist fidelity and/or competence, but have not used predefined cut-off scores for fidelity. However, direct comparisons between the figures for BPI and the two specialist psychological therapies should be interpreted with caution, given that BPI sessions were assessed using a different measure to the CBT and STPP treatment arms. Unlike the CBT and STPP sessions, BPI sessions were not rated blindly for treatment fidelity and had lower levels of inter-rater reliability; therefore, findings regarding levels of fidelity to BPI must be interpreted with some caution.

In addition to demonstrating good levels of protocol fidelity, treatment differentiation was established between all three treatment arms. In line with our hypotheses regarding ratings on the CPPS, CBT sessions were significantly lower than STPP on the psychodynamic interpersonal subscale, STPP sessions were significantly lower than CBT on the cognitive–behavioural subscale, and BPI sessions were significantly lower than both STPP on the psychodynamic interpersonal subscale and CBT on the cognitive–behavioural subscale. This demonstrates that STPP and CBT sessions were significantly differentiated from each other and BPI sessions were significantly differentiated from both CBT and STPP.

Antidepressant medication

Table 10 shows the number and percentage of participants who were prescribed any antidepressant during treatment and follow-up, by arm, based on data provided by the health economic schedule CA-SUS. Receipt of antidepressant medication during treatment is a potential mediator of the outcome of psychological treatment. Overall, a similar proportion of young people in each arm were prescribed medication during the trial treatment period (< 36 weeks), suggesting that receipt of medication was independent of random allocation. There is therefore no reason to believe that medication during the therapy period would bias estimates of the treatment effect immediately post therapy, unless drug therapy interacted with therapeutic modality.

During follow-up, the proportion of patients receiving any antidepressant medication increased from 27% in the treatment period to 34% for CBT, 35% for STPP and to 40% for BPI, although this difference between arms was not statistically significant (p = 0.584).

Assessment of clinical outcome

Table 11 gives the response rate for the primary outcome measure (MFQ) for each assessment. A total of 16 (10%) patients, 12 (8%) patients and 17 (11%) patients in the BPI, CBT and STPP groups, respectively, had no follow-up MFQ. The CBT group had the highest response rate of 80% at the 86-week assessment. To investigate baseline characteristics that might influence non-response, a logistic GEE model was fitted to an indicator variable of missing primary outcome data at each assessment (6–86 weeks) including the following covariates: randomisation, assessment number, age at randomisation, sex, region (East Anglia, North London, north-west England), baseline MFQ score, SSRI prescription before trial entry, behaviour disorder (a diagnosis of oppositional defiant disorder or conduct disorder), all anxiety disorders combined and a time by treatment by assessment interaction to investigate differential non-response. Two factors appeared to influence non-response: (1) regional centre with a higher response rate in the north-west England regional centre (p = 0.02) compared with East Anglia and (2) behaviour disorders at baseline with a lower response rate for subjects with conduct or oppositional defiant disorder (p = 0.004). Behavioural disorder was not in the original list of prespecified baseline covariates and so was added to all models to support the missing at random assumption for missing outcome data.

Table 12 gives summary statistics for the primary outcome measure MFQ and the secondary quantitative outcomes, which include Revised Children’s Manifest Anxiety Scale (RCMAS), LOI and ABQ, which are self-completed questionnaires. The HoNOSCA is interviewer rated.

On initial inspection of the data, it was clear that the ABQ was highly skewed with the SD larger than the mean at many time points and medians of zero at weeks 52 and 86 for each group. Because of the substantial skewness, the decision was made to dichotomize this scale and compare the proportion of none zero scores. Further analysis of this scale is therefore presented with other binary outcome measures and results are shown in Table 15.

In Figure 8, the mean scores with 95% CIs are plotted against time since randomisation for the quantitative outcome measures. For all scales, lower scores represent a better outcome, hence a greater reduction in one treatment than another represents a beneficial effect. For all four measures, the profile of CBT and STPP is below that for BPI post randomisation across time points.

FIGURE 8.

Mean outcome by treatment group (95% CI). (a) MFQ; (b) RCMAS; (c) LOI; and (d) HoNOSCA.

The hypotheses of the trial are regarding the treatment effect in the post-treatment period (≥ 36 weeks).

The study had four primary hypotheses.

When comparing CBT with STPP:

1. CBT will show non-inferiority effects compared with STPP at 52 weeks

2. STPP will show superiority effects compared with CBT at 86 weeks

and when comparing CBT and STPP with BPI:

1. the specialist intensive interventions (CBT/STPP) will show superiority effects compared with BPI at 52 weeks

2. the specialist intensive interventions (CBT/STPP) will show superiority effects compared with BPI at 86 weeks.

As described in Chapter 8, the mean differences were estimated at weeks 36, 52 and 86 post randomisation (1) between the two specialist psychological therapies (STPP vs. CBT = intensive therapies) and (2) between the two theory based treatments and BPI [(CBT or STPP) vs. BPI]. These mean differences were estimated using linear mixed-effects (LME) model fitted to post-treatment responses only (≥ 36 weeks from randomisation). As well as a main effect of treatment, a time by treatment interaction and other prespecified covariates were included in this model – details of which are given in the data analytic strategy (see Chapter 8).

Estimates of the time by treatment interaction from these models and the main effects of models are given in Appendix 1, Table 24. In total, 392 young people provided 1013 assessments of the primary outcome measure (MFQ), so the model includes 84% (392/465) of all participants contributing, on average, 2.6 assessments. To address the four hypotheses, the time point-specific treatment effect averaged over baseline covariates, sometimes called the marginal treatment effect, was estimated from the model including a time with treatment interaction. The mean differences between treatments at the three time points (36, 52 and 86 weeks post randomisation) are tabulated in Table 13. As higher scores reflect worse outcome for all scales, negative differences indicate a positive effect.

Presenting results chronologically, we first considered a secondary hypothesis concerning clinical outcome at 36 weeks, that is at the end of trial treatment phase. The difference between CBT and STPP was negative for the primary outcome (MFQ) and the secondary outcomes (RCMAS, LOI and HoNOSCA) (see Table 13). For MFQ, the differences showed a non-significant reduction of –0.18 (p = 0.929). For RCMAS, LOI and HoNOSCA the respected difference were also non-significant –0.86 (p = 0.621), –0.82 (p = 0.167) and –0.62 (p = 0.567), respectively. When the intensive treatment (CBT + STPP) was compared with BPI, the difference was –3.23 (95% CI –6.61 to 0.14; p = 0.061) for MFQ at 36 weeks. For both RCMAS and LOI, the reduction was statistically significant at a 2.5% level, suggesting benefit to the specialist psychological therapies at this time point on two of the secondary measures.

Moving to the hypothesis that CBT will show non-inferiority compared with STPP at 52 weeks (hypothesis 1), we need to consider the upper confidence limit of the difference in outcomes between CBT and STPP. For non-inferiority to be concluded, we need the upper limit to be smaller than a clinically important difference. For the purpose of the sample size calculation, this was assumed to be five points on the MFQ scale.

For all four measures in Table 13, the point estimate is negative and this represents a beneficial effect of CBT compared with STPP. For MFQ at 52 weeks, the treatment effect estimate is –0.307 (95% CI –3.774 to 3.161). As the upper limit (3.161) is less than a clinically important difference of 5 units, we can conclude that CBT is non-inferior to STPP at this time point. The corresponding upper limits for RCMAS, LOI and HoNOSCA were 2.354, 0.452 and 1.078, respectively, which would appear also to support the hypothesis of non-inferiority.

Consider now the hypothesis that STPP will show superiority effects compared with CBT at 86 weeks (hypothesis 2). Table 13 shows that the treatment effect for CBT compared with STPP is –0.578 (95% CI –4.100 to 2.950). With the point estimate representing a slight benefit for CBT compared with STPP, there is no evidence to support this hypothesis (p = 0.748). Similar conclusions can be drawn for RCMAS (p = 0.878), LOI (p = 0.906) and HoNOSCA (p = 0.394).

Hypotheses 3 and 4 consider superiority of the specialist psychological treatment compared with BPI at 52 and 86 weeks. At 52 weeks (hypothesis 3), the difference is –2.81 (95% CI –5.79 to 0.18; p = 0.065) for MFQ. Although this suggests that there is a benefit at this time point for the specialist psychological treatment, this difference is smaller than the 5-point difference hypothesised as clinically significant in the sample size calculation and is not statistically significant at a 2.5% level. Similar effects are observed for RCMAS (adjusted mean difference –2.81, 95% CI –5.43 to –0.21), LOI (adjusted mean difference –1.12, 95% CI –2.01 to –0.23) and HoNOSCA (adjusted mean difference –1.15, 95% CI –2.60 to 0.29), with the effect being close to the 2.5% level for RCMAS (p = 0.035) and statistically significant for LOI (p = 0.014), but less so for HoNOSCA (p = 0.118). At 86 weeks (hypothesis 4), the beneficial effect of the specialist psychological treatment compared with BPI is attenuated with the treatment effect for none of the four outcome measures being statistically significant at a 2.5% level.

The intraclass correlation coefficient (ICC) for therapists was estimated for the models by centring the time variable in the model. This gives the value of the ICC for the therapist at the mid-point of follow-up. The estimate of the ICC was found to be negligible (< 10^–7) for all models.

Outcome data gathered prior to 36 weeks after randomisation (assessments 1 and 2) were also summarised in Table 12, but these data are not relevant to the hypotheses of this study. Inferential analyses of these data are included in the data analysis supplement (see Appendix 1, Table 25). When STPP was compared with CBT, there were no significant treatment effects for the primary outcome MFQ (p = 0.383), RCMAS (p = 0.681) or HoNOSCA (p = 0.398). Young people receiving STPP had a worse outcome on the LOI scale during this period (adjusted mean difference = 1.19, 95% CI 0.28 to 2.12; p = 0.011). When the two specialist psychological therapies were compared with BPI, there was no evidence of a treatment effect for MFQ (p = 0.382), RCMAS (p = 0.632) or HoNOSCA (p = 0.646), and the outcome for BPI lay between that for CBT (p = 0.309) and STPP (p = 0.120) (see Table 13).

The trial had two binary outcome measures: (1) presence of a MDD as determined by the K-SADS inventory and (2) threshold of primary outcome (MFQ score of > 25) indexing potential clinical caseness. 47,112 As mentioned above, the ABQ was dichotomised for the purpose of analysis owing to having a highly skewed distribution. The frequency distributions of these three measures by treatment and time point are given in Table 14. With a reduced rate representing benefit for all three scales, there was a similar pattern to that observed in the continuous scales (see Table 14), with CBT and STPP showing benefit at assessment 3 (36-week assessment). At assessment 5 (86-week assessment), there is some suggestion of a difference between groups but the pattern was not consistent across measures.

A logistic GEE model was then fitted to all three outcomes for data gathered from week 36 onwards. This model was then used to estimate the difference in proportions at weeks 36, 52 and 86 averaged across covariates, that is the marginal difference in proportions between treatments. These differences with CIs are presented in Table 15.

At 36 weeks, the adjusted difference of percentages in MDD between CBT and STPP was –6.4% (95% CI –20.6% to 7.8%; p = 0.375). The corresponding differences for the MFQ and ABQ thresholds at this time point were 2.9% (95% CI –9.3% to 15.1%; p = 0.642) and –6.8% (95% CI –18.6% to 5.1%; p = 0.263). Considering the hypothesis of non-inferiority of CBT compared with STPP at 52 weeks (hypothesis 1), for MDD the adjusted difference of proportions was –1.8% (95% CI –12.0% to 8.4%), for MFQ threshold it was 1.2% (95% CI –8.8% to 11.3%) and for ABQ threshold it was –4.0% (95% CI –13.5% to 5.5%). Although the adjusted differences are small, the CIs are quite wide and so the evidence to support non-inferiority is weak, indicating that these comparisons are under-powered. At 86 weeks (hypothesis 2), there was no evidence for MDD (p = 0.261), MFQ threshold (p = 0.6777) or ABQ (p = 0.725) of STPP being superior to CBT.

There were no significant differences for the comparison of the specialist psychological therapies against BPI at 36 weeks (p = 0.465), 52 weeks (hypothesis 3) (p = 0.239) or 86 weeks (hypothesis 4) (p = 0.145) for presence of a MDD.

As this is a pragmatic study with no control group, we compared the proportion in remission at 12 weeks with the proportion of remitted patients from the published TADS randomised to a pill placebo control group (n = 111) and also assessed at 12 weeks. 11 This comparison showed that there were 145 out of 305 (48%) patients in clinical remission at 12 weeks in this study compared with 37 out of 111 (33%) pill placebo patients in TADS. We also note that the treatment trial of resistant depression in adolescents reported that 61% of patients met the criteria for diagnostic remission by 72 weeks, compared with 78% who did so by the notional 86-week follow-up in this study. 25 Finally, these result show that 15 (11%) patients relapsed by end of study from the 140 in remission at the 36-week assessment. There were no significant differences in the proportion of relapsed cases between treatment arms [5/43 (11.6%) BPI; 8/49 (16.3%) CBT; 2/48 (4.2%) STPP; p = 0.149]; however, caution is required with these results. As the study was not powered for treatment group comparisons, there are missing interview data at each time point and the control comparison could be achieved at 12 weeks only.

The percentage of subjects with a MFQ score of > 25, the adjusted difference in proportions between intensive treatments and BPI was –10.8% (95% CI –21.7% to 0.05%; p = 0.051) at 36 weeks and –10.2% (95% CI –19.3% to –1.1%; p = 0.028) at 52 weeks (hypothesis 3). This reduced to a difference of –8.5% (–17.7% to 0.66%; p = 0.069) at 86 weeks (hypothesis 4). As this cut-off score is derived from MFQ, one would expect these differences to be broadly consistent with those for this scale (see Table 14).

For the threshold (ABQ score of > 0) the specialist psychological treatments were significantly improved at 36 weeks (adjusted difference –12.8%, 95% CI –23.8% to –1.9%; p = 0.022). There was no significant difference for 52 weeks (hypothesis 3) with p = 0.102, or for 86 weeks (hypothesis 4) with p = 0.389.

In summary, across both continuous and binary outcome measure we conclude that outcome for CBT and STPP were broadly similar at 52 weeks (hypothesis 1) and 86 weeks (hypothesis 2). Comparing the combined specialist psychological treatments with BPI, there was some suggestion that outcome was better at 52 weeks (hypothesis 3), although this effect may not be clinically important. What treatment effects there were appeared to have largely dissipated by 86 weeks (hypothesis 4).

Moderation of treatment effects

Moderator effects on the primary outcome were investigated by adding an interaction between the moderator variable and treatment allocation to the primary analysis model. Table 16 gives the estimates of the treatment by moderator effect for each of the moderator hypotheses proposed (see Chapter 7 for details of measures and Chapter 8 for analytic strategy and hypotheses). A negative estimate in this table indicates that a higher score of the moderator lowered the MFQ for the treatment relative to the comparator, that is an increase in the beneficial treatment effect.

First, we hypothesised that young people with elevated dependency subscale sum scores on the DEQ would have greater reduction in MFQ than those with lower scores if they received STPP rather than BPI or CBT treatment. Before 36 weeks, the direction of the effect was consistent with our hypothesis, but this was not statistically significant (p = 0.168). After 36 weeks there was clearly no evidence of an effect (p = 0.918).

Second, we hypothesised that young people with elevated self-critical subscale sum scores on the DEQ would have a better response than those with lower scores if they received CBT rather than either BPI or STPP treatment. The direction of the effect was consistent with our hypothesis both before and after 36 weeks. Before 36 weeks this was marginally statistically significant (p = 0.053), with the effect being much smaller after 36 weeks (p = 0.384).

Finally, we hypothesised that higher total scale scores for rumination response style of thinking when depressed (RRS) would show a better treatment response for CBT than for BPI or STPP treatment. There was little evidence of an effect either before (p = 0.671) or after (p = 0.976) 36 weeks. In this case, the direction of the effect was consistent with our hypothesis only when comparing CBT with BPI after 36 weeks.

In summary, there is some indication that CBT might be more beneficial for young people with elevated DEQ self-critical scores, but this effect was applied to the treatment period only, rather than a long-term benefit as evaluated in the follow-up phase.

Adverse events

We undertook a brief examination of potential side effects of psychological treatment defined using responses to a set of somatic items selected from existing self-report scales in the study. This provides a brief proxy measure of the potential for psychological treatment to have negative effects and we focused on physical symptoms only. A description of the items and the findings are given in Appendix 1.

Data completeness

At 86 weeks, full CA-SUS service use data were available for 94 participants (61%) in the CBT group, 91 (58%) in the STPP group and 92 (59%) in the BPI group, which was 60% of the total number randomised. Baseline demographic and clinical characteristics of cases with complete and missing service use data were compared using standard t-tests and analysis of variance, as appropriate, including age, sex, ethnicity, region and MFQ. No significant differences were identified.

Outliers

The cost data were examined to consider the impact of highly influential observations, defined by Weichle et al. 113 as those whose exclusion result in major changes in the results. Two observations were identified as above the 99th percentile for total costs, but only one of these would have increased parameter estimates by a factor of 1.4. Therefore, this one observation was removed from the main analysis as recommended. 113

Resource use

All resources used over the 86-week follow-up period are summarised by group in Table 21.

Trial treatment

For the sample of participants with full service use information, the average number of treatment sessions attended by the young people was 7.97 in the BPI group, 9.73 in the CBT group and 13.85 in the STPP group. The numbers differ slightly from those reported in Chapter 9 because they are the results for the subsample of participants for whom we had full service use data. On average, the number of sessions attended was lower than the number of sessions planned (BPI 12 sessions, CBT 20 sessions, STPP 28 sessions).

Other health and social services

Overall, there was little difference between randomised groups in levels of service use over the 86-week follow-up (Table 17). Levels of mental health admissions were low (< 2%) across all randomised groups. There were slight variations in non-mental health admissions, with 13% of the STPP group being admitted compared with 8% in the BPI group and 5% in the CBT group. Overall, up to one-fifth of participants had a non-mental health admission (BPI 23.40%, CBT 12.63% and STPP 19.57%).

General practitioners were the most widely used service, accessed by 66%, 72% and 64% of participants in the BPI, CBT and STPP groups, respectively. Use of community mental health services, excluding the trial interventions, was highest in the BPI group (46% of BPI participants), compared with 38% and 29% of the CBT and STPP groups, respectively. Rates of social services contacts were also highest in the BPI group.

Antidepressant medication

Over the course of the study, patients were allowed to receive a SSRI in addition to psychological treatment if they met NICE guidelines13 for combined treatment to aid clinical remission by end of treatment. The proportion of participants prescribed antidepressant medication at any point over the 86-week follow-up was approximately 30% in each group.

Total cost

Outcomes

Cost-effectiveness analysis

Cognitive–behavioural therapy versus brief psychosocial intervention

For the CBT versus BPI comparison, CBT is less costly but slightly less effective in terms of QALYs than BPI. As a result, the replications produced in the scatterplot in Figure 9 are mainly in the south-west and south-east quadrants, reflecting lower costs in the CBT group (points below the x-axis) and the very small difference in outcomes between the two groups (points evenly spread across the y-axis). The CEAC in Figure 10 shows that for all levels of willingness to pay per QALY there is a higher probability that CBT is more cost-effective than BPI.

FIGURE 9.

Scatterplot of differences in costs vs. differences in QALYs for CBT vs. BPI.

FIGURE 10.

Cost-effectiveness acceptability curve showing the probability that CBT is cost-effective compared with BPI for different values at which a decision-maker might be willing to pay for improvements in QALYs.

Short-term psychoanalytic psychotherapy versus brief psychosocial intervention

For the STPP versus BPI comparison, costs were, on average, £403 higher in the STPP group than in the BPI group and QALYs were equal. The bootstrapped replications for STPP versus BPI are shown in Figure 11. The majority are in the north-east and north-west quadrants, reflecting the higher costs in the STPP group (points above the x-axis). The CEAC in Figure 12 shows that there are no willingness-to-pay values for which the probability of STPP being cost-effective compared with BPI is > 23%, within the £20,000–30,000 ceiling level of willingness to pay that is considered acceptable by NICE. 66

FIGURE 11.

Scatterplot of differences in costs vs. differences in QALYs for STPP vs. BPI.

FIGURE 12.

Cost-effectiveness acceptability curve showing the probability that STPP is cost-effective compared with BPI for different values at which a decision-maker might be willing to pay for improvements in QALYs.

Cognitive–behavioural therapy versus short-term psychoanalytic psychotherapy

Comparing the two specialist psychological treatments, CBT and STPP, total costs per participant over the 86-week follow-up were, on average, £703 lower in the CBT group than in the STPP group and outcomes were 0.02 QALYs worse. As a result, the replications in the scatterplot in Figure 13 are mostly in the south-west quadrant. The CEAC shown in Figure 14 suggests that the probability that CBT is cost-effective compared with STPP for all willingness-to-pay values is > 50%.

FIGURE 13.

Scatterplot of differences in costs vs. differences in QALYs for CBT vs. STPP.

FIGURE 14.

Cost-effectiveness acceptability curve showing the probability that CBT is cost-effective compared with STPP for different values at which a decision-maker might be willing to pay for improvements in QALYs.

Cognitive–behavioural therapy versus short-term psychoanalytic psychotherapy versus brief psychosocial intervention

The three interventions were compared head to head in a three-way comparison. The CEACs in Figure 15 show that for all values at which a decision-maker might be willing to pay for a QALY, CBT has the largest probability of being cost-effective.

FIGURE 15.

Cost-effectiveness acceptability curve showing the probability that BPI, CBT and STPP are cost-effective for different values at which a decision-maker might be willing to pay for improvements in QALYs.

Sensitivity analysis

The results of the sensitivity analyses are detailed in Tables 22 and 23. Multiple imputation did not alter the direction of the differences in cost, nor did reanalysis using cost per week rather than cost over the entire follow-up period. Including an estimate of the cost of sessions that were scheduled but that the young person did not attend altered the order between the three interventions.

TABLE 22 - Sensitivity analyses for costs (£) over the 86-week follow-up

Cost characteristic	Treatment group, mean (SD)
	BPI	CBT	STPP
Base-case analysis	2678.39 (2881.89)	2379.01 (3643.85)	3081.70 (3573.17)
Non-attendance at 50% cost	2907.30 (2939.08)	3050.05 (5891.69)	3364.14 (3563.08)
Multiple imputation	– (–)	– (–)	– (–)
Total cost per week	28.76 (31.63)	25.25 (38.35)	32.42 (35.84)

TABLE 23 - Between group differences for sensitivity analysis at the 86-week follow-up

All adjusted for region and baseline EQ-5D score, behavioural disorder and antidepressant use.

Comparisona	Sensitivity analysis	Coefficient	95% CI	p-value
CBT vs. BPI	Base case	–338.54	–1333.17 to 656.09	0.503
	Non-attendance at 50% cost	185.15	–392.71 to 1657.16	0.225
	Multiple imputation	–425.07	–1384.58 to 534.43	0.381
	Total cost per week	–3.95	–14.58 to 6.68	0.464
STPP vs. BPI	Base case	609.55	–406.73 to 1625.83	0.238
	Non-attendance at 50% cost	632.21	–392.71 to 1657.16	0.225
	Multiple imputation	448.95	–609.77 to 1507.66	0.399
	Total cost per week	6.12	–4.47 to 16.72	0.256
CBT vs. STPP	Base case	–709.23	–1836.04 to 417.58	0.216
	Non-attendance at 50% cost	–429.79	–1955.24 to 1095.65	0.579
	Multiple imputation	–891.47	–1951.81 to 168.86	0.098
	Total cost per week	–7.46	–19.10 to 4.17	0.207

For the sample with full economic data, the average number of sessions that were offered but were not attended was three in the BPI group, 14 in the CBT group and six in the STPP group. The inclusion of the cost of these sessions (at 50% of the cost of a full session) resulted in the average cost of CBT (£3050) becoming more expensive than the BPI average cost (£2939), with STPP remaining the most costly group (average cost £3364). Although there are no statistically significant differences in cost between the groups, this change in direction has an impact on the cost-effectiveness analyses for the comparison of CBT and BPI, with CBT being dominated by BPI (costs higher and outcomes very slightly lower). Figure 16 shows the scatterplot for this comparison; the majority of the replications are in the north-east and north-west quadrants, denoting higher costs in the CBT group (points above the x-axis). The very similar outcomes mean that the CEAC in Figure 17 suggests that the probability that CBT is cost-effective compared with BPI is < 50% for all values at which a decision-maker might be willing to pay for a QALY. Figure 18 shows a head-to-head comparison of all three groups in terms of cost-effectiveness and including a cost for sessions missed. It demonstrates that there is a higher probability of BPI being cost-effective than CBT and STPP for all values of willingness to pay.

FIGURE 16.

Sensitivity analysis: scatterplot of differences in costs vs. differences in QALYs for CBT vs. BPI with non-attendance at 50% session cost.

FIGURE 17.

Sensitivity analysis: CEAC showing the probability that CBT is cost-effective compared with BPI for different values at which a decision-maker might be willing to pay for improvements in QALYs with non-attendance at 50% session cost.

FIGURE 18.

Sensitivity analysis: CEAC showing the probability that CBT, STPP and BPI are cost-effective for different values at which a decision-maker might be willing to pay for improvements in QALYs with non-attendance at 50% session cost.

Timing of outcomes

The results showed a similar trend, seen across several symptom measures of depression, anxiety and obsessionality, for outcome scores in the specialist psychological treatment groups (CBT or STPP) to be somewhat lower than that of BPI at 52 weeks after randomisation, which was also seen at 36 weeks corresponding approximately to end of treatment. This is partly illustrated in Figure 7, showing the differences in individual scale scores for each treatment group. Although individual differences are not significantly different, there is a suggestion that the combined analysis (CBT or STPP) reveal that differences were close to statistical significance at the 2.5% level in Table 13. However, none of these mean differences was large and the clinical significance of somewhat lower sum scores seems uncertain. For example, the depression symptoms were significantly lower for the specialist treatment group with a mean difference of –3.28 at 36 weeks, –2.80 at 52 weeks and –1.98 at 86 weeks of follow-up assessment. Furthermore, although the mean difference is < 5 points on the MFQ, which we identified a priori as of clinical relevance, the confidence limits hint at patients showing greater individual differences in treatment response that may decline by 86 weeks.

Although multiple analyses across measures will tend to inflate type I errors, we note the consistency of the pattern of these results rather than the statistical or clinical significance of any one analysis. The average effect sizes across the four quantitative outcomes (see Table 13) were similar, ranging from 0.17 to 0.21 at 52 weeks. A slightly larger effect was observed at 36 weeks, ranging from 0.18 to 0.29. Equally, there was a consistent trend across these symptom measures for this difference to be much smaller and entirely non-significant by 86 weeks with effect sizes ranging from 0.04 to 0.17. The findings provide a pointer to the need for further research into timescales for symptomatic recovery from depression. The fall in symptoms deserves further investigation to establish whether or not it represents a meaningful difference in rate of recovery for some of those receiving specialist psychological treatments compared with BPI or, alternatively, that the observed difference in rate of recovery is not important. Reducing the period of time that an individual is suffering may be of value to them and their family. Further investigation may also reveal whether or not there are individual differences in the trajectory of symptomatic response to treatment and whether or not psychosocial adjustment correlates with or decouples from symptom change over time.

Economic analyses

On average, the observed cost of the trial treatments of those for which full data were available in the main analysis was lowest for CBT and highest for STPP. Although BPI was intended to be brief compared with both alternative treatments, in practice the average number of BPI treatment sessions attended was not substantially lower than in the CBT group (probably due to non-attendance of the planned longer-term therapies). Coupled with higher average therapist costs (as the majority of BPI was delivered by psychiatrists in specialist CAMHS training), BPI was estimated to cost somewhat more than CBT, although none of these differences was statistically significant. However, the inclusion of a cost for non-attendance reversed this finding, making CBT more expensive than BPI as a result of a larger difference between sessions offered and sessions attended. With a higher number of sessions attended per person, STPP remained the most expensive of the three treatments. Use of all other health and social services over the follow-up period was broadly similar, thus differences in total costs were primarily influenced by the cost of the trial treatments. In terms of cost-effectiveness, differences between groups were marginal and sensitive to the inclusion of the cost of sessions offered but not attended. Fewer therapy sessions attended could be taken either as an indicator of treatment effectiveness (fewer sessions needed before achieving a desired outcome) or as an indicator of lower treatment acceptability. Current results cannot in themselves clarify between these two alternatives or if both apply to different patients.

Brief psychosocial intervention

The protocol for high-quality active clinical care, referred to in this report as BPI, was developed for this study in order to ensure that the comparison condition was coherent and informed by all available research, and that it did not include active technical components of either CBT or STPP. As BPI was designed as a high-quality ‘active’ control condition, it was not evaluated against an intervention predicted to be less effective. This means that we cannot determine whether or not it was efficacious (as there is no passive control condition) or clinically effective in the absence of an appropriate clinical control group such as waiting list. However, as outcomes at 86 weeks in the BPI group were non-inferior to specialist therapies, BPI merits further consideration as a potential treatment. BPI was a multifaceted intervention with several elements (see Chapter 4 for details) that may have contributed to improvement in self-reported depression, including psychoeducation, support for increased activities, and attention to the young person’s family and school environment. It remains to be determined whether some or all of these elements may contribute to treatment effectiveness and also whether or not other components of BPI were important. For example, BPI may have been more tailored to variations in the young persons’ wishes or problems than the theory-based specialist treatments, which is consistent with evidence for advantages of personalised approaches to adolescent depression. 114

Treatment resistance and the maintenance of clinical high-risk status

A substantial proportion of patients (approximately 25%) continued to meet diagnostic criteria for unipolar major depression by 86 weeks. A further 15% reported depressive symptoms higher than the cut-off score (> 26) for potential cases. However, only 285 (60%) of the sample were available for full clinical assessment at this time point. This finding may suggest that, as there were no treatment group differences, there is a degree of treatment resistance or non-compliance in this cohort overall. Current therapeutic interventions used in this study appear to be potentially ineffective in a proportion of cases, an observation that resonates with prior trial findings on depressed adolescents. 11,12 This is a serious negative outcome that requires further investigation as providing potentially ineffective treatments to depressed adolescents is not good clinical practice. There is a marked lack of understanding regarding treatment failures and resistant depressions in this age range. There is a suggestion that more clinically severe presentations, the presence of suicidality and obsessive–compulsive disorder at presentation are associated with less treatment response by 28 weeks post randomisation. 115 Furthermore, the presence of non-suicidal self-injury at randomisation may be a predictor of increased suicidality and, therefore, a risk for treatment non-response. 116 These observations require replication and further investigation than they have received hitherto as identifying patients unlikely to respond to available therapies is an important clinical priority. The process of recovery may also be of interest. Numerous recent studies have identified rapid initial response to therapy (the first four or five sessions) as a critical indicator of ultimate benefit from therapy in adults. Further research with young people will be needed to investigate whether or not an early therapeutic response similarly indicates the absence of treatment resistance in a youth population.

Reducing depression symptom recurrence risk following treatment

Preventing clinical diagnostic relapse by maintaining low depressive symptoms is potentially of substantial clinical and cost value because even successfully treated adult patients may suffer, on average, between five and nine episodes over their life course at considerable personal and economic cost to both the individual and society. 2,117,118

The likelihood of relapse following successful treatment in depressed adolescents is also substantial, occurring in 50–75% of successfully treated patients. 29–31

These studies emphasise the potential value in lowering prospective diagnostic risk by reducing depressive symptoms in the medium term. Elevated symptom levels above the population norm predict the emergence of major depression in adolescents, correlate with persisting depressive disorders and predict relapse in adults with a history of depression. 93,119–122

Therefore, the current findings are encouraging as the lowered depressive symptoms by 86 weeks that is associated with these psychological treatments suggests progression towards prevention of diagnostic relapse for all three forms of interventions.

Strengths and limitations

This study had many strengths, including that participants were representative of depressed adolescents referred to a comprehensive health provision across diverse regions of the UK, that they all met research diagnostic criteria for DSM-IV MDD, that they were randomised remotely from the research team, and that reliable follow-up assessments were completed blind to treatment group assignment in > 75% of those randomised with no differences between the treatment groups. The most common reason for blindness break was the researcher being told by the respondent what treatment they were having. Breaks to blindness of treatment allocation were approximately 10–15% during the treatment phase but subsequently < 5% in the follow-up phase (36- to 86-week assessments). The sample size was greater than in any previous study and this is the first time, to our knowledge, that a trial of depressed adolescents has used follow-up as late as 86 weeks post randomisation as an end point of the trial. The participation rate was 84% for the primary outcome analysis, which is comparable to other studies in this field with similar duration of follow-up. Each of the three treatments was manualised, and fidelity and differentiation assessments demonstrated expected differences between them. Although there was a higher number of psychiatrists in specialist CAMHS training delivering the BPI treatment, overall clinicians who delivered each type of treatment were characteristic of those who deliver these treatments in routine clinical practice, which adds confidence to the findings.

There were limitations regarding the of outcome measures used. A decision had to be made between the research burden on vulnerable young people and the depth of information acquired. The relatively high level of consistency in trajectories between behavioural measures over the course of the study reassures us but acknowledges that functioning and adaptation was not as thoroughly assessed in the trial as symptomatology.

The use of a self-rated questionnaire has both advantages and disadvantages. We believe that, in the case of this study, the advantages offered were substantial. In particular, self-reporting reduced attrition rates over time as a significant number of participants did not want to attend follow-up face-to-face assessments beyond the treatment phase, but agreed to complete self-report questionnaires.

The overall trajectory of improvement seen with the self-report MFQ was similar to that seen with the interview-led HoNOSCA, suggesting multimodal measurement congruence between self-report and interviewer-led measures that focused on symptoms and their related functions, respectively. This was consistent with the findings published in ADAPT, for which self-reported depression and anxiety scores, together with research interview-led HoNOSCA and clinician-led clinical improvement ratings, were highly congruent in reporting responses to treatment. The multimodal measures used in this study (patient self-report, interviewer-led HoNOSCA and interviewer-led K-SADS together with self-report on depressive symptoms of anxiety, obsessive–compulsive and antisocial behaviour) provide a degree of acceptable validity and added value in measuring change over time in this treated cohort. However, this is limited in that there were no reliable and valid biomarkers of underlying pathophysiology available at the time of design or implementation of this study. Therefore, like all current treatment trial designs, we are limited to changes in the observed manifest symptoms following treatment. Advances in understanding aetiological and pathophysiological mechanisms for major depression are urgently required so that future treatment trials can determine their value in predicting treatment response and advance the need for more precision in determining what treatment works best for which patient.

The attendance rates through the trial are similar to previous findings in the UK ADAPT trial and not dissimilar to other large US trials, such as TORDIA (Treatment Of Resistant Depression In Adolescents). 25 The UK ADAPT and IMPACT trials were both pragmatic trials based in the NHS and the attendance rates reflect NHS service delivery and take-up rates by young people.

The inclusion of a manualised comparison condition, BPI, made it more likely that the quality of the treatment was similar across arms, enabled fidelity to be assessed and clarified the focus for supervision. It also ensured that the comparison condition did not include several of the significant features of STPP and CBT. The issue of overlap between therapies was evaluated with satisfactory findings from analysis of audio tapes, suggesting that the therapists were delivering treatment as per protocol and that these were substantially different between the three treatments. Nevertheless, it is acknowledged that some overlap in therapy processes may occur. In the absence of specific knowledge about the mechanisms of particular psychological treatment, it may be that there are commonalities that are shared and that exert sufficient clinical effects to diminish differences between specific treatment protocols in this study. In addition, BPI as an active manualised therapy does not constitute a low-level ‘treatment as usual’ concept. Further research on the mechanisms that lead to change when we expose depressed adolescents to psychological treatment is suggested for these results to disaggregate general from specific therapy effects.

Furthermore, we note the less than predicted attendance rates for each treatment arm. Understanding the precise relationships between duration of treatments, number of sessions attended and outcomes is an important topic for future studies. Finally, future studies should determine whether or not BPI may be of value in community child mental health settings, such as primary health care, schools and local authority services for adolescents, where depression is identified.

The economic results were limited by missing data, which were higher than for the primary clinical outcome measure (60%). However, multiple imputation of missing values did not change the results of the analysis, giving some confidence in the conclusions. The economic results, shown to be sensitive to the inclusion of the cost of participant non-attendance, are also limited by the reliance of this sensitivity analysis on accurate reporting by therapists. Data were calculated as the number of sessions offered minus the number of sessions attended. However, more detailed information (e.g. on whether sessions offered were cancelled or rearranged) was unavailable. These data are, therefore, a proxy for did not attend rates and so should be interpreted cautiously. The assessment of cost-effectiveness on the basis of the EQ-5D raises questions about the validity of the measure in an adolescent sample with depression. Although there is some evidence to support the use of the measure in this population,63 the evidence is relatively weak and further research into measures appropriate for young people with mental health problems is needed.

Summary of recommendations for future research

We may consider the following potential research questions on the basis of the findings reported above.

Conclusions

This RCT demonstrates no superiority for the specialist psychological therapies CBT and STPP compared with a high-quality active comparator, BPI, in maintaining the reduction of depressive symptoms after treatment through to 86 weeks post randomisation. All three treatments were associated with similar clinical effectiveness and cost-effectiveness and improved quality of life over the follow-up period.

Overall, these psychological treatments for unipolar major depression in adolescents differed markedly in theoretical and clinical approach but are associated with a similar degree of clinical improvement, cost-effectiveness and subsequent maintenance of lowered depressive symptoms. BPI offers an additional patient choice, alongside CBT and STPP, for depressed adolescents attending routine specialist CAMHS.

Contributions of authors

Ian M Goodyer (Professor of Child and Adolescent Psychiatry, University of Cambridge) designed the study, obtained the research funding and wrote papers arising from the study.

Shirley Reynolds (Professor of Evidence Based Psychological Therapies University of Reading) designed the CBT protocol, contributed to the study design including the fidelity and differentiation study and wrote papers arising from the study.

Barbara Barrett (Senior lecturer in Health Economics, King’s College London) undertook the health economics analysis and wrote the health economics chapter.

Sarah Byford (Professor of Health Economics, King’s College London) designed and undertook the health economics analysis, contributed to the grant application and wrote the health economics chapter.

Bernadka Dubicka (Consultant Psychiatrist and Lecturer, University of Manchester) contributed to measurement and study design and the BPI clinical protocol.

Jonathan Hill (Professor of Child and Adolescent Psychiatry, School of Psychology and Clinical Language Sciences, University of Reading) contributed to measurement, selection and study design while Professor of Child and Adolescent Psychiatry at the University of Manchester.

Fiona Holland (Research Associate in Statistics, University of Manchester) undertook data analysis.

Raphael Kelvin (Consultant Child and Adolescent Psychiatrist, Cambridge and Peterborough NHS Foundation Trust and Associate Lecturer, University of Cambridge) wrote the clinical manual and protocol for BPI, designed the fidelity and differentiation study and wrote papers arising from the study.

Nick Midgley (Child and Adolescent Psychotherapist, Anna Freud Centre and Lecturer at University College London) contributed to measurement and methods, contributed to writing the STPP protocol and manual, designed fidelity and differentiation study, and wrote papers arising from the study.

Chris Roberts (Professor of Biostatistics, University of Manchester) contributed to this grant application, designed the data analysis and wrote papers arising from the study.

Rob Senior (Senior Research Fellow, University College London, Medical Director and Honorary Consultant, Tavistock and Portman NHS Foundation Trust) contributed to study design and was the lead on recruitment in North London.

Mary Target (Professor of Psychoanalysis, University College London), contributed to the manual and clinical protocol of STPP, contributed to study design and wrote papers arising from the study.

Barry Widmer (Project Manager, University of Cambridge) designed the database and website, and managed the study.

Paul Wilkinson (University Lecturer, University of Cambridge) contributed to measures, contributed to writing of the BPI manual and wrote papers arising from the study.

Peter Fonagy (Freud Memorial Professor of Psychoanalysis, University College London) designed the study, obtained the research funding and wrote papers arising from the study.

We are grateful to the adolescent patients, former patients and their parents/guardians who actively contributed to the study design.

Publications

Goodyer IM, Tsancheva S, Byford S, Dubicka B, Hill J, Kelvin R, et al. Improving mood with psychoanalytic and cognitive therapies (IMPACT): a pragmatic effectiveness superiority trial to investigate whether specialised psychological treatment reduces the risk for relapse in adolescents with moderate to severe unipolar depression: study protocol for a randomised controlled trial. Trials 2011;12:175.

Goodyer IM, Reynolds S, Barrett B, Byford S, Dubicka B, Hill J, et al. Cognitive behavioural therapy and short-term psychoanalytical psychotherapy versus a brief psychosocial intervention in adolescents with unipolar major depressive disorder (IMPACT): a multicentre, pragmatic, observer-blind, randomised controlled superiority trial. Lancet Psychiatr 2017;4:109–19.

Data sharing statement

All trial data are held in a repository within the University of Cambridge. Access was made available following the publication of the main findings of the study in 2016. To obtain access please contact 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 HTA 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 HTA programme or the Department of Health.

Adverse events

The five items from the RCMAS subscale (breathing problems, sleep disturbances, drowsy/tiredness, nausea and sweating) and one from the MFQ subscale (restless/overactive) were used to generate a physical adversity score ranging from 0–6, for which each item with a response of ‘never’ was assigned a zero score, otherwise, this was assigned a value of one. Pro-rating was used if one or two of the six items were missing using all the available data for the MFQ or RCMAS subscales, respectively. The summary statistics are shown in Table 27.

Inspection of the data shows no observable differences between treatment groups over the course of the study. The decline in adverse event reporting over the five assessments is relatively consistent with a persistent lessening of positive responses from baseline recruitment to the end of follow-up.

Although the results show no treatment differences in somatic side effect profile, they are unlikely to provide a comprehensive estimate of psychological and social side effects or adverse effects that may accrue from psychological treatment of depressed adolescents. However, currently, there are no formally adopted methodologies for the measurement of such experiences, although the field is beginning to recognise the need for such methodologies. 123,124