Behavioural activation therapy for post-stroke depression: the BEADS feasibility RCT

Stroke

Stroke is a condition in which interruption of the blood supply to the brain causes brain damage. This leads to the impairment of motor, sensory and cognitive abilities. Cognitive impairments include disorders of communication, such as aphasia, and problems with attention, memory, visuospatial abilities and executive function. These cognitive impairments have a negative effect on recovery and long-term outcomes. 1–3

Depression after stroke

Many stroke patients experience emotional consequences, including depression, anxiety, post-traumatic stress, anger, apathy and frustration. Depression is the most commonly investigated emotional consequence of stroke. 4,5

Several recent reviews5–8 have reported that about one-third of stroke patients have depression at any time point. Two of these reviews1,2 included a meta-analysis. Ayerbe et al. 1 analysed data from 43 studies published between 1983 and 2011. They reported an average prevalence of depression among stroke survivors of 29% [95% confidence interval (CI) 25% to 32%] of stroke survivors with depression, with a prevalence of 28% (95% CI 23% to 34%) within 1 month of stroke, 31% (95% CI 24% to 39%) at 1–6 months, 33% (95% CI 23% to 43%) at 6 months to 1 year and 25% (95% CI 19% to 32%) at > 1 year. Hackett and Pickles2 conducted a similar review and identified 61 studies. They obtained a pooled prevalence of depression among stroke survivors 31% (95% CI 28% to 35%) at any time, up to 5 years after stroke. Hackett et al. 3 also highlighted that this figure was not significantly different from the proportion in their earlier review (33%, 95% CI 29% to 36%; difference of 2%, 95% CI < 1% to 3%), suggesting that the management of the problem had not substantially improved over the previous 10 years.

Factors associated with depression after stroke

Effective treatment of depression is important because depression is associated with worse rehabilitation outcomes4–7 and increased disability. 1 Stroke survivors who are depressed may engage less in rehabilitation, which, in turn, can lead to decreased functional recovery. 4 Depression is also associated with increased mortality. 1,8,9 Not only does depression affect stroke survivors themselves but it also has an effect on their carers. 10 It has cost implications for the NHS because it is associated with increased health-care utilisation. 9

Most studies have assessed depression and its potential outcomes at the same time point, making it unclear whether depression is a cause or consequence of the outcome variable. Ayerbe et al. 1 reviewed only those studies in which depression was assessed at an earlier time point than the outcome and found that disability, lower quality of life (QoL) and mortality may be outcomes of depression in stroke survivors. In a more recent review, Towfighi et al. 11 concluded that the most consistent predictors of post-stroke depression are physical disability, stroke severity, history of depression and cognitive impairment. They also reported that post-stroke depression is associated with higher rates of health-care use after stroke. Therefore, in addition to improving mood, effective treatment of post-stroke depression is important as this has the potential to improve functional outcomes and QoL and reduce health-care costs.

Many studies of depression after stroke are based on clinical interviews or questionnaires to assess depression but these may not be appropriate for those with communication problems. About one-third of stroke survivors have aphasia,12,13 which may affect all communication modalities, namely speaking, understanding, reading and writing. Studies that have used measures of depression appropriate for those with communication problems have reported that stroke survivors with aphasia may be particularly susceptible to post-stroke depression. 14,15

Psychological treatments for depression after stroke

Previous research16 indicates that a high proportion of depressed stroke patients are likely to be taking antidepressants and so suggests that antidepressants have not resolved the mood problem. Previous research also suggests that few stroke survivors receive ongoing psychological treatment. 17 The Communication and Low Mood (CALM) trial of behavioural activation (BA) for low mood in people with aphasia17 found that, at 3-month follow-up, only 14% of participants who had been identified as having low mood after stroke had received mental health treatment in the past 3 months (from a mental health nurse, counsellor, psychologist, or psychiatrist). Although Improving Access to Psychological Therapies (IAPT) has extended its remit to include people with physical health problems,18,19 the current uptake by stroke survivors is unknown.

Among the several psychological approaches to the treatment of depression, cognitive–behavioural therapy (CBT) is the most widely used psychological treatment for depression in clinical practice20 and may be appropriate for those with stroke. 21,22 There is evidence from single case design studies that some patients with post-stroke depression improve following CBT. 23,24 However, a randomised controlled trial25 (RCT) of CBT for post-stroke depression found no significant difference between those participants who received CBT, an attention placebo or usual care. One of the possible reasons for the lack of efficacy was that psychological treatments need to be tailored for people with aphasia and cognitive impairment. 22,25 A systematic review26 of the modifications to CBT that were required for people with cognitive impairments caused by acquired brain injury reported promoting an understanding of how specific changes to cognition, affect and behaviour occur as a result of brain injury and the use of memory aids. 27 However, a randomised trial of augmented CBT, in which CBT was adapted to suit those with stroke, also found no evidence of benefit in comparison with a cognitive training control group. In addition, a trial28 of CBT at different time points after stroke found no overall effect of CBT in comparison with usual care, although there was some evidence that CBT improved mood in those who were recruited > 9 months after stroke. Therefore, there is currently little evidence to support the provision of CBT after stroke.

Effectiveness of psychological interventions for depression after stroke

Other psychological interventions that may be appropriate for those with depression after stroke include counselling, motivational interviewing and problem-solving training. Some of these have been provided early after stroke in an attempt to prevent the development of depression,29–35 whereas others have been provided later to those who have developed depression. 36,37

There is currently limited evidence for the clinical effectiveness and cost-effectiveness of these psychological therapies for treating post-stroke depression. 38 Towfighi et al. 11 identified seven trials (n = 775 participants) of psychological interventions for depression after stroke and concluded that these trials suggest that brief psychosocial interventions may be useful and effective in the treatment of post-stroke depression. Two of these trials36,37 evaluated a brief psychosocial behavioural intervention, but details of the content of the intervention are limited. Motivational interviewing31,35 has also been shown to reduce post-stroke depression but studies recruited participants early after stroke and excluded those with severe communication or cognitive problems, so these findings may not be applicable to the broad range of stroke survivors with post-stroke depression.

Primary objective

The primary objective was to determine the feasibility of proceeding to a definitive trial.

Secondary objective

The secondary objective was to determine the feasibility of delivering BA to people with post-stroke depression.

Trial design

This study used a parallel-group, feasibility, multicentre RCT design with nested qualitative research and economic evaluation to compare BA therapy with usual stroke care for patients with post-stroke depression. Participants were allocated to BA or usual stroke care at a ratio of 1 : 1.

Ethics approval

Ethics approval was granted on 29 January 2015 by the National Research Ethics Service (NRES) Committee East Midlands – Leicester (reference number 15/EM/0014).

Important changes to the methods after feasibility trial commencement

Changes made to the essential documentation during the trial and following ethics approval on 29 January 2015 can be found in Appendix 1.

Early in the recruitment phase substantial amendment 2 [Protocol v2.1, Research Ethics Committee (REC) approved 8 July 2015] provided additional options in the recruitment process. This allowed the therapists to contact patients directly by telephone following initial consent to be contacted. This streamlined the recruitment process by providing the opportunity for the therapist to explain more about the research and arrange a home visit to complete screening measures. This amendment also broadened the recruitment routes to include potential participants on acute outpatient caseloads.

In substantial amendment 3 (Protocol v2.2, REC approved 7 September 2015), an additional exclusion criterion were added to exclude patients who were currently receiving psychological intervention. Five out of the 49 participants were recruited prior to this exclusion criterion being added. Furthermore, it was specified that participants in the intervention group could be withdrawn from the intervention if it was subsequently agreed that the patient needed immediate clinical psychology input. This amendment also clarified that participation in the study would not compromise access to other services (i.e. psychological input) that were part of usual care.

Minor amendments were also made including changes in study personnel and contact details. In addition, another secondary end point was added to estimate the sample size for a definitive trial and clarification that two or fewer missing items within the Patient Health Questionnaire-9 (PHQ-9)58 could be imputed if required.

Participants and eligibility criteria

Settings and locations where the data were collected

The University of Nottingham sponsored the trial. Co-ordination of the trial was undertaken by the Sheffield Clinical Trials Research Unit (CTRU). Participants were identified and recruited from hospital and community stroke services at Sheffield Health & Social Care NHS Foundation Trust, Derby Teaching Hospitals NHS Foundation Trust and Let’s Talk Wellbeing at Nottinghamshire Healthcare NHS Foundation Trust. Screening and identification of potential participants was supported by Clinical Research Network (CRN) staff based at Sheffield Teaching Hospitals NHS Foundation Trust, Derbyshire Community Health Services NHS Foundation Trust and at Sherwood Forest Hospitals NHS Foundation Trust. Baseline data were collected and BA therapy was delivered by the therapists in participants’ own homes. Six-month follow-up data were collected in participants’ own homes, for those with aphasia, by blinded outcome assessors. For those without aphasia, data were collected by post unless help was requested.

Interventions

Feasibility criterion

Changes to trial outcomes after the trial commenced, with reasons

An additional secondary end point, to estimate the sample size for a definitive trial, was added to the protocol after the trial commenced, as an inconsistency was identified between the statistical analysis plan (SAP) and the protocol.

Sample size

As this was a feasibility study, it was not powered for efficacy and no formal interim analyses of efficacy were conducted; the sample size was adequate to estimate the uncertain critical parameters [standard deviation (SD) for continuous outcomes; consent rates, event rates and attrition rates for binary outcomes] needed to inform the design of the definitive RCT with sufficient precision. The sample size of 60 participants allowed SDs for continuous outcomes, such as the PHQ-9 and VAMS ‘Sad’ item, to be estimated to within precision of approximately ± 19% of its true value (with 95% confidence). Allowing for 15% attrition by 6 months post-randomisation follow-up, 72 participants needed to be recruited. To achieve the target sample size of 72 over the 12-month recruitment period, with three centres, we needed to randomise two participants per centre per month.

In addition to this, we estimated that we would recruit a total of 65 carers and three therapists to the study. The carer estimate was based on the CALM study,17 in which approximately 90% of people with stroke had an informal carer present who completed the study outcome assessments.

Further information on both the quantitative and the health economic analyses is provided in the SAP. This covers both the procedures for missing, unused and spurious data and definitions of populations for which data were analysed.

Explanation of any interim analyses and stopping guidelines

The overall study could have been stopped because of safety concerns or issues with study conduct at the discretion of the sponsor. There were no formal statistical criteria for stopping the trial early. Decisions to stop the trial early on grounds of safety or futility would have been made by the Trial Steering Committee (TSC) on the basis of advice from the Data Monitoring and Ethics Committee (DMEC). No early stopping was planned, but the study could have been terminated early if, in the view of the TSC, no useful information was likely to be obtained by continuing. The criteria for assessing this were primarily the feasibility outcomes listed in Feasibility criterion. The TSC could also have recommended the closure of a centre but that the trial as a whole continued, on the same grounds. Unblinded adverse event (AE) data were reviewed by the DMEC, who could have recommended to the TSC that the trial was stopped if, in their opinion, there was evidence of harm in the intervention group. As this was a feasibility trial, it would not have stopped early for efficacy.

Method used to generate the random allocation sequence

Randomisation was conducted using a computer-generated list with random permuted blocks of varying sizes, created and hosted by the Sheffield CTRU in accordance with their standard operating procedures and was held on a secure server. Once a participant had consented to the study, the therapist logged into the remote, secure, internet-based randomisation system and entered basic demographic information. The allocation for that participant was then revealed to the researcher.

Type of randomisation and details of any restriction (such as blocking and block size)

Block randomisation with randomly varying block sizes of two, four and six was used so that the sequence of allocation could not be predicted. The block sizes were determined by the trial statistician and block size was not revealed to any other member of the study team. Participants were allocated to BA or usual stroke care at a ratio of 1 : 1. Randomisation was stratified by site.

Allocation concealment mechanism

Access to the allocation sequence was restricted to those with authorisation. The sequence of treatment allocations was concealed until interventions had been assigned and recruitment, data collection and analyses were complete.

Neither the participants nor the therapists were blind to which treatment the participants were receiving. The outcome assessors were blind to the treatment received and there was no requirement for them to know the treatment allocation at any stage. As a result, a procedure for breaking the code was not necessary.

Blinding

Participants were randomised at baseline (after consent and baseline assessments) in equal proportions to BA or usual stroke care. It was not pragmatically possible for the participant or therapist to be blind to the group allocation, but the researchers completing the 6-month outcome assessments were blinded and had no involvement in any other aspects of the trial. The researchers were asked to record whether or not they thought they were unblinded and were also asked to guess the group allocation. We followed guidelines71,72 to minimise unblinding during RCTs of rehabilitation.

The trial statisticians remained blind until data freeze, at which point data checks were carried out on unblinded data.

Statistical methods

As the trial was a pragmatic, parallel-group RCT, data were reported and presented in accordance with the CONSORT 2010 Statement. 55 All statistical analyses were performed in R version 3.3.1 (The R Foundation for Statistical Computing, Vienna, Austria). 73 As a feasibility study, the main analysis was descriptive and focused on CI estimation and not formal hypothesis testing.

Analysis populations

The intention-to-treat (ITT) population includes all participants for whom consent was obtained and who were randomised to treatment, regardless of whether they received the intervention. This is the primary analysis set and end points were summarised for the ITT population unless otherwise stated.

Patient and public involvement

The BEADS trial received input from the patient and public involvement (PPI) group (including one patient with significant aphasia) on aspects of the design and methods development as well as study oversight. Two patients and one carer attended five scheduled meetings to discuss feedback on how the study was being conducted, including ideas about recruitment, study documents, ensuring the well-being of the patients and carers, and supporting the therapists. The meetings were attended by the PPI group members, the trial manager and the chief investigator. Meetings took place prior to receiving ethics approval when study materials were being developed, during recruitment and intervention delivery, and after the study to discuss the study results. The meetings each had an agenda that was agreed by the group. A summary of the discussions was written up by the trial manager and chief investigator and was circulated to the group for them to add any points and to ensure that it was an accurate summary of the meeting. Suggestions were put into practice with the creation of a short study summary card and a spiral-bound version of the aphasia-friendly participant information sheet. At the suggestion of the patient and carer representatives, the therapists were invited to join the PPI meetings. In these meetings, the PPI group members were able to ask the therapists questions and give suggestions. PPI group members asked the therapists if they found their job difficult and whether or not there were any challenges with delivering the intervention. The therapists explained that they were well supported and that they enjoyed their role. The PPI group members felt that this was crucial to the success of the study. Another question asked by the PPI group was whether the therapists felt that having the carer being present during therapy was helpful or not. The therapists explained that the carer provided support to the patient. One of the therapists said that they made sure they addressed any discussion or questions to the participant directly, so that they could choose when they wanted their carer to answer on their behalf.

Background

The health economic analysis had two key components that related to the primary end points of the trial:

1. assessing the feasibility of collecting data that may be used in a health economic analysis

2. conducting an economic evaluation and a value-of-information analysis in order to provide information on the potential value of conducting the definitive trial.

Overview

For the feasibility analysis, the number of participants who had complete data for each of the key measures is reported for each time point by treatment group and overall. For patient and carer questionnaires, the item response rate at each visit (baseline and 6 months) is reported. Response rate was measured as a fraction of the total number of items. This provides an analysis of the feasibility of collecting data required to complete a health economic analysis. For the health economics analysis, the data of most relevance are those from the:

• EQ-5D-5L – standard version (completed by participants who are able)

• EQ-5D-5L – aphasia-accessible version (completed by all participants)

• EQ-5D-5L – completed by the carers of participants for themselves

• EQ-5D-5L – completed by the carers of participants on behalf of the participant

• resource use questionnaire.

For the economic evaluation, a series of cost-effectiveness analyses were conducted:

• within-trial analysis from a NHS and Personal Social Services (PSS) perspective

• within-trial analysis from a societal perspective

• model-based analysis from a NHS and PSS perspective

• model-based analysis from a societal perspective.

Owing to the importance of carers for people with post-stroke depression, it was important to include analyses undertaken with a societal perspective to supplement the NHS and PSS analyses.

The within-trial analyses were undertaken both with and without multiple imputation, which was used to estimate values for missing data. Patient-level costs and outcomes were assessed over the full length of the feasibility study and this was supplemented with the construction of a simple economic model to examine the longer-term cost-effectiveness of treatment and priorities for future research. Costs and utilities were estimated for individual patients using data collected at baseline and follow-up, based on responses to EQ-5D-5L and resource use questionnaires, combined with standard cost and valuation sources. 79,80 Differences between costs and quality-adjusted life-years (QALYs) in the two groups were described and the incremental cost-effectiveness ratio (ICER) was calculated.

The main aim of the BEADS trial was to assess the feasibility of conducting a future definitive RCT to investigate the clinical and cost-effectiveness of BA therapy for people with post-stroke depression. Therefore, our analysis cannot provide conclusive cost-effectiveness results. However, early cost-effectiveness modelling remains of value because it provides insight into the likely cost-effectiveness of the intervention and demonstrates the value of pursuing further research, particularly when value-of-information analyses are included. 81,82 The value-of-information framework allows the maximum value of further research to be estimated, taking into account the uncertainty in the parameters included in the economic model. 83,84 We estimated the expected value of perfect information (EVPI), representing the maximum value of further research on all uncertain parameters in the economic model, and also estimated the expected value of perfect partial information (EVPPI), representing the maximum value of obtaining more information on each specific parameter (or group of parameters) included in the model.

Resource use

Costs were estimated for each participant, including intervention costs (based on staff time and number of sessions) and health-care resource use. Questionnaires were tested as a method for collecting resource use data. The resource use questionnaire included questions about a participant’s use of health services over the previous 3 months, representing the final 3 months of the follow-up period. As data were required for the entire 6-month follow-up period, we assumed that costs for the first 3 months were the same as for the final 3 months. In the questionnaire, resources were split by services, such as inpatient, outpatient, primary care and community services, and, where necessary, included average appointment length. Participants were also asked to record dosages of medication relating to depression, and information about carer time and employment. This information was used to calculate total medical costs and societal costs.

Unit costs

Resource use data were combined with unit cost data from the latest versions of the Personal Social Services Research Unit (PSSRU) unit cost publication,85 NHS reference costs79 and the British National Formulary86 in order to calculate costs for inclusion in the economic analysis. When appropriate values were not available from the latest version of the PSSRU unit costs publication, earlier versions were consulted87–89 and prices were inflated using the hospital and community health services index. 85

The unit costs used to estimate the costs associated with the resource use observed in the trial are presented in Appendix 3.

Outcomes

Participants who did not have moderate or severe language problems were asked to complete the standard version of the EQ-5D-5L as well as an amended (and as yet unvalidated) accessible version (based on pictures). 57 Participants who had moderate to severe language problems were asked to complete the accessible version of the EQ-5D-5L. In addition, for participants who had carers, the carer was asked to complete a standard EQ-5D-5L by proxy. This allowed us to test alternative methods for collecting data from which to calculate QALYs.

Analysis

Model-based analysis

The trial-based analysis was supplemented with an analysis using a simple decision-analytic model, used to estimate the cost-effectiveness of the intervention over the lifetime of participants. This was populated using the trial data combined with unit costs and mortality rates as well as assumptions regarding the maintenance of the treatment effect over time. The base-case analysis was undertaken from a NHS and PSS perspective, but a supplementary societal analysis was also undertaken.

The structure of the model (Figure 1) was based on that used to estimate the cost-effectiveness of computerised aphasia treatment compared with usual stimulation in the CACTUS study. 70 A simple three-state Markov model was used to extrapolate the data from the trial to a simulated cohort over a lifetime horizon. Participants entered the model in the no response state. Each month, they could remain in this state or transition to the good response state or death. Once in the good response state, participants could remain in this state or move back to the no response state or to death.

FIGURE 1.

Markov model.

Transition probabilities were primarily based on the trial data. The primary clinical outcome measure was the PHQ-9; therefore, we based our definition of ‘good response’ on PHQ-9 scores. Specifically, participants moved from the ‘no response’ state to the ‘good response’ state if they achieved a 4.78-point decrease in PHQ-9 score from baseline to follow-up. This definition of a response was chosen based on the minimum important clinical difference of PHQ-9 reported by Löwe et al. 75 In the model, we assumed that the intervention would be given over 4 months, as in the trial, and a response (if achieved) would occur after 3 months. As the trial had only one follow-up time point, it was not possible to estimate a relapse rate for a good response. Hence, in the base case, it was assumed that the relapse rate was zero. A sensitivity analysis was carried out to estimate the effect on cost-effectiveness of changing the relapse rate; in this analysis, it was assumed that participants in the ‘good response’ state could move back to the ‘no response’ state after 6 months.

Transitions from the ‘no response’ and ‘good response’ states to death were based on evidence on long-term survival following stroke,96 combined with background mortality rates from the Office for National Statistics,97 reflecting the approach taken in a previous economic evaluation of an intervention for people with aphasia. 70 The same mortality rate was used for the ‘no response’ and ‘good response’ states.

The HRQoL utility scores applied to each health state were reduced over time on the basis of multipliers estimated by Ara and Brazier. 98 QALYs were estimated for each cycle of the model by combining utility scores with life-years, allowing the total QALYs associated with each treatment strategy to be calculated. Costs and QALYs were discounted at a rate of 3.5% each year, in line with recommendations made by the National Institute for Health and Care Excellence (NICE). 99

Distributions were placed around each of the uncertain parameters included in the model for use in probabilistic sensitivity analysis (PSA), which allowed the estimation of cost-effectiveness acceptability curves (CEACs) and a value-of-information analysis. Gamma distributions were used for costs, log-normal distributions for utilities, and beta distributions for probabilities, with dispersions based on numbers observed in the trial. The PSA was supplemented with deterministic scenario analysis on the relapse rate as this was not observed directly in the trial.

The EVPI and EVPPI analyses were undertaken assuming a cost-effectiveness threshold of £20,000 per QALY gained (based on NICE decision rules99) over a 10-year period (assuming that it might take 10 years before a new treatment for these patients is developed), using a 3.5% discount rate. The Sheffield Accelerated Value of Information tool100 was used to estimate the EVPI and EVPPI.

Interviewer characteristics

Three women and one man conducted the interviews with patient and carer participants. They were Doctor of Philosophy (PhD) students and research assistants who were registered allied health professionals (one was a speech and language therapist) or working towards gaining professional registration as an allied health professional (psychologist). They had experience of working with people with neurological conditions and had been trained to conduct interviews with patients and carers, including those with reduced language and cognitive ability. Dr Gogem Topcu, who has expertise in health psychology and research into long-term conditions, conducted the therapist interviews. She is an experienced qualitative researcher who has adopted a realist pragmatic approach to research; she had no previous experience of conducting BA therapy.

Relationship with participants

The interviewers’ role was to guide and facilitate the interview, rather than impose how the interview proceeded. The interviewers had no involvement in the delivery or provision of care for the participants and participants were aware of this. The person who interviewed the therapists did not have any prior relationship with the therapists. The interview analysts did not have a personal view on the benefits or limitations of the intervention or about participants’ experience of taking part in the study.

Theoretical and thematic framework

A framework approach was adopted, which is a hierarchical, matrix-based method for ordering and synthesising qualitative data. 101,102 This approach enables in-depth exploration of the data while simultaneously maintaining an effective and transparent audit trail. 101,103 Adopting this approach consolidated the rigour of the analysis and the credibility of the findings.

The thematic framework was constructed iteratively from the interview objectives and existing literature (e.g. previous trials of BA) as well as the issues the participants raised during the interviews. We followed Ritchie and Spencer’s102 approach in arriving at the final thematic framework, which began with familiarisation of the transcripts and audio recordings of the interviews to gain an overview of the data, noting salient points and recurrent ideas. The thematic framework was further consolidated by amalgamating the notes taken during the familiarisation process and the a priori issues covered in the interview schedule and extant literature. However, we did keep an open mind to incorporate material that did not fit within our predefined structure, and iteratively checked whether or not additional themes were warranted or whether or not some predefined themes needed to be minimised in terms of importance or relevance, based on the interview (see Appendix 4 for a worked example of an audit trail of how a framework was developed and amended).

Participant selection

We used a purposive, maximum-variation sampling strategy to select participants and their carers for the interview phase. We attempted to recruit a heterogeneous sample based on their demographic and stroke characteristics. The selection strategy was designed to balance the sample of participants in the following categories:

• recruitment site

• treatment arm

• sex of participant

• level of depression (mild, moderate or severe)

• aphasia status

• recruitment (early or late to the study).

As there was a large number of stratification factors, we prioritised selecting participants by site and treatment arm, and then by the other factors. The sample size was guided by our previous experiences of soliciting feedback about trials from stroke survivors and patients with neurological conditions (e.g. das Nair and Lincoln104), whereby we felt that we were able to achieve sufficient detail related to each construct under investigation, and the categories had ‘conceptual depth’. 105 We were also guided by the limits placed by the recruitment process of the main trial and our framework analysis method. Our primary criterion for conceptual depth related to the question, ‘Do we have sufficient data for each key question that represents rich, nuanced, wide ranging experiences, that resonate with (agree or depart from) the extant literature?’ (cf. Nelson106). Therefore, we did not seek to achieve ‘data saturation’ per se. The trial therapists from each participating site were invited to take part in interviews about their experiences of working on the trial and delivering the intervention.

Data collection

The interviewers had sufficient knowledge of the interview schedule to loosely follow the questions in line with interests and views offered by the participant. Although a predetermined and structured schedule was used to conduct the interviews, the interviews were flexible, allowing the emergence of issues relevant and important to interviewees that were not in the initial schedule. This allowed participants to say what they felt was important to them, while maintaining a basic framework of inquiry at the same time. Prompts were used throughout interviews to provide cues when participants had difficulties, to clarify questions and encourage responding.

The interview schedule included general questions for all participants and carers concerning their experiences of participating in the study, and specific questions to those who had received the intervention and those who were in the control group (see Appendices 5 and 6). An accessible version of the interview questions was also developed for participants with aphasia. The interview schedule for therapists included questions relating to therapists’ experiences of delivering the therapy, views on the trial procedures and practical aspects of delivering the therapy (see Appendix 7). Interviews were transcribed verbatim. The transcripts did not include any personal identifiers and the recordings were deleted on completion of the transcription.

Data analysis

Data analysis was undertaken by Gogem Topcu using the framework approach,101,102 for which the data were mapped onto the constructed thematic framework. If required, the framework was amended to include new concepts or themes introduced during the interviews. To map the data onto the theoretical framework, we relied on indexing various sections of the data to specific thematic constructs. This, again, was an iterative process, requiring Gogem Topcu to go back to previously analysed transcripts to check whether or not the newly emerged construct was also evident there. After mapping all the data, a matrix was generated in which the data were charted to summarise each main theme. This matrix was then used in the interpretation of the data in addition to the notes made during the coding process. The interpretation process, like the other processes, was iterative and relied on consultation between Gogem Topcu and Roshan das Nair regarding the viability and relevance of a theme, to interrogate theoretical constructs, and to unpack nuances within the data.

To ensure rigour and credibility of the findings, Roshan das Nair reviewed the generated matrix and checked the quotations for their relevance to the themes. Disagreements were resolved by discussion. We provided information regarding the context in which the interviews were conducted, some pertinent descriptions of the interviewers and interviewees, and some verbatim quotations from our participants, to ensure transparency. To ensure quality of study reporting, we used the COnsolidated criteria for REporting Qualitative research (COREQ) checklist (see Appendix 8).

Implementation summary

The trial was due to start recruitment at the beginning of May 2015 but, owing to delays in set-up, recruitment did not start until 27 May 2015 in site 3, 20 August 2015 in site 1 and 3 September 2015 in site 2. This resulted in a total of 27 recruitment months available rather than the planned 36 months. Recruitment ended as planned on 30 April 2016. Follow-up was completed in November 2016. There were a number of issues in regard to recruitment, data collection and delivery of the interventions, which are discussed below.

In order to maintain anonymity, the sites will be referred to as ‘site 1’, ‘site 2’ and ‘site 3’ rather than by their geographical name.

Recruitment to the trial

The study set out to assess feasibility of recruitment by utilising a number of different recruitment routes including hospital database, community database, stroke ward, community caseload, voluntary group, self-referral and outpatients. The CONSORT flow diagram (Figure 2) shows the flow of participants through the trial. See Chapter 2, Feasibility criterion, for the primary end points referred to in the analysis section.

FIGURE 2.

Study CONSORT flow diagram. a, One participant was randomised in error (found to be ineligible within days of being randomised), this participant was excluded from analyses; b, one participant withdrew consent 5 days after randomisation, this participant was included in analyses.

Table 2 displays a summary of recruitment flow by screening route and Table 3 shows the number identified by the different screening routes for each site. The number randomised signifies the success of each route; however, it is necessary to discuss each method in turn to understand its individual efficiency fully. Overall, we randomised 49 participants to the trial in 27 centre-months of recruitment (one participant was randomised in error and was not included in analysis, and one participant withdrew consent 5 days after randomisation and was included in analysis). Recruitment by centre can be seen in Table 4 and numbers recruited by the recruitment route can be seen in Table 5.

Table 6 shows that 28 out of a possible 33 carers were recruited to the study. These carers were evenly spread across treatment arms and centres. Table 7 shows the reasons why a small number of carers (n = 5) refused to consent to the trial.

Protocol non-compliances

Table 8 shows protocol non-compliances reported, by site and category. All of the issues identified as non-compliances could be incorporated into site staff training for a future definitive trial.

We also reported a non-compliance relating to the inconsistency in using the CST during recruitment; this is discussed in further detail in Challenges with recruitment and data collection.

In addition to the non-compliance reports, there was one instance of unblinding occurring at a 6-month outcome visit in which a participant revealed their allocation to an outcome assessor.

Losses and exclusions after randomisation

Table 9 shows the number and percentage of participants who dropped out of the trial by treatment arm, site and length of time since stroke. Attrition is defined here as the number of participants who did not complete the primary outcome (PHQ-9) at 6 months post randomisation. (See Appendix 9 for an outline of the reasons, where given, for dropout.) A total of nine participants dropped out of the trial during the 6-month follow-up period. Of these nine participants, seven were in the intervention arm. All of the participants who dropped out of the study had their most recent stroke between 3 and 12 months previously.

Dates defining the periods of recruitment and follow-up

Primary clinical outcomes

A total of 39 out of the 48 randomised participants had valid PHQ-9 outcomes at follow-up (18 participants in the intervention arm and 21 participants in the control arm). Table 17 shows a summary of the results of the primary effectiveness analysis. Following adjustment for baseline and centre, we observed a mean difference in PHQ-9 of –3.8 points (95% CI –6.9 to –0.58 points) at 6 months post randomisation. This represents a reduction in depression in the BA arm.

Multiple imputation by chained equation was used to impute this missing outcome data; this was a sensitivity analysis that increased the sample size back up to 48 (25 in the intervention arm, 23 in the control arm). Adjusting for baseline and centre, we observed a mean difference in PHQ-9 score of –3.4 points (95% CI –7 to 0.094 points) at 6 months post randomisation. This is shown in Figure 3. Adjusted models included baseline measure and centre as a random effect (n = 39 in unadjusted and adjusted analysis, n = 48 in MI adjusted and unadjusted analysis).

FIGURE 3.

Results from unadjusted and adjusted primary analyses showing mean difference between intervention and control at 6 months.

Previous research has suggested that the minimum clinically important difference or change for the PHQ-9 is 4.78 points on the scale. 75 The 95% CI for the mean difference in PHQ-9 scores between the groups includes a difference of ≥ 5 points, suggesting that the likely treatment effect is within a clinically relevant range.

Figure 4 shows that in more participants in the intervention arm than in the control arm the PHQ-9 category was reduced (horizontal dashed lines) from baseline to follow-up. Horizontal dashed lines represent the cut-off scores for depression categories. This information is also presented in Table 18. Most participants in the intervention arm were in the ‘moderate depression’ category at baseline (n = 9), whereas participants in the control arm are distributed across three depression categories: ‘moderate’ (n = 7), ‘moderately severe’ (n = 8) and ‘severe’ (n = 6). Most of the participants in the intervention arm reduced their level of depression by one (n = 7) or two categories (n = 7) at the 6-month follow-up compared with baseline. Most participants in the control arm stayed in the same category (n = 9) or reduced by one category (n = 6).

FIGURE 4.

Line plot showing individual participant change in PHQ-9 score from baseline to follow-up, by treatment arm. (a) Control; and (b) intervention.

Table 18 shows that a greater number of participants in the intervention arm than in the control arm decreased by one or two categories.

Figure 5 shows the mean PHQ-9 score in the intervention and control arms at baseline (n = 48) and 6 months (n = 39). Despite the intervention arm having a lower mean score at baseline, the decrease in PHQ-9 score overall was greater in the intervention arm. The dashed lines represent the cut-off points for PHQ-9 depression categories and show that the mean score in the intervention arm has reduced by one depression category.

FIGURE 5.

Mean PHQ-9 score at baseline and 6-month follow-up, by randomised group by treatment arm (n = 48).

Figure 6 shows the relationship between PHQ-9 and VAMS ‘Sad’ item at baseline and 6 months post randomisation. Correlation was found to be moderate at both baseline and follow-up (r = 0.45 and r = 0.57, respectively). The level of agreement would be considered too low for VAMS ‘Sad’ item to be used to impute PHQ-9. 107

FIGURE 6.

Scatterplot showing correlation between standardised PHQ-9 and VAMS ‘Sad’ item at baseline and 6-month follow-up (n = 48 at baseline and n = 39 at 6 months). (a) Baseline (r = 0.45) and (b) 6 months (r = 0.57).

Secondary clinical outcomes

Table 19 shows the analysis of patient-reported secondary outcomes. In terms of the patient-reported secondary outcomes, most of the differences suggest that the intervention has a small positive effect. Table 19 shows the adjusted mean differences and their relative 95% CIs for the patient-reported outcomes including VAMS ‘Sad’ item, NLQ, NEADL and EQ-5D-5L. VAMS ‘Sad’ item and NLQ both suggest that the intervention has a positive effect although the 95% CIs include a difference of zero. The NEADL does not demonstrate any difference between intervention and control arms. EQ-5D-5L score is lower in the intervention arm, which suggests a small negative effect.

Table 20 shows the analysis of carer-reported secondary outcomes. CSI score was lower in the intervention group, which represents a positive effect. EQ-5D-5L Carer score was lower in the intervention arm, which suggests a small negative effect of the intervention. SADQ score showed very little difference between groups. CIs are wider because of smaller numbers of carers than participants. CIs around the adjusted mean differences for carer-reported outcomes are very wide because of low numbers of carers who completed the 6-month follow-up (n = 22).

Standardised mean differences and their CIs can be seen in Figure 7, where arrows on the y-axis indicate the direction of a positive intervention effect. Arrows on the y-axis indicate the direction of a desired effect. Outcomes have been standardised onto a 0–100 scale.

FIGURE 7.

Results from secondary analyses showing mean difference between intervention and control at 6 months, adjusted for baseline and centre.

Decision on the primary end point and sample size for a definitive trial

Delivery and receipt of the intervention

Table 22 shows the attendance at BA sessions in participants who were randomised to the intervention. Attendance of participants to therapy sessions was high. The mean number of therapy sessions received in the intervention arm was 8.1 (SD 3.4). In total, 92% of participants randomised to the intervention arm received at least two sessions, 88% received at least five sessions, 64% received at least eight sessions and 40% received at least 10 sessions. The mean number of missed sessions was 0.9 (SD 1.3), with 52% of participants missing at least one scheduled therapy session.

Site 3 had a higher rate of missed sessions, probably owing to a higher number of withdrawals. Figure 8 shows that participants who withdrew from the intervention attended a low number of sessions. There were also three participants who attended eight or more sessions and were lost to follow-up.

FIGURE 8.

Number of sessions attended by participants who were randomised to the intervention arm (n = 25).

The timing of therapy sessions was fairly regular, with sessions generally occurring weekly.

Number of missing values/incomplete cases

Missing data in the outcome measures and missing items within these outcomes are presented in Table 23. Baseline completion rate was high in all outcomes. The 6-month follow-up rate was around 80% in most outcomes. Two participants had two or fewer missing items for the PHQ-9 at 6 months post randomisation. Imputation was used on these missing items as specified in the SAP.

Challenges with the delivery of the intervention

Based on the experiences of delivery of the intervention, we have gained some important insights for a definitive trial:

• The therapists delivering the intervention identified that participants in the early days post stroke had more practical goals, such as walking and speaking, and were less concerned with their mood.

• It was difficult to find activities for participants with limited mobility.

• Getting participants to adapt their goals was challenging and one therapist felt that, at this stage, a talking therapy to deal with acceptance would be more beneficial.

• One of the therapists had experience of delivering CBT in a previous role. Therefore, it may be important to tailor therapist training for those who are inexperienced in delivering psychological therapies and those who are experienced in delivering psychological therapy but not specifically BA therapy. This would ensure that both experienced and inexperienced therapists deliver the therapy in accordance with the manual.

• This feasibility study recruited only participants who were mildly affected by aphasia and, therefore, did not provide us with knowledge of delivering the intervention to those with aphasia and cognitive impairments.

Challenges with recruitment and data collection

There were a number of issues experienced with recruitment that resulted in lower than planned recruitment rates. However, the final sample recruited was sufficient for this feasibility study, which was powered not for efficacy but to inform the design of the definitive RCT. The original recruitment target of 72 participants was based on three sites recruiting two participants per month over 12 months, giving a total of 36 recruitment months. However, owing to delays in site set-up, none of the sites was able to utilise the full 12-month recruitment period. Instead, a total of 27 recruitment months were available across the three sites.

There were three main challenges for the delays in site set-up: (1) approval of excess treatment costs (ETCs), (2) the knock-on effect on appointment of study therapists in the NHS (as advertisement of the posts was contingent on ETCs) and (3) a change to the IAPT provider close to the time that the study was due to open. Although we under-recruited slightly, this had little impact on determining the feasibility of the trial. Recruitment challenges included the fact that the therapists had a dual role, which meant they were recruiting participants in addition to delivering the intervention. This also meant that they were limited to being able to recruit only as many participants as they could deliver the intervention to at one time.

A second recruitment challenge was the inevitable competition in highly engaged clinical research centres for the resources to recruit participants. The CRN research nurses at sites were able to assist in screening participants from the hospital database and stroke wards. They issued invitation packs to participants they identified as fitting the initial eligibility criteria, but this then required further time to follow up potential participants to discuss the study after issuing the invitation packs.

The third recruitment challenge was the centres being in different settings and, therefore, encountering different problems with recruitment. Recruitment was particularly successful where the therapist was based in the hospital setting and so was able to work closely with the CRN nurse in recruiting from the hospital database and stroke wards. In contrast, the bulk of recruitment in the case of the therapist based in a community setting was gained through the outpatient, community caseload and voluntary group routes as there was limited resourcing to recruit from the hospital database. The therapist based in the IAPT service was heavily reliant on the participants identified from the hospital database and stroke ward by the CRN nurses and so was constrained by the resourcing available. For a definitive RCT, it would be beneficial to fund more research assistant/research nurse time to support recruitment.

Quality of life

The standard version of the EQ-5D-5L was used to calculate QALYs as no participants had severe aphasia. All participants were able to complete the standard version, including the five participants with moderate aphasia. At the 6-month follow-up, EQ-5D-5L utility scores were slightly higher in the control group than in the intervention group (Table 30).

As a supplementary analysis, the pairwise correlation between the different utility measures was calculated. There was a strong correlation (Pearson’s correlation coefficient 0.91) between values of the aphasia accessible version of the EQ-5D-5L and standard version. The correlation between responses to the EQ-5D-5L completed by carers on behalf of the study participant (carer proxy) and the standard version completed by the participants was moderate (Pearson’s correlation coefficient 0.59).

Costs

The cost of the intervention was estimated to be £57 per hour. 85 This was based on the intervention being provided on a one-to-one basis by a grade-5 Agenda for Change (AfC) mental health nurse, using estimates from the PSSRU. PSSRU incorporate training and clinical supervision costs in this estimate and, therefore, the training and supervision received in the BEADS study was not costed separately. Direct intervention costs were calculated by multiplying the hourly intervention cost by the average number of sessions for each individual patient.

There were two participants in the control group and seven in the intervention group with missing cost data at follow-up. One of these participants also had cost data missing at baseline. As there was the same level of missing data across resource use categories, data were imputed on the total cost level. The missing data at 6 months were imputed separately, based on whether or not the participant had achieved a ‘good response’ as measured by the PHQ-9. Analyses that incorporate imputed data are referred to as ‘full data set (with imputation)’ analyses. Analyses that do not use imputed data are referred to as ‘complete-case’ analyses.

Average costs at follow-up for the whole trial period (6 months) are shown in Table 31; these figures contain no imputed data. For all of the categories except societal costs, the average cost per patient in the intervention group was higher than in the control group (although no differences were statistically significant, see Table 32). Inpatient costs were incurred by two participants in the intervention group, compared with no participants in the control group. It is possible that these costs were chance events rather than being related to the intervention. For outpatient hospital services, two participants in the intervention group had costs of > £1000, whereas there were no such outliers in the control group. These costs were associated with psychiatry and psychology outpatient visits. It is unclear whether or not these costs were related to the intervention.

Home help was the main factor that caused the difference between the control and intervention groups in primary and community care costs. Four participants in the intervention group received home help compared with three participants in the control group. The intervention group received 295.5 hours more home help than the control group over the last 3 months of follow-up. There was one participant in the intervention group who received the equivalent of approximately one visit a day for 3 hours per day, which was more than that received by any other participant in the trial. See Appendix 14 for the distribution of costs relevant to the NHS and PSS perspective for the control and intervention groups.

The human capital approach was used to estimate societal costs. The societal component of costs consisted of work hours lost, carer time, travel expenses, private health-care costs and charity-provided services. These were added to the costs from the NHS and PSS perspective to form costs for the societal perspective. Owing to the nature of the population, there were only two participants in the control group and three participants in the intervention group either working or looking for work. In addition, three carers were reported to have lost work hours as a result of their caring responsibilities. However, data were incomplete, with some carers not recording whether or not they worked, and some carers not recording if they lost any time from work. For this reason, we allocated a cost to all carer time reported for the societal perspective analysis, whether or not it resulted in time being taken off work. Carer time accounted for 93.3% and 78.0% of societal costs for the control and intervention groups, respectively. See Appendix 15 for the distribution of costs relevant for the societal perspective for the control and intervention groups. Aggregated costs for the 6 months of the trial are presented in Table 32.

Results from the within-trial analysis

The within-trial analysis was undertaken on both the complete-case and the imputed data sets. Results are presented in Table 33. Based on the imputed data set, when including all costs relevant to the NHS and PSS [labelled ‘NHS & PSS perspective (1)’] costs were £1316 higher in the intervention group than the control group over the course of the study. To illustrate the influence of inpatient hospital stays we reanalysed the data excluding inpatient costs if the hospital stay was for > 1 night [labelled ‘NHS & PSS perspective (2)’]. In this analysis the incremental cost associated with the intervention reduced to £980. The 95% CI for both these estimates includes negative values, indicating that there was no statistically significant difference in costs between the intervention and control groups.

TABLE 33 - Results from the within-trial analysis

Incremental costs differ from those presented in Table 32 as a result of the covariate adjustment used in the SUR.

Cost perspective	Incremental costs (£)	Standard error (£)	95% CI (£)	Incremental QALYs	Standard error	95% CI
Complete case (38 participants)
NHS & PSS perspective (1)	1634	1174.0	–667.39 to 3934.63	–0.0107	0.015	–0.04 to 0.02
NHS & PSS perspective (2)	1228	951.9	–638.13 to 3093.27	–0.010	0.015	–0.04 to 0.02
Societal perspective	–6479	7727.6	–21,625.08 to 8666.45	–0.012	0.015	–0.04 to 0.02
Full data set (including imputed data; 48 participants)
NHS & PSS perspective (1)	1316	1275.15	–1191.66 to 3823.16	–0.0204	0.021	–0.06 to 0.02
NHS & PSS perspective (2)	980	1245.94	–1469.93 to 3430.61	–0.0205	0.021	–0.06 to 0.02
Societal perspective	–8281	7945.87	–23,879.93 to 7318.81	–0.0207	0.021	–0.06 to 0.02

For all analyses, incremental QALYs were negative, meaning that participants in the intervention group accrued fewer QALYs over the 6-month trial period than participants in the control group. Again, differences were not statistically significant. Therefore, based on an NHS and PSS perspective and a within-trial analysis, the intervention is expected to be dominated by the control, that is, it results in higher costs and lower QALYs than usual care.

It was estimated that there would be lower costs (by £8281 per patient) in the intervention group than in the control group, when taking a societal perspective and valuing all carer time regardless of whether it resulted in the carer taking time off work. This results in an ICER of £539,917 saved per QALY lost based on the complete-case analysis, and an ICER of £400,048 saved per QALY lost based on the imputed data analysis. The interpretation of these ICERs is different from the standard interpretation because the intervention is situated in the south-west quadrant of the cost-effectiveness plane (see Figure 12) – it results in cost savings but fewer QALYs.

Figures 9 and 10 show the point estimate of the ICER and CIs around this based on the ‘NHS & PSS perspective (2)’ analysis and the societal analysis, respectively, using the full data set with imputation. Figures 11 and 12 show the corresponding probabilities of the intervention being cost-effective at different cost-effectiveness thresholds for these two analyses.

FIGURE 9.

Confidence ellipses (controlling for baseline utility and costs) for ‘NHS & PSS perspective (2)’ analysis, full data set.

FIGURE 10.

Confidence ellipses (controlling for baseline utility and costs) for the societal perspective, full data set.

FIGURE 11.

Cost-effectiveness acceptability curve (controlling for baseline utility and costs) for ‘NHS & PSS perspective (2)’ analysis, full data set.

FIGURE 12.

Cost-effectiveness acceptability curve (controlling for baseline utility and costs) for the societal perspective, full data set.

Model inputs

Results from the model-based analysis

Deterministic sensitivity analysis

Cost-effectiveness analysis results are often more sensitive to the values of certain parameters than others. A one-way sensitivity analysis allows us to see how changing one parameter and keeping other parameters constant would influence the ICER. Owing to the relapse rate not being included in our base-case analyses (as it was not possible to calculate it directly from trial data), the base-case ICERs may be optimistic. Figure 13 demonstrates that when a relapse rate was incorporated within the model, the ICER increases above £30,000 per QALY gained for relapse rates of > 7% per month, under an NHS and PSS perspective.

FIGURE 13.

ICER vs. relapse rate per month, NHS and PSS perspective.

From a societal perspective, the intervention continues to dominate for all possible relapse rates, all else remaining equal.

Probabilistic sensitivity analysis

The results of the deterministic analysis have to be interpreted with caution as they do not account for parameter uncertainty. PSA assigns distributions to parameters and runs the model many times, each time selecting a value from the distribution of each parameter. Estimates from each run of the model are combined and averages are taken to calculate the probabilistic ICER. CEACs are estimated, which represent the proportion of simulations in which the intervention was associated with an ICER below a defined cost-effectiveness threshold. We ran the model a total of 10,000 times in a Monte Carlo simulation. The result of each simulation is presented on the cost-effectiveness plane in Figure 14, for the analysis that took an NHS and PSS perspective and incorporated imputation. The probabilistic ICER was £18,979.82 per QALY gained.

FIGURE 14.

Model-based analysis: probabilistic results on the cost-effectiveness plane, NHS and PSS perspective.

Owing to the substantial uncertainty around monthly costs associated with each health state, incremental costs for each model run varied widely, with approximately 70% of runs estimating that the control group would have lower costs, and approximately 30% estimating that the intervention group would have lower costs. CEACs, showing the probability of the intervention being cost-effective at various thresholds for the NHS and PSS perspective, are plotted in Figure 15. At a threshold of £20,000 per QALY gained, the intervention had a 51% probability of being cost-effective, and at a threshold of £30,000 the probability was 60%. Owing to a small number of model runs producing a QALY decrement for the intervention group, the probability of the intervention being cost-effective does not reach 100%, irrespective of how high the threshold is. For cost-effectiveness thresholds of > £19,000 per QALY gained, it was estimated that the intervention would be most likely to be the cost-effective treatment option.

FIGURE 15.

Cost-effectiveness acceptability curves, NHS and PSS perspective.

From the societal perspective, the probabilistic analysis estimated that the intervention was dominant – the vast majority of model runs resulted in cost-effectiveness estimates that lay in the south-east quadrant of the cost-effectiveness plane (Figure 16). The probability that the intervention was cost-effective at a threshold of £20,000 per QALY gained was 82.8% (Figure 17).

FIGURE 16.

Model-based analysis: probabilistic results on the cost-effectiveness plane, societal perspective.

FIGURE 17.

Cost-effectiveness acceptability curves, societal perspective.

Supplementary analysis

It is important to reiterate that the base-case deterministic and probabilistic results incorporated a relapse rate of zero, with no uncertainty around this. We conducted a supplementary analysis that incorporated a hypothetical relapse rate of 10% per month, with a beta distribution with a standard error of 0.2, and lower and upper bounds of 0.01 and 0.8, respectively. Clearly this is an important assumption. The standard error represents an inflation of the standard errors associated with the response rate in the experimental and control groups, which were 0.09 and 0.08, respectively. This resulted in a probabilistic ICER of £21,626.49 per QALY gained for the NHS and PSS perspective. From the societal perspective the model continued to estimate that the intervention would dominate, leading to cost savings and QALY gains. The cost-effectiveness plane scatters and the CEACs for these analyses are presented in Figures 18–21. At a cost-effectiveness threshold of £20,000 per QALY gained, the probability of the intervention being cost-effective was 42.6% under an NHS and PSS perspective, and 81.0% under a societal perspective.

FIGURE 18.

Model-based analysis: probabilistic results on the cost-effectiveness plane, NHS and PSS perspective with relapse rate.

FIGURE 19.

Cost-effectiveness acceptability curves, NHS and PSS perspective with relapse rate.

FIGURE 20.

Model-based analysis: probabilistic results on the cost-effectiveness plane, societal perspective with relapse rate.

FIGURE 21.

Cost-effectiveness acceptability curves, societal perspective with relapse rate.

Value of information

We conducted our value-of-information analysis based on the supplementary probabilistic analysis that incorporated a non-zero relapse rate, as it was considered to be unrealistic to assume a zero relapse rate with no uncertainty. We estimate that the per-patient value of perfect information is £1348.85, given a cost-effectiveness threshold of £20,000 per QALY gained and taking an NHS and PSS perspective.

We extrapolated the per-patient value-of-information estimates to a population level by estimating the number of participants that would be likely to receive the treatment over a 10-year time period. It is estimated that there are 1,202,053 stroke survivors in the UK. 109 Approximately 29% of these people will also have depression. 1 To be conservative in our estimation of the population size that could benefit from BA therapy, we assumed that the intervention would not be suitable for people with severe depression. No published estimates of the proportion of stroke survivors who had mild, moderate or severe depression were found. Hence, to estimate this we used the proportion of participants with severe depression in the BEADS trial (29.2%); thus, 70.8% of stroke survivors with depression were estimated to have mild or moderate depression and, therefore, to be eligible for the intervention. In total, we estimate that the prevalent population in the UK that could benefit from the BA intervention is 246,922 [1,202,053 × 0.29 × 0.71].

In addition, it is estimated that approximately 152,145 people in the UK have a stroke each year,110 with 67–75% of these estimated to be first (i.e. not recurrent) strokes. A total of 78% of people experiencing a stroke are expected to survive for longer than 6 months. 111 Again, approximately 29% of these people will also have depression and we estimate that 70.8% of these would have mild or moderate depression and be eligible for the intervention. To be conservative, taking the lower bound of the stroke proportion that represents a first stroke, we estimate that the incidence population in the UK that could benefit from the intervention is 16,333 per year (152,145 × 0.67 × 0.78 × 0.29 × 0.71).

In total, we estimate that 410,251 people in the UK could be eligible to receive the intervention over a 10-year time period (246,922 + 10 × 16,333), with, on average, 41,025 treated per year. Based on this, we estimate a population-level EVPI of £552M.

The EVPI was higher when undertaking the analysis from a societal perspective. The per-patient EVPI was £1881 at a cost-effectiveness threshold of £20,000 per QALY gained and the population-level EVPI was £771.8M.

Gains and changes

Therapy-specific experiences (intervention group only)

Experiences of delivering the therapy

Challenges

The therapists also discussed the challenges they encountered while delivering the therapy. Dealing with the individual differences was reported to be one of the most challenging parts of delivering the therapy. Therapists did not rigidly follow the session-by-session plan as not everyone worked at the same speed, and the physical and cognitive capabilities and personalities of people varied. Some participants were more resistant to change or sceptical, whereas others were open to take on board new ideas:

I had a couple of people where I had to take it a little slower, needed a little bit more time or a little bit more time for their writing, again that wasn’t difficult it was just something that I had to be aware of . . . I think some people are a little bit more resistant to change depending on where they are in their journey after stroke. Some people are ready to take on board anything that you want to do others are a little bit more sceptical even though they’ve decided to try it . . . So trying to get people to understand the rationale and why we were doing it, some people I had to repeat that for a lot more and really y’know try and keep them on that frame of mind.

In order to deal with the individual differences, therapists tried to be responsive to individual preferences and tailored the session structure based on individuals’ needs:

I just tried to be responsive to the individuals as I kind of do in my day-to-day therapy and I think again as I mentioned, going back to the rationale of the therapy and the protocol to try and remind them what we were trying to do and why . . . just thinking around the individual needs and exactly sort of what people were frame of mind that they were in at the time.

Two of the therapists also found the process of identifying meaningful and achievable activities difficult in some cases owing to the participants’ levels of disability and unrealistic expectations:

I‘d say the hardest things was just working round practicalities of what people could and couldn’t do now, I think people tended to still get very focussed in their initial goals of ‘I just want to be as I was in the past’ and so the hardest thing was trying to get people to look at perhaps having to adapt that and having a different goal slightly or working with that.

Keeping people engaged with the therapy-related tasks was also challenging. One therapist suggested that this was due to people not being able to come to terms with the stroke and being ready to change. They also argued that, if they were in a clinical situation, they would withdraw the intervention for such people (after a collaborative decision). However, as this was a research study they felt the need to keep pushing to engage people:

Quite a few people that I worked with it seemed like they weren’t quite ready to be increasing their activity and what they needed more at this stage in their recovery was the talking therapy and work on acceptance with kind of the stroke and the changes that have happened in their life so there was some difficulties there in terms of people’s physical capabilities and also engagement with the intervention because they kind of felt like there wasn’t anything that they could do . . . quite a few of the people I worked with were telling me that they didn’t really want to plan things out they didn’t want to plan activity and plan out their lives . . . and I think in a real life clinical situation there would have been a couple of points where I think I would have had quite a frank conversation with somebody about their level of engagement with the intervention and where we’d to go from here and if they really want to carry on.

Improvements needed/suggestions

In order to improve the therapy, two therapists suggested having a standardised interactive therapy notebook or a folder that might be easier for participants to use and might also facilitate their engagement with the tasks, as noted in the following quotation from one of the therapists:

I tended to recommend to people or suggest they consider having like I say a bit of a therapy book because we would obviously have sheets that I could copy for people and things like that and that can become a bit disparate and people end up with piles of paper all over. So one thing I’d say perhaps just to consider, and a couple of people did, was to have a therapy book, erm a notebook that they can make any questions for the session, notes at the end and set the goals in it and kind of almost do that together and then they can sort of slip their diary in it. And one thing for therapy in the future if that wanted to be quite standardised across therapists is that could always be something that came with the therapy so perhaps a bit of a workbook for the clients that BEADS produce so that all the therapists were doing that in the same way. So even if it was just a notebook with like a couple of sections or something like that or even just a folder that sheets could go in or just to standardise it if that’s something to keep quite the same across a large number of people.

Some therapists also argued that delivering the therapy 3 months post stroke was too soon as patients were still in the process of adjusting to stroke, and behavioural change was not a priority for these participants. Therefore, they suggested that delivering the therapy 9 or 12 months post stroke might be more beneficial:

I thought that maybe having the therapy 3 months post stroke was a bit too soon, just because people are still adjusting to what’s happened and having to adapt their homes . . . I felt like it wasn’t so much of a priority for them to do the BA. They were always wanting to walk or improve their speech.

Experiences of participants from therapists’ perspectives

Support provided to therapists

Experience of working on trial within therapists’ departments

Therapists had varying experiences of working on the trial within their departments. One therapist was autonomous, working completely separately from their clinical team. Apart from receiving supervision and updating the team, they mostly worked independently from home:

I think for me it was completely separate. So I had my supervision from a clinical lead who was the PI for our site, my team do something very different so apart from I kind of let people know what I was doing and sort of gave a bit of information, but generally it was just completely autonomous . . .

One therapist felt integrated within a larger clinical team within the department:

My team . . . were brilliant as well, so yeah I really enjoyed it. I had a lot of help from [names person] . . . everyone was really supportive and always replied to the e-mails and yeah I got a lot of support so it was really good . . . there was always someone there to help and give you any supervision if you needed it and . . . so there was always a wealth of knowledge to ask if ever I needed it so yeah, I couldn’t have asked for better team to be, to be part of.

In contrast, one therapist did not feel integrated within the clinical service and expressed that they were mostly working in isolation:

I was kind of working in isolation a little bit which isn’t ideal . . . so I didn’t really feel integrated into the service in anyway and the way that referrals came to me from the service it seemed I wasn’t really involved in the triage process or really that involved with the other professionals in that team.

Integrating the trial practice into wider service

Rationale of the study

Participants and carers recognised the importance of the study and there was a sense of them having an understanding of its rationale:

I think it’s useful information and the only way you can get it is by talking to people like me, that’s guide through it [sic].

Patient A6/001: control group

You can’t move forward in these situations without these projects and things.

Carer B1/087: intervention group

However, there were a few concerns about what benefit the study and filling out different questionnaires might offer participants on a personal level:

Well, since this [stroke] has happened I’ve had all sorts chucked at me. Different types of questionnaires. Varying different types of questionnaires. I thought it was good, it’s got its good points. Erm, but ultimately what do I achieve from it?

Patient C7/017: intervention group

Motivation and reason to participate

Altruism was one of the most commonly mentioned reasons for participating. Some participants joined the study as they felt grateful to have been invited. Other participants explained that they thought it was better than doing nothing and that there was nothing to lose:

I went into this open-minded and I thought, I’ve got nothing to lose, I’m only sat in the house, I’m not doing anything. If nothing else, it’s somebody to talk to.

Patient C7/001: intervention group

We said yes, you know it’s the least we can do in return for all the care and attention that [patient] had.

Carer C1/047: control group

However, it was evident that some people had specific expectations of the BEADS study. Some wanted to find out about their problem and what could be done to improve their mood. Others chose to participate to see if the treatment would work, and some saw the study as an opportunity to talk to others about their problems:

But err so I’m interested in you know trying to sort out well for ourselves trying to find out what is it frustration or is it depression that [patient]’s got. Is there anything that we can do about it? . . . But yes I am I am in a way desperate or I am keen to try and find a way that we can bring [patient] a little bit more back to what she was prior to certainly the last two strokes, if not all three strokes.

Carer C1/047: control group

Well I thought, well I’ll see what I can do and try and, try and help like, you know. And see if it help me to get me head straight a bit.

Patient A4/002: control group

Some participants reported having no expectations regarding the study when they first got involved:

I don’t think we had a thought about whether it was going to help or not. And I, and I don’t think actually, one way or the other.

Carer A1/124: control group

I didn’t know what to expect, I was just – I’ll do it, I’ll do it, I’m not doing anything else. So yeah, I thought yeah, I’ll give it a go.

Patient A6/001: control group

For others, it was the hospital or the health professionals who suggested the trial to the patient:

It was from [staff name] at [hospital name], and she came out as sort of a stroke counsellor to try and help me adjust with you know the stroke. Because it’s difficult. It’s tough when you’ve had a stroke you know. And she came out and it were [sic] her that suggested it.

Patient A4/002: control group

Understanding the research process

Participants’ views of the outcome measures

Recruitment

Each therapist experienced different issues with recruitment depending on the setting in which they were based and their links to recruiting trusts:

I think it was quite difficult in that I couldn’t get hands-on and do things. So we struggled with getting packs out and recruitment to start with . . . so having permission to go in and work with the team and the research nurses to send out packs and then contact patients and follow that up would be really helpful . . . if we could have a few extra staff as it were for the first couple of months to really plough out that invites, information contact people and then even if that dropped off later because we didn’t need that support, to kind of almost do it the other way round that would have helped.

They also found that recruitment from voluntary groups was ‘unfruitful’ as the members were already functioning well or were several years post stroke:

[we] tried a number of approaches for different groups so like the voluntary groups and things like that, which was completely unfruitful. So that didn’t really work because the groups mainly held people who were either functioning quite well because they were active in the groups or they were sort of 15 years post stroke so it was a very long time ago.

One therapist found the process challenging and experienced problems at the start, as the sites had not been set up as planned. They felt stressed when targets were not met as they believed that it was their responsibility to recruit the target number of people:

We had a little bit of an issue with recruitment at the start of the study in [names place] and other sites weren’t quite set up on according to the schedule for the study and that kind of set things back a little bit . . . the impression I had was that recruitment was and the feeling that I got was that recruitment was kind of solely my burden almost it was all my responsibility and kind of it was quite stressful at time when targets weren’t, weren’t being met because of lack of referrals . . . and I think one thing that helped towards the end of that was getting some more support from the admin team that I was kind of working with the [name of] admin team which I guess just didn’t figure out was available at the start.

When the therapist was integrated into the wider stroke research team at the trust, they reported that this went well as they had access to patient data and were able to support all recruitment activity:

I had a lot of help from the stroke co-ordinator . . . she’d given me a database to work from, send out recruitment packs which really helped to kick-start recruitment for [names site] . . . being able to access that data and managing all the recruitment myself and sending out the packs and knowing where was at with everything really helped because I wasn’t having to rely on other people to do it who had their own jobs to do as well so I think that was a real benefit for [names site] because I was able to look at the patient records and send out the invitation packs so it was really good.

Two therapists argued that working part-time limited their ability to recruit more people and slowed down the process, as they did not have sufficient time to exhaust all the available recruitment routes at the same time as delivering the therapy.

Study procedures

As it was their first research experience, two therapists had initial concerns about following the protocol, but they both felt that they were able to ask questions and received support from the trial team, which they found useful:

There was a few things that I had questions on so just to check back on, yeah as I say I’m very clinical based so a lot of the paperwork it was all new to me quite how much paperwork and forms and everything there was. So I was a bit terrified of getting something wrong, but I was, I felt able to ask questions and anytime something was slightly wrong.

Two therapists found the protocol comprehensive, easy to follow and useful as a reference guide. One therapist thought that it needed some amendment to allow for flexibility and streamlining in terms of the processes of screening, baseline assessments, recruitment and consenting:

It [protocol] seemed to work really well, it was nice having that as a reference guide to refer back to, it was very thorough and so yeah I think that worked well, I can’t think of anything to change.

There were a couple of things in terms of recruitment and the recruitment routes themselves I think there’s a couple of amendments there to be made to allow for that flexibility but I think just a little bit more flexibility being inherently more in the protocol would have been useful . . . I think the process for screening, baselines and recruitment and consent could have been streamlined a little bit I think that there were some cases I found that it was a little bit disjointed.

One therapist identified that completing therapy treatment sheets and transferring them to the online system duplicated effort and was time-consuming:

So, for most of the paperwork, like assessment paperwork, you would do it with the client ‘cause you were doing there and then and then you would transpose it onto the system when you got home which was fine because you can understand that, but the therapy treatment write-up sheets just felt like a real duplicate of time because you would write up your notes with the patient, make some notes as you were going along and then you would do a paper copy and then you would do it on the computer. So it was almost like ‘cause that was quite time-consuming.

Measures

Therapists thought that all the measures were relevant and comprehensive and fit with the purpose of the study:

Well I know all the different assessments are capturing different bits of data for various things that they’re going to be looking at. I think they’re all relevant . . . I think the carer questionnaires were particularly good, I think the carers that I worked with really liked being involved in the study and that way being invited to take part as well and have their view . . . They seemed fit for purpose, there wasn’t anywhere I was thinking that it was I didn’t understand why it was being done or that I was thinking it wouldn’t be useful there was just that one where there was a doubling up of the aphasia-friendly and the standard one.

One therapist thought that there were too many questionnaires, but felt that they were not tedious and were relevant for the purpose of the study. Another therapist did not think that there were too many measures to complete and noted that, when needed, the assessment session was split:

There was a lot of them [questionnaires], but I could see why they were there. They didn’t feel too tedious, as I say I never had anyone with great difficulties sort of aphasia or cognitive, they could have been quite overwhelming if there was, it would have taken time.

I didn’t find that it was too much. I just had one lady who did get tired halfway through so we had to split the session up . . . I just thought that that was the information that was needed to collect . . . if participants were struggling I’d read the questions out for them and do and the checking and things like that.

One therapist highlighted the doubling up of the aphasia-friendly version and standard version of the EQ-5D-5L questionnaire at baseline and thought that it would be more efficient and more accurate if people completed them independently at first and got help from the therapist if needed:

There’s one that there was a standard version and an aphasia-friendly version as part of the baseline booklet which I always kind of felt that it was a bit of a confounding variable that some of those baselines were done in person with the therapist as opposed to done by the individual themselves because a lot of the time the baselines were carried out with me going through them, me reading in which case the aphasia-friendly version and standard version it kind of, it’s a bit meaningless if it was me going through them anyway.

Feasibility trial

We randomised 49 participants to the trial (one in error) in 27 centre-months of recruitment. Recruitment was lower than the anticipated sample size of 72, although this calculation was based on 36 centre-months of recruitment. The highest number of participants was recruited through hospital databases, whereas recruiting from stroke wards was the least effective method. Nine participants (19%) dropped out of the study during the 6-month follow-up; seven of these were in the BA arm. Delivery of the intervention was good, with high attendance (90%).

Participants received a median of 9 and a mean of 8.1 BA therapy sessions (range 0–14 sessions), with sessions each lasting a mean of 57.3 minutes. Excluding the participants from the BA arm who did not receive the intervention, 75% of participants received at least eight sessions. A positive effect was found in the intervention arm for the primary clinical outcome, the PHQ-9 score. We did not test for statistically significant differences between the groups because this was a feasibility trial but we note that the 95% CI does not include zero. In terms of the secondary outcomes, the intervention had a positive effect for participants on VAMS ‘Sad’ and the NLQ and for carers on the CSI, although these results are not considered statistically significant. There was no difference between intervention and control groups on NEADL. Small negative effects were found for the patient-reported EQ-5D-5L and SADQ.

Fidelity assessment

The therapy record forms, which documented the proportion of time (in 10-minute units) spent on individual components, indicate that the components of BA described in the manual were delivered to participants. The distribution of time on the different components over the course of therapy was as expected. However, there was little use of graded tasks assignments and training in problem-solving documented. This may be because graded tasks were often used as a between-session task and so were coded as such. The video recordings of sessions were an important adjunct to the record forms as they highlighted some aspects not otherwise recorded.

Qualitative

Therapists, participants and carers observed that the most helpful aspects of therapy were identifying new and meaningful activities, reflection during the sessions, having weekly sessions and having the chance to talk with someone. Most participants felt that both the number and the duration of sessions were appropriate, although some participants suggested that follow-up sessions would help to maintain the gains made. Interestingly, some control participants also found participation in the study helpful as it provided opportunities to talk about their experiences. However, others were uncertain as to why they had been randomised to usual care.

The therapists found delivering BA a positive experience and would recommend it to other therapists because of its simplicity and usefulness. The therapists found the manual and training helpful but suggested having a summary of each session and an interactive notebook or workbook for participants. The therapists may also benefit from having a quick reference guide alongside the full treatment manual. The therapists also believed that therapy would be more appropriate for those who had come to terms with their stroke (i.e. those who had a stroke > 3 months ago). The biggest challenge identified was the variation in individual presentations, although the therapy and manual allowed sessions to be tailored for individuals’ needs. However, identifying meaningful and achievable activities was difficult for some owing to their level of disability and unrealistic expectations. The therapists reported different experiences of recruiting participants to the trial, reflecting local site differences.

Health economics

Behavioural Activation Therapy for Depression after Stroke has shown that it is feasible to collect the data to conduct a rigorous economic evaluation. Uncertainty is high and influential, such that very high values of obtaining further information are estimated, irrespective of the perspective of the analysis, but conducting a full trial would appear to represent good value for money.

Population/recruitment

The primary objective of this multicentre feasibility trial was to assess feasibility of recruitment in the study population for a definitive trial. A total of 756 participants were screened across all sites, of whom 49 were recruited, giving a recruitment rate of 1.8 participants per centre per month. Delays in set-up meant that we had a reduced number of months of recruitment as we still closed the study on the planned date. Had recruitment been open for 36 months rather than 27 months, at our recruitment rate of 1.8 participants per centre per month, we would have achieved a sample size of 62. We have identified ways to improve recruitment in future studies. The delays in site set-up have been outlined in Chapter 3 and were due to delays in receiving approval for excess treating costs, with the knock-on effect on the appointment of study therapists in the NHS and the IAPT provider changing close to the time at which the study was due to open. This suggests that a longer set-up time should be scheduled for a definitive trial, particularly given the increased number of sites needed.

The sample size for a definitive trial was calculated with two main options, one based on a difference in PHQ-9 score of 4 points between treatment and control at 6 months and the other based on a difference in PHQ-9 score of 3 points between treatment and control at 6 months. Both options represent moderate standardised effect sizes that are appropriate for a trial of this kind. If a definitive trial were to be powered on the smaller effect size to detect a smaller effect, this would mean that, to have a total sample size of 580 participants, based on a recruitment rate of 1.5 participants per centre per month, 16 centres and 24 months of recruitment would be required. However, recruiting 16 centres capable of participating and delivering the intervention, while maintaining a steady pace of recruitment over 24 months, may be a potentially significant challenge. Given the variability in psychological support services for people with stroke, it is not yet clear if 16 centres would be achievable. Data would need to be collected to explore this aspect further.

The attrition rate, at 19%, was higher than the anticipated 15% and was highest in the BA group. The reasons given for dropping out were varied and included having insufficient time (for those in the BA group) and a deterioration of health. However, people did not have to give a reason or may not have given the real reason for drop-out. The attrition rate was higher than in studies of CBT for post-stroke depression25 and motivational interviewing after stroke. 31,35 The sample size was too small to enable subgroup analysis to explore whether or not particular characteristics were associated with drop-out. One possible factor is that in BEADS we recruited people at a wider range of time points post stroke.

Recruitment varied across the three sites. Recruitment was most successful at site 1, where the therapist was based in the hospital with the CRN research nurse, which enabled close working to facilitate recruitment. This was reflected in the qualitative interviews, in which therapists noted that it was important for them to be integrated within clinical teams to facilitate recruitment. The therapists were generally positive about the trial procedures but suggested that the process of screening, baseline assessments, recruiting and consenting could be clearer in the protocol.

From our observations in the feasibility trial, the use of IAPT as a main site for recruitment in a definitive study may not be ideal, particularly if this is separate to the trust delivering wider stroke services. This is due to the nature of IAPT services, which can be actively sought out by participants, and it was possible that people would be less likely to refer potential participants owing to the risk of an unfavourable randomisation outcome. The IAPT site had lower recruitment, possibly as a result of having fewer stroke patients available to them than the other main sites. This made them heavily reliant on patients identified from the hospital database at the companion site. This adds to the challenge of recruiting the large number of sites required for a full trial.

Recruitment varied across the different routes. The highest number of participants (42.9%) was recruited using hospital databases. Although this requires a large number of potential participants to be contacted, it may be considered an efficient approach in terms of staff time to send out the invitation packs. This approach alone may not be sufficient to sustain recruitment over the duration of a definitive trial depending on the number of patients on the database and how many years previously the details are available (e.g. if recruiting up to 5 years post stroke). The highest uptake from the number of patients screened was via community caseload (13/28 participants; 46.4%), which is possibly attributable to the personal approach by the clinician who had the opportunity to explain the study and answer questions, although this is more time-consuming than using databases. We did not recruit through GP databases, and this could be explored as a future option in order to approach people who are later post stroke (e.g. 2–5 years) to reach the large patient population. We would first need to establish the likely proportion of people on the database who have had a stroke within the past 5 years and the time and resources needed to send out invitation packs. A further suggestion for recruiting more patients who are later post stroke (e.g. after 1 year) and therefore unlikely to be in contact with stroke services would be to increase self-referral. This could be attempted through advertising the study more widely in the press and through social media.

It is possible that equipoise, or lack of individual equipoise, may have affected recruitment when we were reliant on clinicians as the gatekeepers notifying potential participants about the study. Interviewing the PIs and key clinicians from the sites would have provided insight into how they perceived equipoise and to what extent it influenced how they explained the study to participants. We did get an indication, however, that there were issues of equipoise at one site early on. At this site objections were raised about the treatment-as-usual arm not receiving any active intervention, and whether this was ethical. This was resolved with the explanation that it was unknown whether or not the intervention arm would be shown to be beneficial. However, additional guidance for clinicians may help them to explain equipoise to eligible participants. 112

Therapists reported that, as a result of their workloads, they could recruit only the number of participants to whom they could deliver the therapy without undue delay (in the event that participants were randomised to the intervention arm). At one of our sites, the therapist had to delay randomising participants and this meant that there were varied periods of delay between baseline assessments and randomisation. In nine participants this delay was > 30 days. Stratification by site was used but this did not alleviate these delays. To increase recruitment rate and/or reduce this delay, therapists would potentially need to be employed for a greater proportion of the week in order to prevent long delays between consent and intervention delivery.

All of the participants with aphasia had mild aphasia, which may indicate that people with more significant communication support needs may have been missed by the identification and recruitment methods used. It is possible that staff referring to the study simply did not consider people with significant communication difficulties. In addition, the written information sent out to potential participants (covering letter, participant information sheet, reply slip, PHQ-9, VAMS ‘Sad’) may not have provided sufficient support for people with more significant aphasia to understand the study and demonstrate an interest. For a definitive trial, the presence of these difficulties could be one of the factors that help decide whether to send a pack or whether to arrange a home visit to discuss the study.

The therapists reported not using the CST if they felt that the potential participant seemed to be communicating well with them. Yet, how well a person appears to understand can be misleading and is subject to the judgement made by the therapist. If people with more significant communication difficulties are identified for a future trial, it will be important for therapists to use a tool such as the CST routinely with all participants. This would ensure consistency and identify any communication difficulties that need to be supported. It is to be noted that the FAST, which is similar in presentation to the CST, was completed well as it was part of the CRF. However, FAST indicates only the presence and severity of aphasia, whereas the CST identifies the levels of communication ability and thus eligibility for the study and also links to support strategies. These issues could be addressed during staff training in a future study.

Generalisability

The information gained from this feasibility trial was adequate to inform the sample size needed for a definitive trial. Our inclusion criteria were broad to reflect clinical practice and address the issue of the generalisability of the results. In order to recruit a sample representative of the wider stroke population, we identified participants through a range of routes. In most participants (62.5%) the time since stroke was 3 months to 1 year, although one-quarter of participants were between 1 and 2 years post stroke. We included people who had their stroke up to 5 years previously because there is evidence that depression is still common after this time113 and to reflect our original PPI feedback.

The therapists did attend voluntary stroke and aphasia groups in an effort to identify people > 5 years post stroke but found that they were less likely to need support for low mood and were often more than 5 years post stroke. For a future study there is the potential to access hospital and community databases to include more people between 2 and 5 years post stroke.

It is an important point that people with aphasia have higher rates of depression than the rest of the stroke population and it was disappointing that only mildly aphasic participants were recruited. Of the 48 participants recruited, 18 had aphasia, which reflects the incidence rates of aphasia in this population114 (it should be noted that all of these patients were mildly aphasic). This may be a reflection of the recruitment approaches used. We have already shown in a previous study17 that BA can be delivered to a sample of participants with aphasia and can improve mood. At baseline, 78.7% of participants were scored as having mild cognitive impairment on the MoCA, which is in keeping with previous findings that up to 75% of participants have cognitive impairment after stroke. 115

The team discussed the possibility that it is not uncommon for clinical staff to ‘pre-screen’ patients and not identify those who they believe might struggle with the intervention or they feel they want to protect, which is common with patients with communication and cognitive difficulties owing to their increased vulnerability. A possible way to ensure inclusion of a wider range of aphasia severity in a future trial may be to provide training to the clinical teams about ensuring that all patients identified as eligible are given the choice about whether or not to participate in the trial, irrespective of a personal clinical viewpoint. In addition, it may be of benefit to engage more fully and regularly with the speech and language therapists within each clinical team, as they are best placed to identify people with aphasia.

Half of the people (69/138) who completed the PHQ-9 for screening prior to recruitment scored as depressed. This is higher than the prevalence of one-third reported in the literature11 and suggests that our recruitment approaches were able to identify depressed people who were willing to respond to the invitation for the study. The mean PHQ-9 score of 16.8 points (SD 4.7 points) in the recruited sample at baseline falls in the category of major depression, moderately severe. In the sample at baseline, 39.6% of participants scored as having moderate depression, 31.2% were scored as having moderately severe depression and 29.2% were scored as being severely depressed. We excluded people who were receiving medical or psychological treatment for depression when they had their stroke and those currently receiving psychological intervention. Therefore, we would not expect to have many people with severe depression.

We recognise that all but one participant was from a white British ethnic background. We would therefore need to improve the representation of the multicultural, multi-ethnic UK population in a definitive trial. Approaches to improve this representation could involve the selection of a diverse range of study sites as well as the inclusion of people from different ethnic backgrounds in our PPI group.

For the purposes of this trial, only English speakers were offered the intervention as delivering the intervention to non-English speakers is not straightforward. This would need to be addressed if the intervention was shown to be effective with English speakers initially. It is important to explore methods of broadening this intervention out to non-English speakers, for example by using therapists trained in BA who speak the same language as the patient, using translators or using family members (this would need to be explored with caution as translators often do not translate verbatim and, therefore, the content of the therapy delivered may not always be consistent with that received and so careful fidelity measurement would be required).

Intervention

The feasibility of the intervention with regard to delivery was very promising as the results demonstrated that the intervention could be delivered as intended across sites and was acceptable to participants, carers and therapists. Of the 225 intervention sessions scheduled, 202 sessions were completed (i.e. 90% of sessions scheduled were delivered). Reasons for missed sessions included changes to participants’ availability (n = 14, 61%), illness (n = 4, 17%) and change of therapists’ availability (n = 3, 13%). Overall, the therapy seemed to be acceptable to participants, carers and therapists and could be delivered by an AP or IAPT therapist under supervision from an experienced mental health practitioner. The intervention was manualised to support delivery of the intervention by the therapists, to facilitate consistency across sites and therapists and to allow future replication. The therapists did comment in the interviews that the manual was quite large and that an overview summary to refer to would be helpful. They also recommended that an interactive therapy notebook/workbook might facilitate participants’ engagement with therapy. Some participants who received BA suggested that follow-up therapy sessions would be useful to maintain gains. This would be consistent with the booster sessions offered in the study comparing BA and CBT for depression in primary care. 49

The therapy record forms indicate that the components of therapy in the manual were delivered to participants and that the distribution of time was as expected. The therapy record form may need to be modified to reflect the content of between-session tasks, as the infrequent coding of problem-solving and graded tasks on the record form may be because they were covered as between-session tasks. This is supported by the fact that the therapists and participants did refer to problem-solving and graded tasks (i.e. breaking down a bigger goal or task into smaller steps), which suggests that these approaches were used.

In the qualitative interviews, therapists felt that the timing of therapy was too early for some participants as they were still adjusting to stroke, although it is important to note that this was not commented on in participant interviews. Participants had to be a minimum of 3 months post stroke but could be up to 5 years post stroke, so we did not constrain recruitment to the early months post stroke. The sample size is not sufficiently large to allow subgroup analysis of whether time post stroke was related to differences in outcomes. With regard to whether the intervention is sufficiently adaptable for patients who were perceived to come up with unrealistic goals, we could attempt to address this in a definitive trial by including more time in the therapists’ training to provide guidance on goal-setting, using examples from BEADS. We could also seek feedback from our PPI group on how to better explain the purpose of the therapy to patients and provide examples of goals that may be appropriate.

Outcomes

Follow-up for a definite trial could be done via telephone (where appropriate), which may improve follow-up rates. We observed that some participants needed to be reminded by telephone call a few times to complete and send back outcome assessment packs to the central study team; this could be practically challenging in a larger study. Postal outcome assessments supported by telephone reminders with the option of resending questionnaires, completing questionnaires over the telephone or offering a home visit have been used previously31,35 and Parker and Dewey116 highlighted that administration time was important to maximise the quantity and quality of information obtained by postal questionnaires. Follow-up home visits could be offered to all participants; however, this would significantly increase the cost of the study, particularly considering the number of centres and participants needed for a definitive trial. Text message reminders could also be used.

Completion rates of individual questionnaires were good and there are no reasons to change the outcome measures. The outcome measures were generally felt to be appropriate in terms of content and length.

The quality-of-life measures used in the BEADS trial are worthy of further consideration. The standard EQ-5D questionnaire was included, as was a proxy version completed by carers on the behalf of participants, an accessible, unvalidated, aphasia-friendly version of the EQ-5D, and the standard version completed by carers to report their own QoL. Because nearly 98% of participants were able to complete the standard version of the EQ-5D at baseline, this measure was chosen as the basis for the economic evaluation, with neither the proxy version nor the accessible version used any further. It is relevant to note, however, that data completion was slightly worse for proxy measures, and that the agreement between proxy measures and patient-completed questionnaires was only moderate. Agreement between patient-completed standard and accessible measures appeared to be much higher, although this does not represent an appropriate validation of the tool. This reinforces the view that proxy measures should be avoided when possible, hence the recognised need for accessible preference-based HRQoL measures for participants who are unable to complete the standard EQ-5D questionnaire. 117

Carer QoL was collected marginally less successfully than patient QoL in the BEADS trial. Completion rates were 84.8% at baseline and 63.6% at the 6-month follow-up time point. QoL of carers has rarely been taken into account in economic evaluations, but excluding it may underestimate the value of an intervention owing to health spill-over effects not being incorporated, as the health and well-being of a carer can be affected by the health of the person for whom they are caring. 118 Including these spill-over effects can affect the cost-effectiveness of an intervention. 119,120 There is currently no consensus on how spill-over effects should be incorporated, but a conceptual framework has been developed to suggest how they could be. 121 The BEADS trial has shown that it is feasible to collect information on the HRQoL of carers, and this could be incorporated into future analyses to help inform decision-making on the allocation of scarce health-care resources.

Owing to the small mean difference (and 95% CI) found between the intervention and control groups at follow-up on the NEADL, this measure could be omitted from a full trial. Overall, the items on the NEADL may not be sensitive to detect the changes in the types of activities addressed in BA. In the qualitative interviews the therapists commented that some participants showed an increased confidence during the therapy. One participant also suggested that questions about fear and confidence were missing from the measures they completed. A measure of confidence after stroke has recently been published122 and may be suitable as an outcome measure in a full trial, although, as a new measure, its responsiveness to change has not yet been determined.

Adverse event reporting

Trials of non-pharmacological health interventions vary widely in their approach to recording AEs and making attributions in terms of causality. We recommend that the approach taken in any future trial is appropriate to the level of risk and feasible. Given appropriate resource, we would record AEs but an option would be to avoid labelling them as ‘adverse’, specifically to avoid the need to assess causality. For a definitive trial we would recommend that non-serious AEs are not assessed for relatedness to stroke or the intervention as these are difficult to attribute to the intervention. As hospital admission details are not easily accessible for rehabilitation trials and an AE cannot always be assessed for its relatedness to the intervention or condition, we would propose that events are not recorded as AEs but rather as safety outcomes, which are not assessed in the same way and for which no action is taken. These safety outcomes could then be reported regularly to the DMEC for monitoring. We would record and report SAEs.

The feasibility trial

Behavioural Activation Therapy for Depression after Stroke was a multicentre RCT. Allocation concealment was ensured using internet-based randomisation hosted by Sheffield Clinical Trials Unit. The outcomes assessors were blind. It was not possible to blind participants or therapists to group allocation. The SAP was agreed prior to data lock. In addition to the feasibility RCT, we had nested qualitative research, fidelity assessment and a health economics evaluation. As outlined previously, delays in site set-up meant that the start of recruitment was delayed. However, we kept the original recruitment end date and so were able to deliver the trial, intervention and follow-up assessments in the time period originally agreed, albeit with reduced numbers. Issues with recruitment and recommendations for improvements are detailed in Chapter 3, Challenges with recruitment and data collection.

Participants were willing to be randomised, although in the interviews some control group participants seemed to misunderstand why they were randomised to usual care (e.g. they thought that they did not qualify for the intervention). This suggests that the purpose of randomisation and result of the allocation needs to be explained more clearly. We were able to recruit a diverse sample in terms of demographic and clinical characteristics.

The control arm in this trial was treatment as usual; therefore, there was no active control arm or any attempt to control for attention. We decided against including an attention control group, as there is already evidence that the benefit from BA to treat depression in other settings is a specific effect of treatment rather than a non-specific effect of extra attention. In a trial of CBT for post-stroke depression25 it was difficult to deliver an attention placebo intervention that was credible to participants, easy to facilitate and included none of the active intervention strategies. Further consideration should be given to including an attention control third arm, but the benefits of this would have to be weighed up against the difficulties associated with the associated increase in sample size that this would require.

As noted recently by Popp and Schneider,123 there is no ‘gold standard’ for attention placebo controls in trials of psychosocial interventions. Our main argument against including an attention control group is that there is already evidence for the effectiveness of BA for treating depression in other settings. In particular, in a recently updated meta-analysis of RCTs of BA for depression,124 6 of the 25 studies comparing BA to controls used treatment as usual for the control, 15 used a waitlist control, 3 used a psychological placebo intervention and 1 study had a waitlist and a placebo as controls. The meta-analysis found that BA was superior to controls. A smaller effect size was found in studies using a placebo intervention compared with a waitlist or usual care, but these effect sizes were still statistically significant. There is therefore up-to-date evidence that BA is an effective treatment for depression in the general population.

In the qualitative process evaluation from the COBRA trial (non-inferiority trial comparing CBT with BA for depression),49 although common factors, such as being listened to and feeling understood, were valued by participants, the researchers found that specific aspects associated with the CBT and BA were important in addition to these factors. In a definitive trial evaluating BA for treating post-stroke depression, it would therefore be important to use a process evaluation to focus on the mechanisms of change that may be taking place within BA. For example, in an ongoing cluster RCT of BA by mental health nurses for treating late-life depression in primary care, BA is being compared with treatment as usual rather than an attention control condition and one of the research questions is to explore the mechanisms of change that account for the effectiveness of BA compared with treatment as usual. 125

We do not have detailed information on what usual care was at each of the sites. It was intended that treatment-as-usual data would be gathered from the resource use inventory but, as this questionnaire covered all services received, whether treatment was as usual, was sought by participants or paid for, etc., it does not provide the specific usual-care information that would be valuable to collect in a definitive trial.

Regarding the disappointment expressed by some participants who were randomised to the control group, it is worth noting that the attrition rate was lower in the usual-care group than in the BA group. This suggests that allocation to the usual-care group did not deter people from remaining as participants in the study and completing the outcome questionnaires. To alleviate the disappointment for participants allocated to the usual-care group, we would recommend that the training provided to staff who are explaining the study and taking consent should include additional time spent explaining the concept of randomisation in lay terms. In addition, we would recommend that the training includes additional time spent on how to explain the randomisation outcome to participants, in particular to those who have been allocated to the usual-care group (and who will therefore not be receiving the BA therapy). The materials for this training (such as example scripts and role-play tasks) would be developed in collaboration with the PPI group.

Fidelity assessment

We assessed the fidelity of the delivery of the intervention through therapy record forms for every session and video recordings for a sample of sessions. The video analysis was limited as recordings were incomplete and did not include all sites. We had intended to select participants and sessions iteratively using purposive sampling but this was not possible because of practicalities and resources. We had only one video camera for the study; this meant that each therapist had the camera for only a short period of time and they could record available sessions only for those patients who consented to this. Technical issues such as the camera battery running low meant that some sessions were only partially recorded. Therefore, we cannot be sure that all sessions at all sites were delivered in accordance with the manual; in a future study it would be logical to review videos during the trial and ensure sufficient funding for all sites to have a video camera throughout the study. However, in addition to the training and the therapy manual, weekly clinical supervision by a local clinical psychologist and monthly teleconferences with the chief investigator and a NHS consultant clinical psychologist were provided. These meetings did not raise any concerns suggesting that the therapists were not following the manual.

We did not assess the competencies of the therapists in the study (although appropriate clinical supervision was provided). To our knowledge, a validated competency assessment for BA is not currently available. We note that Ekers et al. 62 developed a fidelity checklist for use in their trial evaluating BA (delivered by generic mental health workers) for treating depression and there may be scope for this to inform training and monitoring of therapy in a definitive trial.

Health economics

The BEADS trial has shown that it is feasible to collect the data to conduct a rigorous economic evaluation – including data to help inform an analysis taking a societal perspective, including indirect costs, and potentially including spill-over HRQoL effects. The collection of data required for the economic evaluation was successful and it would be feasible to collect similar data in a full trial. However, it is important to note that participants reported some difficulty answering the resource use questionnaire owing to issues with recall; this should be carefully considered when designing the data collection approach in a definitive trial. In addition, given the importance of carer costs, consideration should be given to collecting information on carer time over a longer time period, to provide more reliable information for a societal analysis. Resource use diaries represent an alternative approach, but also have limitations.

Our analyses suggest that there would be a very high value to obtaining further information on key parameters within the economic model – this value is likely to far exceed the cost of running a full trial. It would be most valuable to obtain more information on the NHS, PSS and societal costs associated with the health states included in the economic model. Often it is difficult to collect such information with precision even in definitive trials, but there remains substantial value to obtaining further information on other parameters such as the relapse rate, utilities and response rates, which would be best achieved through a RCT.

It was not our aim to provide definitive estimates of the cost-effectiveness of the BA treatment. However, our preliminary analysis of the cost-effectiveness of the intervention suggests that it may represent a dominant treatment strategy (i.e. cost saving and QALY gaining) from a societal perspective, but is of borderline cost-effectiveness from a NHS and PSS perspective. However, these results should be interpreted with some caution, particularly because our within-trial analysis resulted in QALY losses for the BA treatment. Owing to the higher response rate observed in the intervention group and the assumed higher utility score associated with good response, our modelled analysis provided an opposite result, but there is clearly considerable uncertainty around this.

The qualitative research

The nested qualitative research has helped our interpretation of the feasibility results. As well as interviewing participants who had received the intervention, we interviewed carers and also participants who were randomised to the control group in order to gain their perspectives. Participants were selected purposively to represent the centres, range of severity of depression, and stroke survivors with and without aphasia.

Essential/high priority

• Recruit through stroke services rather than using IAPT as the main site for recruitment.

• Provide at-site support or central monitoring of recruitment.

• Although there were monthly teleconferences between the CI and the therapists, and regular newsletters to update the sites of key changes and phases of the study, for a definitive trial, regular teleconferences could be arranged for the site staff and PIs so that they feel more engaged. This may improve recruitment at sites.

• Send out regular newsletters to all participants informing them of the study’s progress to help them feel more engaged in the study. This may increase retention as some participants in the control arm commented that they did not think they needed to complete the follow-up assessment that they were sent as they did not realise that they were still involved in the study.

• Amend the therapy record form so that the content of the between-session task is recorded, as some components of therapy (e.g. graded tasks) may be covered as a between-session task.

• Ensure that study staff and CRN staff resourcing is allocated accordingly for the most effective recruitment routes (e.g. hospital databases require a large number of letters to be sent out but this is efficient in terms of sending out a mailing to potential participants).

Moderate priority

• The therapists were in post only until the end of recruitment and so the study manager had to co-ordinate the reporting of AEs and SAEs, including finding information on the causality, during the follow-up period. It would be worth considering funding an administrator at the sites during the follow-up period. This would allow the sites to be more involved in follow-up and in the reporting of AEs and SAEs and would ensure that related activities, such as dealing with database discrepancies, study completion forms, etc., were dealt with. Having at-site support for the full duration of the study would also help with collating information about treatment as usual at the sites and any other issues that arise after the end of the intervention.

• Explore GP databases and social media as other sources to recruit participants, particularly those no longer receiving rehabilitation.

• Ensure that data on NHS, PSS and societal costs are captured.

• Ensure that sufficient data are collected to allow estimation of a relapse rate.

• Improve the collection of usual-care data. For example, study leads at sites could contact participants regularly (e.g. every 2 to 3 months) to find out whether they have had any other intervention and follow up further details with staff who delivered other interventions.

• Ensure that therapists have the capacity to start treatment sessions before randomising participants.

• Consider including a booster session(s) to support the maintenance of therapy gains.

• Consider providing a quick reference guide for the therapists to use alongside the full treatment manual.

• Ensure that the training for researchers includes additional time spent on how to explain the randomisation outcome to participants, in particular to those who have been allocated to the usual-care group (and who will therefore not be receiving the BA therapy). The materials for this training (e.g. example scripts and role-play tasks) should be developed in collaboration with the PPI group.

• Develop a fidelity checklist to be used to inform the training of therapists and the monitoring of the videos of therapy sessions during the trial.