Clinical effectiveness and cost-effectiveness of collaborative care for depression in UK primary care (CADET): a cluster randomised controlled trial

Article history

This issue of Health Technology Assessment contains a project originally commissioned by the MRC but managed by the Efficacy and Mechanism Evaluation Programme. The EME programme was created as part of the National Institute for Health Research (NIHR) and the Medical Research Council (MRC) coordinated strategy for clinical trials. The EME programme is funded by the MRC and NIHR, with contributions from the CSO in Scotland and NISCHR in Wales and the HSC R&D, Public Health Agency in Northern Ireland. It is managed by the NIHR Evaluation, Trials and Studies Coordinating Centre (NETSCC) based at the University of Southampton.

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 the material published in this report.

Declared competing interests of authors

David A Richards receives funding support from the National Institute for Health Research (NIHR) Collaborations for Leadership in Applied Health Research and Care. Simon Gilbody was a NIHR Health Technology Assessment Clinical Evaluation and Trials Board member during the conduct of this study (tenure 23 June 2008 to 30 September 2014). Glyn Lewis is currently a NIHR Efficacy and Mechanism Evaluation Board member. Peter Bower reports personal fees from the British Association for Counselling & Psychotherapy, outside the submitted work.

Copyright statement

© Queen’s Printer and Controller of HMSO 2016. This work was produced by Richards 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.

Depression

Depression results in substantial disability and is recognised as a major health problem; it is currently the second largest cause of global disability. 4 Around 350 million people are impacted by depression across the world and each year up to 5.8% of men and 9.5% of women will suffer from an episode of depression. 5 Depression has a very significant impact on physical health, occupational functioning and the social lives of sufferers. 6 Often anxiety is also present, causing further disability. 7

Depression is the acknowledged reason for two-thirds of all suicides. 8 The nature of depression is that it can frequently be chronic, with regular bouts of relapse and subsequent new episodes. After one depressive episode, around 50% of people will experience additional episodes. The risk of subsequent relapse is 70% after a second bout of depression and as much as 90% following three or more episodes. 9 Among other diagnostic criteria, depressive symptomatology incorporates depressed mood, loss of interest or pleasure in activities, insomnia or sleeping too much and fatigue or loss of energy. 10

Although effective pharmacological and psychological treatments for depression are available, people are often treated with a less than optimal programme. Internationally, there is often poor patient adherence to pharmacological treatment11 and further problems caused by organisational barriers between generalists and specialist mental health professionals. 12,13 There is often very limited support for primary care doctors when treating participants with both psychosocial interventions and pharmacological methods. Such support may be critical given that, in systems such as that in the UK and elsewhere, the general practitioner (GP) is the sole responsible medical clinician for 90–95% of patients. 14

Collaborative care

In a previous systematic review of 36 studies testing organisational interventions,15 it was concluded that guidelines, practitioner education and other simple interventions were not effective in studies attempting to improve the management of depression. However, there is better evidence for the role of organisational interventions in improving the management of a range of chronic conditions generally. Organisational strategies have been used in the management of depression, including ‘collaborative care’. This is a complex intervention developed in the USA that has been supported by previous reviews. Collaborative care has been identified as the most effective of the range of organisational approaches studied. 15–20

Collaborative care incorporates a multiprofessional approach to patient care; a structured management plan; scheduled patient follow-ups; and enhanced interprofessional communication. 21 In practice, this is achieved by the introduction of a care manager into primary care, responsible for delivering care to depressed patients under supervision from a professionally qualified mental health specialist and for liaising between primary care clinicians and specialists. Care management has been described as a health worker taking responsibility for proactively following up a patient, assessing patient adherence to psychological and pharmacological treatments, monitoring patient progress, taking action when treatment is unsuccessful and delivering psychological support. 18 Care managers work closely with the primary care provider (who retains overall clinical responsibility) and can receive regular supervision from a mental health specialist. 15,22 The specific disciplines vary by country context but can include counsellors, paraprofessionals or nurses as care managers, and psychiatrists, psychologists and mental health nurses acting in the specialist role.

Systematic reviews23,24 demonstrate that collaborative care improves depression outcomes, with some studies showing benefit for up to 5 years. Before developing the CollAborative DEpression Trial (CADET), our 2006 systematic review24 of 28 collaborative care studies showed collaborative care to be effective [standardised mean difference (SMD) –0.24, 95% confidence interval (CI) –0.17 to –0.32]. The I² estimates of inconsistency were 80% for antidepressant use and 54% for depressive outcomes. In metaregression analyses three intervention content variables predicted improvement in depressive symptoms, recruitment by systematic identification (p = 0.061), care managers having a specific mental health background (p = 0.004) and provision of regular supervision for care managers (p = 0.033), which reduced the overall heterogeneity (I²) from 54% to 48% for systematic identification, 43% for case manager background and 49% for supervision.

More recently (after the initiation of the CADET trial) we have undertaken a Cochrane review. 23 Our new analyses show greater improvement in depression outcomes for adults with depression treated with the collaborative care model compared with usual care in the short term (0–6 months) [SMD –0.34, 95% CI –0.41 to –0.27; relative risk (RR) 1.32, 95% CI 1.22 to 1.43], medium term (7–12 months) (SMD –0.28, 95% CI –0.41 to –0.15; RR 1.31, 95% CI 1.17 to 1.48) and long term (13–24 months) (SMD –0.35, 95% CI –0.46 to –0.24; RR 1.29, 95% CI 1.18 to 1.41). However, these significant benefits were not demonstrated into the very long term (≥ 25 months) (RR 1.12, 95% CI 0.98 to 1.27). In metaregression of this significantly larger study data set (n = 79) collaborative care that included psychological interventions predicted improvement in depression (beta-coefficient 20.11, 95% CI 20.01 to 20.20; p = 0.03). These new data include the results of our CADET trial along with another nine UK studies and a greatly expanded study data set. We include them here for completeness and refer to them further in the discussion.

In 2008, at the commencement of the CADET trial, collaborative care had generally been developed and tested in the USA within managed health-care settings. It is possible that the overall effectiveness of collaborative care programmes might vary when it is implemented and evaluated in non-US settings. In other areas of mental health care results from US-developed organisational interventions have not generalised outside the original health-care context. 25 For collaborative care, there was some supportive evidence from other contexts, including the developing world,26,27 but prior to the CADET trial there has been uncertainty around the standardised effect size in UK trials (SMD 0.24, 95% CI –0.060 to 0.547) and elsewhere. 24 These limited non-US data and the relatively small effect size in trials of patients with depression alone led the UK National Institute for Health and Care Excellence (NICE)28 to issue a research recommendation that ‘The efficacy of organisational interventions, such as chronic disease management programmes or other programmes of enhanced care for depression, should be tested in large-scale multicentre trials in the NHS’ (research recommendation 5.6.8.1, p. 103). This provided us with the rationale to undertake a fully powered UK evaluation of collaborative care.

Development of the CollAborative DEpression Trial

In published studies there is considerable between-study heterogeneity in terms of the duration and intensity of collaborative care and the training and background of care managers. Therefore, to investigate the clinical effectiveness and cost-effectiveness of collaborative care in the UK, we conducted a series of Medical Research Council (MRC)-funded preparatory studies. We wished to develop an intervention in anticipation of a fully powered randomised controlled trial. We carefully developed our collaborative care intervention to be applicable outside the USA, in health-care systems with a well-developed primary care sector. 29–31

In our Phase II testing of this intervention,30 we found preliminary evidence that collaborative care adapted to the UK was acceptable to patients and clinicians and may be effective outside the USA, but that a cluster randomised controlled trial was required to guard against potential contamination between trial arms. We amended the clinical protocol studied in our pilot trial to take account of acceptability data in our qualitative interviews and designed a cluster randomised controlled trial of sufficient power to detect clinically meaningful and achievable differences between collaborative care and usual care. We now report the results of this pragmatic cluster randomised controlled trial1 to determine whether or not collaborative care is more clinically effective and cost-effective than usual care in the management of patients with moderate to severe depression. This report is divided into chapters detailing the methods and results of our primary clinical effectiveness and cost-effectiveness questions followed by similar chapters for our process evaluation. We have undertaken an additional long-term follow-up of clinical outcomes and report this in a separate chapter. Finally, we conclude with a discussion chapter summarising our results and considering their implications for the management of depression.

Research question

Is collaborative care more clinically effective and cost-effective than usual care in the management of participants with moderate to severe depression in UK primary care?

Study design

The CADET trial was a multicentre, two-group, cluster randomised controlled trial with allocation of general practice clusters to two trial arms: collaborative care (experimental group) or usual care (GP management). We chose a cluster design given that our Phase II trial30 described in the previous chapter demonstrated that a participant-randomised trial of collaborative care could be vulnerable to contamination and open to type II error, underestimating the true effect size of the intervention through potential intervention ‘leakage’.

Patient and public involvement

We involved patient and public representatives at all stages of the project. A patient and public involvement (PPI) advisor (CM) was a trial applicant, investigator and full member of the Trial Management Group (TMG). He attended all meetings of the TMG and advised on patient-facing materials including ethics materials and participant therapeutic manuals and on the conduct of the trial including project management, questionnaire development, data collection and project dissemination. There were two PPI representatives on the Trial Steering Committee (TSC), one from a depression consumer advocacy group and another with lived experience of depression. Both provided important checks and balances as part of the independent TSC oversight of the trial.

In addition, the trial was initially co-ordinated from the Mood Disorders Centre at the University of Exeter and latterly from the University of Exeter’s Medical School. Both the Mood Disorders Centre and the Medical School operate within a culture of PPI, guided by published theories of participation, empowerment and engagement, through the Mood Disorders Centre’s 20-strong Lived Experience Group and the National Institute for Health Research (NIHR) CLAHRC (Collaboration for Leadership in Applied Health Research and Care) for the South West’s patient involvement group PenPIG (Peninsula Public Involvement Group).

Setting and participants

We recruited participants between June 2009 and January 2011 from the electronic case records of primary care general practices in three UK sites: Bristol, London and Greater Manchester.

Recruitment

Potential participants were identified by clinical studies officers (CSOs) or practice staff from July 2009 to January 2011. These workers searched the computerised records of participating practices over a 19-month period, looking for records of people with at least one identification code for depression recorded against their name by their GP in the previous 4 weeks. We searched for those codes most widely used by GPs to classify participants as depressed.

The lists of people generated by the searches were screened by GPs to remove the names of anyone whom GPs knew would not meet our inclusion criteria or who would be excluded at interview. Staff then sought permission from potentially eligible people for researchers to contact them. Potentially eligible people were sent an information sheet and reply slip in the post. Practice staff or CSOs followed up this letter by telephone after 1 week.

Those who gave research interview consent were contacted by a researcher trained in the specific interview procedures by study investigators and an interview was organised at the convenience of the participant. Interviews took place either in their home or at their GP practice no earlier than 48 hours after they had received the trial information letter. The first part of the research interview consisted of the researcher outlining the trial in detail and answering questions. Once the potential participants were fully briefed and willing to enter the trial, researchers asked them to fill in and sign the trial consent form, following which the diagnostic component of the baseline interview was undertaken.

If the diagnostic component of the baseline interview – the CIS-R33 – confirmed that a potentially eligible person met our depression diagnostic inclusion criteria, he or she was included as a participant in the trial, the research interview proceeded and a full baseline data set was taken.

Intervention and comparator groups

Outcomes

Sample size

The CADET trial was powered at 90% (alpha = 0.05) to detect an effect size of 0.4. This effect size is regarded as a clinically meaningful difference between interventions of this type. 46 The proposed effect size was also within the 95% CI of the effect that we had predicted following our analysis of our feasibility and pilot study. 30 This study had shown a potential effect size of 0.63 with a 95% CI from 0.18 to 1.07. However, an effect size of 0.4 is greater than that in the meta-analysis results of trials published at the time that we were commissioned to undertake the CADET trial (effect size 0.25, 95% CI 0.18 to 0.32). 24

In a two-arm randomised controlled trial in which participants are the unit of randomisation, we would have required 132 participants per group to detect a difference of 0.4 between groups. However, cluster randomisation produces a design effect and inflates the required sample size. In our trial, we planned for 12 participants in each primary care cluster and we used data from our pilot study30 to estimate the intracluster correlation (ICC) as 0.06. The design effect was calculated as 1.65 and we estimated our required sample size without any attrition to be 440. We estimated that the CADET trial might suffer from around 20% participant attrition at our primary end point. In this scenario, to be able to follow up 440 participants, we planned to randomly allocate 550 participants across the trial arms. To deliver this target we decided to recruit 48 practices with up to 14 participants per practice, given that our recruitment rate would not be exactly even between practices.

Statistical methods and analyses

Participant consent and ethical approval

We were granted ethical approval by the NHS Health Research Authority, National Research Ethics Service (NRES) Committee South West (NRES/07/H1208/60). We ensured that informed consent was gathered from participants before they undertook any engagement with the study, including data collection and treatment allocation and receipt. In detail, the process was as follows.

First, potential participants had to indicate their potential interest in the trial. They then consented to a researcher-led discussion. Everyone who reached this stage was sent the full participant information sheet by a member of the CADET research team and an appointment was also made. At the initial appointment the trial was explained in detail by the research interviewer, who also answered potential participant questions.

We informed all potential trial participants that being consented into the trial would not replace or adversely affect usual care delivered by their GP. All interviewees were told that they could avail themselves of other services or treatments and that they could withdraw from the trial without incurring any penalty to their health or treatment choices. Having considered these facts and agreed to trial participation, we asked potential trial participants to sign a formal consent form. All of our researchers undertaking this consent process were trained and supervised by the CADET investigator team.

Participant flow and retention

Allocation of practices

In total, 53 practices were randomised, two of which dropped out after allocation (Figure 1). These practices were removed from the minimisation schedule and their data did not influence later allocations. During recruitment, we found that the cut-off adopted for the IMD had been set far too high, with all practices so far recruited being below the cut-off. We changed this cut-off to one close to the median of practices so far recruited, retaining allocations so far. One practice was found to have been mistakenly recorded in the wrong geographical area; it was moved to the correct group, retaining its allocation. Of the remaining 51 practices, two did not recruit any participants.

FIGURE 1.

Trial CONSORT diagram. a, The number of patients in the collaborative care group who were excluded on interview because they were receiving treatment from secondary care or ‘another mental health provider’ (n = 5) includes one participant who was initially allocated in error and who was subsequently excluded.

Tables 2 and 3 show the geographical distribution of practices and minimisation variables (IMD, number of GPs and number of registered patients per practice), respectively, by intervention group. There was a wide range for all three practice characteristics.

TABLE 2 - Geographical distribution of practices by intervention group

Site	Collaborative care	Usual care	Total
Greater Manchester	7	9	16
Bristol	8	9	17
London	9	9	18
Total	24	27	51

TABLE 3 - Distribution of the minimisation variables in the two intervention groups

Variable	Group	Number of practices	Mean	SD	Minimum	Maximum
IMD	Collaborative care	24	9210	7416	317	27,365
	Usual care	27	8449	6012	265	19,536
	Total	51	8807	6651	265	27,365
Number of GPs (whole-time equivalents)	Collaborative care	24	3.8	2.0	1.0	10.0
	Usual care	27	4.0	1.9	1.0	7.8
	Total	51	3.9	1.9	1.0	10.0
Number of patients	Collaborative care	24	6615	3282	2200	15,000
	Usual care	27	7152	3781	1850	14,528
	Total	51	6899	3530	1850	15,000

Participant recruitment

The mean number of participants recruited for the remaining 49 practice clusters was 11.9 (SD 3.9, range 4 to 20). We recruited 581 participants in total and followed up 505 (87%) and 498 (86%) at 4 and 12 months, respectively. The CONSORT diagram in Figure 1 illustrates the flow of participants through the trial.

Baseline characteristics of participants

The mean age of participants was 44.8 years (SD 13.3) and 72% were women. Fewer than half (44%) of participants were in full- or part-time paid employment. More than half (56%) of the participants fulfilled ICD-10 criteria for a moderately severe depressive episode, with a further 30% meeting criteria for severe depression and 14% meeting criteria for mild depression and 73% of all participants having had depression in the past (Table 4). Almost all (98%) participants had a secondary diagnosis of an anxiety disorder, the most common being generalised anxiety disorder. Almost two-thirds of participants (64%) reported a long-standing physical illness (e.g. diabetes, asthma, heart disease). At baseline, 83% of participants had been prescribed antidepressant drugs by their primary care doctor.

The distribution of PHQ-9 scores at baseline was negatively skewed, with the majority of scores being in the higher part of the range. The mean PHQ-9 score overall was 17.8 (SD 5.1), with the usual care group having a slightly higher average score than the collaborative care group (18.1 vs. 17.4 respectively) (Figure 2 and see Table 2). All subjects were selected for the trial using a different measure of depression (CIS-R) from that used in the analysis to avoid problems of regression towards the mean. Low scores represent random variation because of measurement error.

FIGURE 2.

Distribution of PHQ-9 scores at baseline.

Delivery and receipt of the intervention

A total of 10 care managers provided collaborative care for 276 participants. The mean number of participants managed per care manager was 27.6 (SD 16.42, range 4 to 46).

Patients received a mean of 5.6 (SD 4.01, range 0 to 15) sessions with their care manager. Forty-two (15.2%) participants did not attend any sessions with their care manager, 213 (77.2%) had two or more contacts and 171 (62.0%) had four or more contacts. The mean total time in collaborative care was 3.03 hours (SD 2.18 hours) over a period of 12 weeks (SD 7.75 weeks). For those participants who attended at least one session, the mean duration of the sessions was 34.5 minutes (SD 8.2 minutes). Most participants in both collaborative care and usual care remained on antidepressant medication (74.8% vs. 73.8% at 4 months; 69.7% vs. 69.2% at 12 months).

The mean number of collaborative care sessions per participant in which medication was discussed was 3.2 (SD 3.43, range 0–13). The mean number of sessions incorporating behavioural activation was 5.4 (SD 3.4, range 0–13). The mean number of face-to-face contacts per participant was 1.17 (SD 0.92, range 0–6). The mean number of telephone contacts was 5.3 (SD 3.5, range 0–14).

We collected 220 supervision records reporting a mean supervision time of 35 minutes per week, with six participants discussed on average per session. Participants were discussed in an average of three supervisory sessions over the course of their intervention, at 6 minutes per participant per session. Care managers, therefore, received the intended level of supervision and number of sessions, with the number of participants discussed being dependent on the caseload of individual care managers.

Primary outcome: Patient Health Questionnaire-9 at 4 months

The primary and secondary outcomes at 4 and 12 months are presented in Table 5 and data on recovery, response and numbers needed to treat are presented in Table 6. With regard to the primary outcome (PHQ-9 at 4 months) we found a significant effect of collaborative care. The estimated mean depression score was 1.33 PHQ-9 points lower (95% CI –2.31 to –0.35; p = 0.009) for participants receiving collaborative care than for participants receiving usual care after adjustment for baseline depression. More participants receiving collaborative care than those receiving usual care met criteria for recovery [odds ratio (OR) 1.67, 95% CI 1.22 to 2.29; number needed to treat 8.4] and response (OR 1.77, 95% CI 1.22 to 2.58; number needed to treat 7.8).

Secondary outcomes

Missing data

In Table 8 we present the results after multiple imputation for the effect of collaborative care at 4 and 12 months on the main scales used, which also shows the results of the analyses using the available data, which were reported in the preceding sections. The imputed estimates are very similar to the available data estimates and so we can conclude that for all of these analyses the effects of collaborative care are little affected by missing data. The lower GAD-7 anxiety score at 12 months in the collaborative care group, which is not significant in the available data analysis, is just statistically significant in this simulation. However, the difference in the p-value (0.06 or 0.05) is very small.

Missing data were related to intervention group. Table 9 shows missing PHQ-9 data at 4 months and 12 months by intervention group. At 4 months, missing data were significantly more likely in the collaborative care group. At 12 months the difference between the groups was much smaller and not significant. As far as we can tell, this difference in missingness at 4 months does not produce a difference in the outcome variables between the two groups and does not explain the observed lower PHQ-9 score in the collaborative care group. This difference persists at 12 months, when the difference in missingness is much smaller.

Results of the economic analyses

Our estimated mean cost per participant for the delivery of the collaborative care intervention was £272.50. This cost estimate includes care manager costs at £232 and clinical supervision costs of £40.50. Our probabilistic analyses used to explore uncertainty around the main cost component, drawing from the distribution of contact time for care managers, showed that in 95% of simulations (cost estimates) the estimated cost of collaborative care was between £101 and £592 per participant (median £249 per participant).

Quality-adjusted life-years

In Table 13 we report data on health state values for the EQ-5D and SF-6D and the estimated QALY values over the 12-month follow-up period. When adjusted for baseline and for individual and cluster covariates we found a difference of 0.02 QALYs (95% CI –0.02 to 0.06 QALYs) over 12 months for the EQ-5D and 0.017 QALYs (95% CI 0.000 to 0.032 QALYs) for the SF-6D. Both measures show a QALY gain for collaborative care, although the EQ-5D difference is not statistically significant.

Cost-effectiveness analyses

In Table 14 we present estimates of cost per QALY, based on participants with data on costs and outcomes at follow-up. The base-case cost per QALY for collaborative care was £14,248, adopting a NHS and social care perspective, with uncertainty around this estimate illustrated in Figure 3 (cost-effectiveness plane) and Figure 4 (CEACs). The probability that collaborative care is cost-effective compared with treatment as usual is 0.58 at a willingness to pay of £20,000 per QALY and 0.65 at a willingness to pay of £30,000 per QALY.

FIGURE 3.

Cost-effectiveness plane (payer perspective).

FIGURE 4.

Cost-effectiveness acceptability curves. CEA, cost-effectiveness analysis; INB, incremental net benefit; SA, sensitivity analysis.

Objectives

The objectives of the process evaluation were to investigate:

• the potential moderators of differential participant response62,63 and the possible mechanisms of symptom change in collaborative care

• the process of implementation of collaborative care.

Methods

To investigate moderators and mechanisms of effect we recorded a number of baseline covariates together with intermediate process variables that were targets of the intervention, regressing these against depression severity (PHQ-9) at 4 and 12 months using multilevel multiple linear regression.

Measures

Analysis

Consistent with the mechanisms of change framework described by Kraemer and colleagues for randomised controlled trials, all of our analyses were exploratory, hypotheses generating activities. 62 Analyses were undertaken in Stata 12.1 following a predefined analysis plan.

To explore the role of moderators we analysed the direct effects of our six baseline covariates on depression severity (PHQ-9) at 4 and 12 months using multilevel multiple linear regression, with a preliminary step to assess ‘overall moderation’ (regression including all moderator variables), thereby controlling for type I error inflation as a result of performing a large number of statistical tests. 65 We used the xtmixed command in Stata, modelling surgery as a random effect, to account for nested structure of the data (patients nested within surgeries). Trial site location was modelled as a fixed effect rather than the top random-effects level to reduce model complexity and avoid non-convergence issues. Each of the six baseline covariates and their interactions with the intervention (through which we investigated moderation) were included in the overall model; all variables were appropriately centred. 66 We planned to proceed with individual moderator analyses only in the presence of ‘overall moderation’.

Potential mediating effects of collaborative care were investigated using structural equation modelling (SEM) in Stata. We analysed available data on the effect of medication adherence and behavioural activation at 4 months on PHQ-9 scores at 4 and 12 months and we explored the effect of medication adherence and behavioural activation at 12 months on the 12-month PHQ-9 scores. To explore and control for possible confounding,63,67 for mediators that were found to have a statistically significant effect on outcome, we analysed the effect of pre-randomisation variables on mediation by including all of the baseline covariates (hypothesised moderators) in the respective structural equation model. To investigate the effect of possible post-randomisation confounding variables on mediation, we analysed the direct effects of the collaborative care participants’ care manager and number of treatment sessions on the intervention group’s 4- and 12-month PHQ-9 scores using multilevel multiple linear regressions, with GP surgery modelled as a random effect and trial site as a fixed effect. We planned to proceed with mediation analyses (SEM) allowing for post-randomisation confounding variables only if there was evidence of a direct effect of care manager or number of treatment sessions on depression severity.

In all mediation and confounding analyses we analysed the effect of missing process data in sensitivity analysis, in which we used 1000 sample bootstraps. For the post-randomisation confounding investigation we performed an additional sensitivity analysis in which we controlled for all of the baseline covariates in the regression model and bootstrapped 1000 times.

Results

Introduction

Our original funding and protocol1 included only a maximum 12-month follow-up period. However, we were able to use some grant underspend to facilitate a no-cost extension from the MRC/NIHR to examine the long-term effects of collaborative care.

Our procedures were adapted from those described in the preceding chapters. Given our limited resources, 36-month follow-up interviews could be conducted only over the telephone or by post from a central site in Exeter.

Sample

Participants were those who were recruited and allocated at baseline to the intervention or control groups of the CADET study.

Ethical considerations

We applied for and received ethics approval for this trial extension from the NHS Health Research Authority, NRES Committee South West (NRES/07/H1208/60). Because of the unplanned nature of the 36-month follow-up, participants had not been warned that we would be contacting them at this point. Therefore, we sent them a letter explaining the trial extension with an opt-out slip attached that they could return to the research team if they did not wish to be contacted. If a participant could not be contacted by the research team, the GP and/or clinical commissioning group were contacted as appropriate and all reasonable attempts to re-establish contact were made.

Measures

Our outcomes at the 36-month follow-up point were depression (PHQ-9), quality of life (SF-36) and worry and anxiety (GAD-7).

Analysis

We assessed the baseline characteristics of participants followed up and compared them with the baseline characteristics of those lost to follow-up, using logistic regression predicting follow-up, adjusting for clustering, age, site, practice size and IMD. When differences between those followed up and those lost to follow-up were shown to be significant, we undertook further logistic regressions of treatment group on significant baseline variables, adjusted for age and minimisation variables.

We analysed outcome data by ordinary least squares or logistic regression, allowing for clustering by use of robust standard errors, adjusting at the cluster level for minimisation variables and site and at the individual level for age and, when appropriate, the baseline measurement of the variable.

We analysed the effect of missing data as a sensitivity analysis, estimated by chained regression equations multiple imputation48 using all available scale clinical scores, age, sex, practice variables, site and treatment group.

Results

At 36 months we obtained follow-up data, defined as primary outcome (PHQ-9) data, from 354 of the 581 participants who were observed at baseline (61% of those recruited).

Summary

Our results at 36 months’ follow-up showed that, after 3 years, participants in this trial were greatly improved with regard to depression, anxiety and quality of mental health. There were no longer any treatment differences, suggesting that the usual care participants had caught up with the earlier greater improvements in the collaborative care group. There was little change in physical quality of life; however, we would not have expected this as both collaborative care and usual care in this population were directed towards participants’ mental health. This provides evidence for the validity of the trial.

Clinical outcomes

Collaborative care improved depression compared with usual care at both 4 and 12 months’ follow-up. Our observed effect size was less than that used to power the study, although the 95% CI around it (0.26, 95% CI 0.07 to 0.46) encompassed our original target (0.4). Our result also lies within the 95% CI of the SMD found in the most recent meta-analysis of collaborative care,73 which includes our results (overall SMD 0.28, 95% CI 0.23 to 0.33), and is no different from US (SMD 0.28, 95% CI 0.21 to 0.35), non-US excluding the UK (SMD 0.36, 95% CI 0.13 to 0.59), and other UK (SMD 0.32, 95% CI 0.07 to 0.57) trials. 23 Collaborative care is as effective in the UK health-care system, an example of an integrated health system with a well-developed primary care sector, as in the USA. Our study adds to the emerging international literature from countries such as Chile26 and India27 indicating that collaborative care is a model that reliably generalises outside the USA.

We also found that participants rated their satisfaction with treatment more highly in the collaborative care group than in the usual care group, the largest difference between groups of any of our measures. These results are in line with data showing that most participants were adherent to collaborative care, with the majority receiving more than two contacts with care managers and the average contact rate being between five and six sessions. Collaborative care is therefore not only effective but patients receiving it are more satisfied with their care than those receiving usual care and are adherent to the intervention.

In the long-term, at 36 months, there was no significant difference between the groups. This confirms the picture described in the recent Cochrane review23 in which the clinical benefits of collaborative care were not found beyond 2 years after the intervention. Despite the lack of significant difference it is worth noting that in the collaborative care group > 5% more participants had responded or recovered at 36 months.

Economic outcomes

Although previous reviews74,75 have identified evidence from cost–utility (cost per QALY) studies to support the economic value of collaborative care for depression in the US health-care system, the CADET trial is the first study to estimate the cost-effectiveness of collaborative care in a UK primary care setting. We have shown that collaborative care is cost-effective compared with usual care in treating people with depression in a UK primary care setting when providers are willing to pay up to £20,000 per QALY gained. Furthermore, when taking a broader analytical perspective and including costs associated with informal care, the results show that collaborative care is expected to be cost saving, with expected health gains, and therefore dominates the usual care comparator.

Our cost-effectiveness analyses report an expected modest mean QALY gain for collaborative care at a relatively low cost. Although the mean QALY gain is modest, it is comparable, and favourable, to that recently reported in the evaluation of a UK Improving Access to Psychological Therapies (IAPT) service,76 which estimated a mean EQ-5D QALY gain of 0.014 (SF-6D gain of 0.008). Our estimated costs for health and social care in the CADET trial are similar to those reported in the IAPT service evaluation for IAPT service or comparator mental health-care services. 76 Furthermore, despite differences in the populations, the QALY gain from the collaborative care intervention is in a similar range to that reported in an evaluation of therapist-delivered cognitive–behavioural therapy for depression,77 in which the mean incremental QALY benefit was reported as 0.027 (95% CI –0.012 to 0.066).

Our base-case difference in health and social care (NHS and Personal Social Services) costs over 12 months (£272.50) and the subsequent cost per QALY estimate of £14,248 are heavily influenced by one participant who reported extremely high levels of service use for specialist care, including a 100-day stay in an acute psychiatric hospital. This participant, in the collaborative care group, had an estimated service use cost of £48,522 compared with a mean cost of £1637 for all other trial participants with cost data over 12 months (n = 446) averaged across both groups; 94% of participants had cost estimates of < £5000, all but four participants had cost estimates of < £10,000 and three participants had costs estimated between £10,000 and £24,000. When we excluded this one participant from the analyses, the difference in NHS and personal social services costs between collaborative care and usual care when including the cost of collaborative care was £63, with an estimated cost per QALY of £3334.

Our probabilistic analyses indicate that collaborative care has a 58% and 65% probability of being cost-effective at commonly assumed UK NICE willingness-to-pay thresholds of £20,000 and 30,000 per QALY respectively. When we considered the uncertainty around the cost-effectiveness estimate that excludes one participant with high service use and costs (£3334 per QALY), the probability of collaborative care being cost-effective at these cost per QALY thresholds was > 75%. The most conservative expectation (based on the intention-to-treat principle) would be that the introduction of collaborative care will involve an additional cost of £272.50 per participant for the UK NHS and this potential cost, alongside estimated EQ-5D QALY gains, will result in an expected cost per QALY of £14,342, which is similar to the base-case analysis presented here and represents a cost-effective use of NHS resources. We would suggest that the likely cost-effectiveness of collaborative care in practice might be closer to the estimate in the sensitivity analysis with the very high-cost participant excluded.

Process analyses

The principal finding from our moderation and mediation analyses was that the effects of collaborative care on depression at 4 months’ follow-up can be entirely attributed to the amount of behavioural activation undertaken by participants and the effects at 12 months are strongly mediated by behavioural activation. Despite the fact that collaborative care is a complex intervention that also includes medication management, we found evidence only for the mediating effect of behavioural activation. We found no evidence that depression severity, number of previous depressive episodes, attitudes towards treatment components, physical health problems or socioeconomic status influenced treatment outcome, and nor did we find that the care manager, number of contacts between care manager and participant and adherence to medication influenced outcome, albeit our study was not powered to detect these effects per se.

The second component of our process analysis, interviews with care managers, supervisors and GPs, showed that collaboration around the management of patients with depression in primary care was valued by professionals. However, GPs’ understanding of collaborative care, compounded by long-standing organisational barriers, hindered their engagement in the intervention. It is unclear whether or not more GP engagement would have led to better outcomes. Enhanced supervision as reported in this study may be collaboration enough to result in improved patient care.

Our final process evaluation element, in which we analysed care manager/participant contact audio-tape recordings, revealed variation in the way that care managers engaged participants, how they behaved clinically and how they undertook behavioural activation. Verbal expression of empathy, encouragement and positive feedback seemed to be key in achieving both initial engagement and its maintenance. Some care managers strayed from merely demonstrating empathy into a style of treatment more closely identifiable as counselling. There was variation in how the elements of behavioural activation were demonstrated by care managers, with considerable difference in how behavioural activation was both explained and operationalised. Despite the protocol focus on behaviour alone, some care managers also chose to address cognitions.

Strengths and limitations of the study

The CADET study is one of the largest studies of collaborative care with an integrated economic evaluation. Less than 50% of published collaborative care trials have followed up participants for ≥ 12 months and our levels of attrition at 4 and 12 months are comparable with those in 70% of collaborative care trials and better than those in other trials of brief interventions in this area. 78 There was no evidence that missing follow-up data biased the findings, even at 36 months when, not surprisingly, attrition was higher than at other follow-up time points.

Although our cluster design protected against contamination of the usual care arm by changes in behaviour being tested in the collaborative care arm, cluster trials are prone to selection bias. We minimised this bias by recruiting participants through electronic case note searches rather than clinician referral. Given the nature of the intervention and comparator we could not blind GPs, patients or care managers to treatment allocation but we used self-report outcome measures to minimise the impact of detection bias. The supervisors who we interviewed were also CADET co-investigators and therefore their views are likely to be framed by their academic investment in the study. As CADET trial researchers conducted the qualitative interviews some researcher bias may be evident, as this is likely to have affected the participants’ responses, particularly those of the supervisors and care managers. 79 We attempted to use interviewers from another study site to reduce this bias. Although purposive sampling of GPs was attempted, GPs were difficult to recruit to this qualitative study, with a majority of those who refused citing lack of time or limited involvement in the trial as reasons for this. However, category saturation was achieved within the data, although the difficulties experienced in recruiting GPs may mean that the data may not represent the views and experiences of GPs in all participating practices.

Our within-trial analysis demonstrates cost-effectiveness at the willingness-to-pay threshold of £20,000 per QALY, without the need to extrapolate potential benefits over the longer term. Our analyses used data collected within the trial to estimate resource use and costs associated with delivery of the intervention, but relied on self-report data from interviewer-administered questionnaires to estimate health and social care service use and broader resource impacts. Routinely collected service use data may have provided a more rigorous estimate of service use, particularly for primary care contacts. 80 However, there would be difficulties and costs related to the collection of service use data from 42 general practices and to necessary routine data collection for aspects of care not recorded in GP records and therefore we chose to use participant self-report data. We also relied on self-reported records of care manager contacts and so have no means to assess record accuracy.

Difficulties in collecting detailed data on medication use by self-report methods can lead to errors in self-reporting. These errors include the potential for variation in medication names, variation in reported dose and complexity in relation to the use of medications for a wide range of comorbid conditions. These issues led us to exclude medication costs from the economic analysis plan1 and this may be a limitation in the results presented here, as medication adherence has been shown to be one of the potential benefits of collaborative care. 23 However, most participants in both the collaborative care group and the usual care group remained on antidepressant medication (74.8% vs. 73.8% at 4 months; 69.7% vs. 69.2% at 12 months). Finally, the collection of resource use data using self-report methods over 4-month and 8-month durations may have introduced recall bias and this has not been explored as part of the current analyses.

Although care managers were already employed by organisations providing primary care mental health services in the UK NHS, supervisors were senior members of the investigator group and so it is unclear how much their, albeit minimal, supervision can be generalised beyond the trial. Our intervention was brief and it is possible that a more intensive intervention might have improved outcomes further, particularly for the more complex cases. We could have chosen a different psychological intervention such as cognitive–behavioural therapy81 but a review38 and a randomised controlled trial82 showed that behavioural activation is as effective as cognitive–behavioural therapy, potentially more so for severe cases,83 and can be delivered effectively by junior, less intensively trained health-care personnel. 82

The perspective on the analyses does not extend to the broad welfare and economic impacts of depression, including impact on productivity costs, as such costs are not included in the reference case analyses suggested by NICE56 for UK analyses. However, data collection did cover aspects of care and support and patient costs, which has extended the primary perspective (of NHS and Personal Social Services costs) to a broader patient- and societal-orientated perspective. We accept that the use of a relatively small number of categories for these broader considerations may be a limitation in the analyses. However, as in other studies (e.g. Romeo et al. 84), we found that resources and estimated costs associated with informal care were a dominant aspect when taking a wider perspective and this gives us clear guidance on the magnitude of these wider perspective costs.

Implications of the clinical and economic findings for the NHS

During the time that we undertook the CADET study the number of international trials of collaborative care more than doubled, albeit with many of them still conducted in the USA. Although the generalisability of a US collaborative care model to the UK had been suggested by previous small-scale studies, the CADET trial has provided definitive confirmation of generalisability and that collaborative care is preferred over usual care by depressed patients and is cost-effective in the UK. We have therefore answered a specific research need highlighted in the NICE guidelines for depression,28 providing critical evidence for service delivery improvement. The NICE guidelines can now be reviewed to reassess the place of collaborative care in the stepped care pathway in light of our findings that collaborative care is clinically effective and cost-effective for a range of depressed patients, not necessarily those who fail at other treatments.

Although our results sit within the expected effect range of collaborative care reported in the latest meta-analysis of international collaborative care trials,23 the clinical implications of our results are more difficult to interpret given that the average difference in treatment response was less than that we had anticipated (actual effect size 0.26 vs. 0.4 anticipated). Between-group differences can obscure response rates in individual patients. We have therefore presented the data on meaningful clinical difference using numbers needed to treat and two criteria commonly applied in the depression literature and regarded as clinically meaningful: that of recovery (falling below a recognised point on the PHQ-9 symptom scale) and response (a ≥ 50% reduction in symptoms of depression). Using these metrics it is particularly noteworthy that at 12 months 56% of patients receiving collaborative care were ‘recovered’, 15% more than in usual care. Health services would therefore need to treat 6.5 patients using collaborative care to produce one additional patient with a sustained recovery compared with usual care.

Studies that have achieved higher effects have been undertaken in countries with less developed primary care services26,27 or have used more highly qualified workers such as nurses or social workers. 85 These workers are in acute short supply in the UK and hence this would not have been a translatable model for our health system.

Our results represent a slightly higher recovery rate than that reported by the UK IAPT programme. Recovery rates of around 45% have been achieved in the IAPT programme following an investment of £700M over 6 years. 86 We suggest that integration of our CADET protocol into IAPT services might enhance outcomes for depressed patients receiving treatment and provide guidance to international mental health services that this model can be applicable outside the USA.

The finding that collaborative care is the dominant intervention compared with usual care when we included informal care costs in our analyses is an important issue. Family members and others involved in informal care contribute to the care of depression in a substantial manner. As Richard Layard87 and others have often asserted, the implications of these care costs on productivity are a significant burden on the economic activity of a nation. The introduction of collaborative care has the potential to relieve some of the significant burden that falls on informal carers and reduce this economic load.

Given that collaborative care is more effective over a sustained period of time than usual care and represents value for money to the NHS at commonly used thresholds for cost per QALY, we suggest that commissioners of health care in the UK might review the organisation of their routine depression management services and consider using a collaborative care model. Evidence here indicates that services could benefit from being commissioned to support the management of patients with depression in UK primary care using a collaborative care model, as it would be both clinically effective and cost-effective to do so.

Implications of the results for treatment development and future research

Although collaborative care is an organisational intervention that improves outcomes, much remains to be done to improve the effectiveness of treatments for depression. Even intensive psychological treatments for depression have been shown to achieve only modest gains (effect size of 0.42 in 51 studies88). Our careful selection of intervention ingredients, directed by our identification of components present in the better-performing trials from our previous metaregression,89 did not succeed in achieving the larger effects we had hoped for. In the trial, 44% of participants receiving collaborative care had PHQ-9 scores that remained above the PHQ-9 depression threshold at 12 months’ follow-up.

The strong finding that behavioural activation was the only mediator of effect is important. Behavioural activation is hypothesised to operate by providing people with more opportunities for positively reinforcing experiences and reducing the amount of negatively reinforced avoidance behaviours, thereby improving affect. This hypothesis would appear to be supported by our results. We would encourage future researchers to measure the mediational contribution of behavioural activation to patient outcomes in other trials, to determine the universality or otherwise of our findings. If replicated, this would have considerable implications for the inclusion of activating strategies in other effective depression management strategies. Further, in the context of the findings from our audio-tape analysis that there was variation in the delivery of behavioural activation in care managers’ contacts with participants, we may have achieved better outcomes, more in keeping with our predictions, had care managers been able to deliver a more rigorous and consistent behavioural activation programme. We may have achieved better outcomes, more in keeping with our predictions, had care managers been able to deliver a more rigorous and consistent behavioural activation programme. Our supervision model did not require supervisors to listen to audio tapes and give care managers specific clinical feedback on these tapes. It might be that we could optimise the delivery of behavioural activation to ensure that it is more consistent by amending our supervision model to require supervisors to sample audio tapes and feed back their analyses. It seems clear that clinical and supervisory practice should focus much more on enhancing the quality and consistency of behavioural activation within the collaborative care model. However, our supervisory system was delivered by professionals with considerable expertise in case management and behavioural activation. The ability of routine services to replicate this standard of supervision would need to be tested and then monitored in any implementation programme.

The limited liaison reported between GPs and care managers could suggest that more work is needed to facilitate collaboration around individual patients. Some structural aspects were identified that may facilitate liaison, including shared place of working, shared information technology systems, facilitating opportunities for informal meetings and building in formal collaboration into the collaborative care framework. However, although considered by many to be desirable, it is unclear if such an increase in collaboration between care managers and GPs would improve clinical outcomes. Despite being a core component of the collaborative care model, the extent to which additional collaboration between care managers, supervisors and GPs beyond established communication lines is actually necessary for effective patient management of depression is as yet undetermined. For example, a recent trial of collaborative care for people with long-term physical health conditions and depression90 introduced joint sessions with primary care staff (practice nurses) and mental health-care managers to reflect the comorbid problems that patients had, but the effects were no different from those in the CADET study.

Although the vast majority (98%) of our participants had a secondary diagnosis of anxiety we found no differential effect of collaborative care compared with usual care on anxiety at any follow-up point. Patients in both groups had less anxiety at 12 months in particular but the difference between the two groups at this point fell just short of being significant. It seems that a fruitful area of potential treatment development would be the addition of specific anxiety-directed treatment components to the basic collaborative care package. In addition, the high prevalence of long-term physical health conditions in our study population is worthy of note and suggests that specific attention given to this aspect of collaborative care might reap rewards. Indeed, several studies in the USA91,92 and more recently the UK90 have done this. Unfortunately, the addition of this focus in the UK did not deliver significantly enhanced outcomes compared with the CADET study. 90

Future trials should therefore test enhancements of the basic collaborative care model by developing, testing and delivering better treatments within the effective collaborative care organisational framework or improve the delivery of existing treatments through more rigorous supervision, rather than test collaborative care per se, given that the effects of collaborative care are now firmly established.

Contributions of authors

David A Richards, Peter Bower, Carolyn Chew-Graham, Linda Gask, Karina Lovell, John Cape, Stephen Pilling, Ricardo Araya, David Kessler, Michael Barkham, J Martin Bland, Simon Gilbody, Colin Green, Glyn Lewis and Chris Manning designed the study and were responsible for its conduct.

Jacqueline J Hill, Adwoa Hughes-Morley and Abigail Russell were responsible for study management and data collection.

David A Richards, J Martin Bland, Colin Green, Evangelos Kontopantelis and Jacqueline J Hill undertook data analysis.

All authors had full access to all of the data (including statistical reports and tables) in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis. All authors contributed to the writing and editing of the report.

Publications

Richards DA, Hughes-Morley A, Hayes RA, Araya R, Barkham M, Bland JM, et al. Collaborative Depression Trial (CADET): multi-centre randomised controlled trial of collaborative care for depression – study protocol. BMC Health Serv Res 2009;9:188.

Richards DA, Hill JJ, Gask L, Lovell K, Chew-Graham C, Bower P, et al. CADET: clinical effectiveness of collaborative care for depression in UK primary care. A cluster randomised controlled trial. BMJ 2013;347;f4913.

Coupe N, Anderson E, Gask L, Sykes P, Richards DA, Chew-Graham C. Facilitating professional liaison in collaborative care for depression in UK primary care; a qualitative study utilising normalisation process theory. BMC Fam Pract 2014;15:78.

Green C, Richards DA, Hill JJ, Gask L, Lovell K, Chew-Graham C, et al. Cost-effectiveness of collaborative care for depression in UK primary care: economic evaluation of a randomised controlled trial (CADET). PLOS ONE 2014;9:e104225.

Data sharing statement

The authors confirm that all data underlying the findings are fully available without restriction. The authors have made the clinical and economic data set available through the University of Exeter’s Institutional Repository – Open Research Exeter (see https://ore.exeter.ac.uk). Access to these data is permitted but controlled through requests made via the repository to the chief investigator (Professor Richards: d.a.richards@exeter.ac.uk). Although use is permitted, this will be on the basis that the source of the data is acknowledged (including the funder) and it includes reference to the data set ‘handle’.

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, the MRC, NETSCC, the HTA programme, the EME 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, the EME programme or the Department of Health.