Pilot of a randomised controlled trial of the selective serotonin reuptake inhibitor sertraline versus cognitive behavioural therapy for anxiety symptoms in people with generalised anxiety disorder who have failed to respond to low-intensity psychological treatments as defined by the National Institute for Health and Care Excellence guidelines

Sertraline

The NICE Guidelines Advisory Group proposed sertraline as a first-choice pharmacological treatment, although this agent does not have a marketing authorisation for GAD and there are relatively few randomised trials (only two trials with 706 patients in total19,20). 9 Nevertheless, in terms of risk of discontinuation because of adverse effects, sertraline was the best tolerated antidepressant and was available as a generic brand, making it the most cost-effective choice. Duloxetine (a SNRI) had a greater probability of producing clinical response in a network meta-analysis, but this is not commonly prescribed in UK primary care. The SSRIs paroxetine and escitalopram both have marketing authorisation for GAD, and there is little pharmacological difference between them and the SSRI sertraline. However, paroxetine has a more marked withdrawal syndrome than sertraline21 and escitalopram was still on patent and significantly more expensive at the time of submitting the proposal. There are also more concerns about it extending the QT interval, although it is recognised that sertraline can do this in vulnerable cases. 22 In the two sertraline studies in GAD, sertraline was dosed flexibly between 50 and 150 mg daily (mean dose at the end of treatment about 90 mg). In one study sertraline was started at 25 mg daily for 1 week to improve tolerance early in therapy,23 and this acclimatisation period is also recommended by the manufacturer in the licensed use of sertraline in post-traumatic stress disorder, social anxiety disorder and panic disorder. 24

Cognitive behavioural therapy

There are a number of cognitive behavioural models of GAD. Examples include the cognitive avoidance model,25 the metacognitive model26 and the emotion dysregulation model. 27 Dugas and Koerner28 have also developed a model of GAD known as the intolerance of uncertainty model. Stated simply, the model proposes that negative beliefs about uncertainty (or intolerance of uncertainty) lead to difficulty dealing with real or imagined uncertainty-inducing situations, which can then lead to excessive worry and GAD. Research has shown a consistent and robust relationship between intolerance of uncertainty and GAD; for example, their relationship is not accounted for by shared variance with other anxiety disorders, mood disorders or negative affect. 29,30 Data also suggest that intolerance of uncertainty is a causal risk factor for high levels of worry and GAD; for example, changes in intolerance of uncertainty precede changes in worry over the course of treatment,31 and the experimental manipulation of intolerance of uncertainty leads to corresponding changes in worry and monitoring behaviour. 32,33 Thus, data from correlational, longitudinal and experimental studies suggest that intolerance of uncertainty plays a key role in GAD. The Dugas and Koerner model28 is one of three CBT protocols for GAD, which guides IAPT services in how to carry out CBT effectively and in line with best practice. The treatment aims to help affected individuals develop beliefs about uncertainty that are less negative, rigid and pervasive. This is accomplished with the use of treatment strategies (such as behavioural exposure to uncertainty, problem-solving training and imaginal exposure) that aim to help patients confront uncertainty-inducing thoughts and situations. The treatment has been tested in four published randomised clinical trials, with results showing that it is more efficacious than a waiting list control,34,35 supportive therapy36 and applied relaxation. 37 The findings also show that 60–77% of patients attain GAD remission and that 50–55% achieve high-end state functioning following the treatment. The CBT protocol developed by Dugas and Robichaud (i.e. based on the intolerance of uncertainty model of GAD) was therefore used in this trial. 38

Internal pilot (first 12 months of the recruitment period)

1. To test and refine the recruitment methods for the main trial.

2. To ascertain recruitment rates across sites and the acceptability of the overall recruitment process.

3. To examine the extent of comorbidity between GAD, depression and other anxiety disorders in the population referred into the study.

4. To ensure that the intervention can be delivered in accordance with the protocol in both arms, with satisfactory delivery of training and monitoring procedures.

5. To monitor and assess follow-up rates of the completed primary outcome measure (HADS-A) at 3 and 6 months within the pilot trial.

Overall trial (whole 24 months of the recruitment period including the internal pilot)

1. To recruit sufficient eligible patients with a Diagnostic and Statistical Manual of Mental Disorders-Fourth Edition (DSM-IV)40 diagnosis of GAD willing to participate.

2. To compare the effect of high-quality, reproducible pharmacological and psychological interventions delivered in accordance with clear criteria and evidence-based guidelines. We asked participating GPs to follow established clinical guidelines for delivery of the pharmacological intervention, and the psychological intervention was manualised and quality controlled.

3. To obtain high rates of follow-up data on a minimum of 80% of those recruited into this trial at 12 months in order to provide a definitive answer to the research question and assess the longer-term outcomes of both interventions.

4. To analyse the results in accordance with the Consolidated Standards of Reporting Trials (CONSORT) and Consolidated Health Economic Evaluation Reporting Standards (CHEERS) guidelines.

5. To disseminate the outcomes to the NHS, academic colleagues, relevant service user groups and the wider community.

Setting

The study was community based and linked with local IAPT services. 10 We aimed to work with five recruitment sites in southern England during the pilot phase and up to 15 sites across the whole of England in the full trial.

Recruitment of participants

People who had not responded to step 2 low-intensity psychological interventions for anxiety or depression who were being considered for step 3 interventions within their local IAPT services were considered eligible (Box 1 contains a list of step 2 low-intensity interventions meeting the inclusion criteria). Identification was by low-intensity IAPT workers, who routinely administer the GAD-7 anxiety measure41 and Patient Health Questionnaire-9 (PHQ-9) depression measure,42 reviewing patients. Those patients scoring ≥ 10 on the GAD-7 were given brief details about the trial and if they were interested in finding out more, and possibly taking part, their permission was sought for contact by the research team. If they were unsure about their interest at this stage they were given a brief flyer about the aims of the trial and invited to contact the research team for more information or to discuss things further.

The central research team (trial manager/research assistant) were faxed the details of those who had agreed to be contacted by the research team and aimed to respond within 1 week (preferably by telephone or e-mail or, if not, by letter) offering them an appointment at the IAPT premises or their own home, whichever was preferred. A full study information sheet was sent to potential participants at this stage (see Appendix 1) and the patient’s GP was contacted, with their permission, and asked to complete a Medical Suitability Review form to check that there were no known medical contraindications to them being prescribed sertraline if they were randomised to that intervention arm in the trial (see Appendix 2).

Recruitment appointment/baseline assessment

The baseline interviews and assessments were conducted by a member of the central or local research team (research assistant or clinical studies officer). They checked, at the outset, that potential participants had read and understood the study information leaflet, and answered any queries. They also checked that the patient understood the reasons for the study and confirmed at that point that they were interested in taking part on the understanding that, if randomised to the drug arm, they would receive the SSRI sertraline that, although proposed for use outwith a marketing authorisation for GAD, was recommended by NICE on the basis of its clinical effectiveness and cost-effectiveness in randomised trials. They also ensured prior to the appointment that there were no medical contraindications to receiving the medication sertraline recorded on the Medical Suitability Review form as returned by the GP (see Recruitment of participants, above).

Those agreeing to take part were asked to give fully informed consent before undergoing the eligibility check and baseline assessment. It was expected that most patients would be willing to consent to the study at the baseline visit, but if they were unsure they could be rescheduled for consent and baseline assessment at a later date. The assessment was conducted at the IAPT site or GP surgery in accordance with patient preference. Written consent was obtained by a delegated and appropriately trained member of staff. No clinical trial procedures, including confirmation of eligibility, were conducted prior to taking consent. A copy of the consent form was given to the patient, the original retained in the investigator site file and a further copy sent to the patient’s GP for their medical notes (see Appendix 3).

If potential participants were happy to proceed, then the inclusion and exclusion criteria were checked (see Inclusion criteria and Exclusion criteria), which included administering a pregnancy test to females of child-bearing potential. In assessing whether or not the potential participant fulfilled the DSM-IV criteria for GAD, the relevant sections of the Mini International Neuropsychiatric Interview (MINI) questionnaire43 were administered by the research team member (depression, panic, social anxiety, alcohol and substance misuse, and GAD). If the DSM-IV criteria for GAD were fulfilled, potential participants were asked to confirm whether their GAD or worry symptoms were more severe and of more concern to them than any symptoms that they might have associated with psychological comorbidities, such as depression and other anxiety disorders, and that this was an important problem for them that they wanted to address.

If they fulfilled this criterion and all the other eligibility criteria, the researcher then administered the HAM-A questionnaire12 and asked the participant to complete the primary and secondary outcome measures at baseline, aiming for 100% completion of measures at baseline (see Outcome measures).

Pharmacological intervention: sertraline

The medication sertraline was prescribed by the patient’s GP in accordance with recognised clinical guidelines. The GPs were asked to review these patients regularly (at least six times in 12 months) and patients were to take the medication for 1 year unless they had significant adverse effects. The GPs were given details of the suggested timing and content of each of these appointments with the trial participants (see Appendix 6).

The GP was to act in the best interests of the participant at all times, so was free to refer them to secondary care services or psychological treatments if indicated. We explained that we would prefer patients in the SSRI arm not to be referred to or receive CBT while in the trial, but appreciated this might occasionally happen and should be documented in their GP notes.

If the patient made any interim visits to the surgery to discuss their treatment for GAD or issues to do with the medication being received, we asked for these to be clearly documented in their notes (we costed for up to four additional visits per GP to cover this possibility).

The GP was asked to record any adverse events and both the participants and their GPs were asked to report any serious adverse events (SAEs) or suspected unexpected serious adverse reactions (SUSARs) to the trial team (see Appendix 7).

Psychological intervention: cognitive behavioural therapy

This was delivered by high-intensity therapists from local IAPT services who were trained to deliver 14 (± 2) 50-minute sessions of a manualised treatment developed for use in GAD.

This covered six treatment modules:

1. Psychoeducation and worry awareness training – the first few sessions of treatment are devoted to psychoeducation in which patients begin to monitor their worrying on a day-to-day basis, and learn to distinguish between worries about current problems and worries about hypothetical situations.

2. Re-evaluation of the usefulness of worry – patients identify and re-evaluate their positive beliefs about worry using strategies such as role-play and hypothesis testing. Patients are helped recognise that their beliefs about the usefulness of worry are interpretations and not facts, and begin the process of ‘imagining a life without worry’.

3. Uncertainty recognition and behavioural exposure – participants learn that intolerance of uncertainty contributes to worry and anxiety, and that uncertainty-inducing situations are largely unavoidable. They then learn to seek out and experience uncertainty-inducing situations.

4. Problem-solving training – for worries about current problems, participants learn to use a problem-solving procedure targeting problem orientation, problem definition and goal formulation, generation of alternative solutions, decision-making, and solution implementation and verification. 44

5. Written exposure – in the field of health psychology, a method known as written emotional disclosure has been shown to lead to positive health outcomes. 45 Written exposure sessions are continued until writing about the feared outcome no longer provokes anxiety (typically 8–10 exposure sessions).

6. Relapse prevention – the final component is relapse prevention, the aim of which is to consolidate the attitudes, beliefs and skills acquired during therapy. Patients are encouraged to continue practising their new skills and prepare for stressors that may arise.

Sessions were to be digitally recorded and a random 10% to be assessed for quality (fidelity to the manual and therapist competence) by an independent external assessor according to prespecified criteria (see Chapter 6 for further details).

Usual care by general practitioner

Randomisation was between sertraline prescribed by the participant’s GP and CBT provided within an IAPT service, and both interventions were to be in addition to any other usual care provided by the GP. As always, the patients could be offered other medication or psychotherapy as part of their usual care, although we encouraged the GPs not to change the patient’s medication unless clinically indicated or requested by the patient, and not to refer them for CBT while in the sertraline arm if possible. Usual practice would be to allow the patient with GAD to choose, with the help of their GP, between a SSRI and CBT if they met the criteria for a step 3 intervention and if neither was contraindicated.

Patients in the CBT arm were likely to receive their CBT treatment more quickly than is usual in most NHS settings, and it was a psychological intervention specifically developed for people with GAD that is not current UK practice. NHS waiting lists mean that patients in the SSRI arm who then asked to be referred for CBT were likely to experience significant delays in receiving this. We proposed to record and measure all use of antidepressants and other forms of counselling or psychotherapy, whether NHS or private, and to take account of these in the analysis.

Primary outcome

The primary outcome was the HADS-A measured at 12 months. This is the 7-item anxiety component of the Hospital Anxiety and Depression Scale, a very widely used 14-item scale that can be self-administered. It has high validity and reliability, and the anxiety and depression components have been assessed separately as primary outcomes. 39

The primary outcome measure was initially the GAD-7,41 but this was changed to the HADS-A about 6 months after the study had started and before the recruitment of any trial participants had begun. This was done because we had originally understood, when selecting the GAD-7 as our primary outcome measure for both intervention arms, that it would be possible to ask participants seeing IAPT high-intensity therapists for treatment in the CBT intervention arm not to complete the GAD-7 questionnaire at every CBT session, which is IAPT’s current usual practice. Unfortunately, it was not possible to negotiate this in all the pilot study areas. As a result of this we were concerned that using the GAD-7 in the trial as the primary variable when this is associated with treatment in one of the groups (i.e. the CBT group) was a source of potential bias, and we therefore decided to change the primary outcome measure to the HADS-A.

Secondary outcomes

These were all self-completed measures to be collected by postal questionnaire, apart from the researcher-administered MINI and health service outcomes collected from patient notes.

• HADS-A: the HADS-A was collected at baseline and then as a secondary outcome measure at 3, 6 and 9 months. 39

• HAM-A: this is a 14-item observer-rated anxiety scale, which has been widely used, particularly in pharmacological studies. 12 It was to be administered by a member of the research team at baseline and at the 12-month follow-up.

• GAD-7: a 7-item self-completion questionnaire with very good sensitivity (89%) and specificity (82%) for GAD. 41 It is one of the core measures regularly administered by the IAPT services. 10 It was to be collected at baseline, and at 6 and 12 months.

• PHQ-9: this is a 9-item self-rate scale widely used to monitor the severity of depression. 41 It was to be collected every 3 months for the 12-month duration of the study, along with the HADS-A and EuroQol-5 Dimensions, three-level version (EQ-5D-3L).

• Work and Social Adjustment Scale (WSAS): this is a 5-item self-completion questionnaire that we planned to use to assess participants’ difficulties with physical and social functioning. 46 It was to be collected at baseline and at the 12-month follow-up.

• EQ-5D-3L: a 5-item self-completion measure used to assess quality of life and calculate utility scores for quality-adjusted life-years (QALYs). 47 It was to be collected every 3 months for the 12-month duration of the study, along with the HADS-A and PHQ-9.

• Employment and Social Care questionnaire (ESC): relevant data on services used and productivity losses were to be collected using this modified version of the Client Service Receipt Inventory48 at baseline, and at the 6- and 12-month follow-up.

• Patient acceptability measure: we planned to use the Client Satisfaction Questionnaire (CSQ), a brief 8-item self-completion questionnaire administered at 3 and 12 months. 49

• Patient preference rating scale: we planned to use a Likert scale used by our team in other studies, also administered at baseline and at the 12-month follow-up.

• MINI:43 it was an addition to the original protocol to also administer the MINI questionnaire at the 12-month follow-up to assess the depression, panic, social anxiety and GAD components, with the intention of establishing whether or not the participant met the criteria at follow-up for DSM-IV caseness for GAD or any of the common psychological comorbidities.

• Health service outcomes: we planned to collect health service use data from both intervention arms at baseline, capturing health service use for the preceding 6 months, and again at 12-month follow-up for the following items of health service use during the preceding 12 months: total GP consultations as well as those coded for GAD, psychotropic drug prescriptions from the GP surgeries, and secondary care attendances including mental health and psychological services. The IAPT sites agreed to inform the research team of the attendance rates for CBT sessions attended by trial participants in the CBT intervention arm.

• Serious Adverse Events Monitoring Form: we planned to use the standard SAE template provided by the PRIMENT CTU, modified for use in this study.

Final assessment at 12 months

A member of the research team would have administered the HAM-A face to face to all participants at 12 months and encouraged them to complete the 12-month outcome measures at this point to ensure optimum data collection. This would have involved a different member of the research team from the original assessor, who would have been blind to the participant’s trial allocation. As a check on this, they would have been asked to say which trial arm each participant they assessed had been randomised into.

Summary of baseline data and flow of patients

We planned to follow the CONSORT guidelines in reporting and analysing our data. This included presenting a table of summary statistics for those secondary outcome variables collected at baseline showing clinical characteristics for each group along with (baseline) demographic characteristics. We also planned to create a flow chart that would provide the number of potential participants who were screened, eligible, randomised and followed up at each time point.

Primary outcome analysis

The principal outcome variable was the change from baseline to 12 months in the HADS-A score, which we aimed to compare between treatment groups in a regression model that accounted for baseline scores. Tests and confidence intervals for treatment effects would be based on the normal distribution – an assumption justified by the central limit theorem. Given that there is a single primary outcome, no corrections for multiple comparisons would be required for the statistical inference. The principal analyses would have been conducted according to a prespecified statistical analysis plan to be finalised before database lock. The principal analyses would have been conducted according to the intention-to-treat principle using generalised mixed models. The primary analysis would have used a generalised mixed model accounting for clustering of therapist effects, investigational sites (both as random effects) and a limited number of prespecified patient-level factors, including baseline HADS-A score. The principal analyses would have been based on available data, and supportive analyses would examine the extent to which the principal analyses are robust to the challenge presented by the observed loss to follow-up. Exploratory analyses would have been carried out to describe how patient preferences along with a limited number of other prespecified characteristics of participants may modify treatment effects. 50 Any subgroup analyses conducted would also have been regarded as exploratory.

Secondary outcome analysis

The secondary outcome variables would have been analysed using the same (generalised mixed model) framework as for the primary outcome variable. However, the presentation of the results would have been restricted to the confidence intervals that come out of the analysis, rather than the p-values.

Economic evaluation

We planned to calculate the net monetary benefit (NMB) of CBT compared with sertraline for patients with persistent GAD who had not improved with step 2 low-intensity psychological interventions. A higher NMB indicates greater relative cost-effectiveness. Health- and social-care resource use would have been collected for both interventions over the 12-month duration of the trial using patient GP records, and patients asked to complete a significantly reduced version of the Client Service Receipt Inventory (the Employment and Social Care questionnaire) at baseline and 12 months. The health service resource use data collected would have focused mostly on primary care and psychological therapies. Details of secondary care and mental health resource use would also have been collected. Health-care resource use for the preceding 6 months would have been collected at baseline for adjustment purposes only. Resource use would be multiplied by costs from nationally published sources and summed to calculate the total cost per patient. The health-care resource use associated with the interventions would have been captured in each arm as follows: the cost of sertraline and any follow-up, training or monitoring costs; the cost of CBT based on the number of sessions attended per patient, session duration, the staff type and grade delivering the CBT; and training and any overhead costs.

The mean cost per patient for patients in the sertraline and CBT groups would have been calculated and confidence intervals reported, calculated using non-parametric bootstrapping with replacement and adjusting for baseline service use. The mean QALYs per patient would have been calculated from the EQ-5D-3L51 and the UK algorithm for calculating utility scores. 52 The EQ-5D-3L would have been collected at baseline, 3, 6, 9 and 12 months to allow calculation of the area under the curve over the 12-month trial duration for the SSRI and CBT groups, adjusting for baseline differences. The NMB of both interventions would have been calculated for a range of values of willingness to pay for a QALY. Confidence intervals would have been constructed using non-parametric bootstrapping. A cost-effectiveness acceptability curve would have been used to report the probability that each intervention has the higher NMB for a range of values of willingness to pay for a QALY. One-, two- and multiway sensitivity analyses would have been conducted for any assumptions made. Missing data and clustering would have been handled as specified in the statistical analysis plan.

Sensitivity and other planned analyses

The principal analyses would have been conducted according to a prespecified statistical analysis plan, to be finalised before database lock. The principal analyses would have been based on available data and supportive analyses would have examined the extent to which the principal analyses were robust to the challenge presented by the observed loss to follow-up. Exploratory analyses would have been carried out to describe how patient preferences, along with a limited number of other prespecified characteristics of participants, might modify treatment effects.

Training of high-intensity cognitive behavioural therapists

A 2-day training session in London on 22 and 23 January 2015 was arranged for the IAPT high-intensity CBT therapists and their supervisors from each of the pilot sites, and was delivered by Professor Michel Dugas who came over from Canada to deliver the training (see Appendix 9 and Chapter 6).

Liaison with and training of low-intensity psychological well-being practitioners

In preparation for each pilot site opening for recruitment, training sessions were held with the IAPT psychological well-being practitioners (PWPs) at each site. Their purpose was to prepare the PWPs for their part in identifying suitable participants for the study (see Appendix 10).

Training of research staff to conduct informed consent, eligibility assessment and baseline measures

Two half-day training sessions were conducted for any member of the research staff who might be involved in the baseline recruitment process – this included a combination of the research staff based at the UCL central trial office and lead PWPs or clinical studies officers involved in these procedures at the pilot sites. The first session was held on 5 May 2015 in London and included the trial co-ordinator, Dr Anastasia Kalpakidou, the Bristol Clinical Studies Officer, Joy Farrimond, and lead PWPs from Camden, Islington and Kingston (Tarun Limbachaya, Annie Ormond, Elliott Rose, Rachel Lawrence and Natalie Gunn).

The second training session was held on 16 July 2015 in Nuneaton near Coventry and included the trial co-ordinator, Dr Anastasia Kalpakidou, research assistant Sally Gascoine, ToSCA intern Alessandro Bosco and the Coventry and Warwickshire lead PWP and clinical studies officers (Helen Fletcher, James Tucker, Abayomi Shomoyei and Emily Benson).

Current good clinical practice accreditation was a condition of having a research staff role on the trial (see Appendix 11).

The central research team produced a full training/recruitment manual for participating sites with all the relevant questionnaires and outcome measures as appendices for reference. This was distributed in draft version to participants at the training sessions and the full electronic version sent to the pilot sites when they were open to recruitment (manual available on request).

ToSCA video

The North Central London Research Consortium, which supports primary care and mental health research in north central London where the central research site at UCL is located, also funded the production of a promotional video about ToSCA for GPs to watch. This included some educational material about GAD as well as a brief description of the background to the trial and details of the reimbursement rates for GP practices with patients involved in the trial. It was the first time this methodology had been tried.

General practices in the local area (i.e. Camden and Islington with Kingston) were reimbursed £70 if they arranged to view this video within a practice clinical meeting and could give evidence of the GPs in the practice having watched it. Ten practices were reimbursed for watching the video: seven in Camden (out of a total of 40 practices) and three in Islington (out of a total of 38 practices). Of these, all but one practice in Islington expressed an interest in taking part in the study if any of their patients were potentially eligible.

Interest was also expressed in the video by the other pilot sites where the North Central London Research Consortium was not in a position to reimburse the practices, and they were given the link to the video to watch if they wished. This has now entered the public arena via YouTube (YouTube, LLC, San Bruno, CA, USA). 55

Reasons given by psychological well-being practitioners for the difficulty recruiting participants to ToSCA

• We explained to the PWPs in both the written materials and face-to-face training the importance of having equipoise in explaining the trial to potential participants, and that it was an important unanswered clinical question whether CBT or medication would be the most effective treatment for these patients. However, the fact that they had been trained to deliver a low-intensity CBT intervention is likely to have given them both an overt and also less conscious bias towards psychological therapy treatments.

• A number of PWPs said that they thought many patients they saw were against trying psychotropic medication. This was backed up by many of the documented patient responses to being told about the trial, but it was unclear if the patients may have been influenced in this by the PWPs. We gave the PWPs examples of answers to give to patients who might have been concerned about the side effects of medication (Box 2).

• Several PWPs were concerned that patients would not be offered CBT after trying medication if the medication proved unhelpful, possibly indicating a greater enthusiasm for psychological therapy.

• Some PWPs said they thought GAD was not as prevalent in the clinical population as generally believed and that, in their experience, GAD presentations are often not the focus of psychological treatment.

• Quite a few PWPs wanted to be certain that people they discussed the trial with had GAD, although we had stated that this was not possible at their stage in the process and that we would like them to offer the possibility of the trial to anyone with a reasonable likelihood of GAD, as they would then need to be assessed using the gold standard MINI questionnaire. The PWPs said they were concerned about wasting clients’ time if they did not have GAD.

• Time pressures were a virtually universal problem – all the PWPs had significant clinical caseloads and found it difficult to keep the trial in mind.

• Many PWPs reported that their clients were very anxious about the uncertainty of being referred and allocated to a random treatment. Given that their key problem is likely to have been GAD, with a core component being worry about uncertainty, this illustrates the potential difficulty of recruiting participants to a RCT in GAD. When this barrier was identified, PWPs were encouraged to suggest to potential participants that they meet with the researcher and address their questions to the researcher, rather than the PWP attempting to answer them.

• This problem with worry about uncertainty also made the PWPs concerned about making their patients more anxious when discussing the trial, with implications for their workload, as well as concern for the well-being of their patients.

A further unanticipated barrier identified through these discussions was that the supervisors of the PWPs often considered that a comorbid problem which the patient had, either diagnostic (e.g. social anxiety) or psychosocial (e.g. debt problems or relationship difficulties), was the key clinical issue to address and would direct the PWP to refer the patient for an intervention addressing this problem which meant they were not considered for the study. When this barrier was identified, local PIs and the study team encouraged PWPs and supervisors to refer such patients for a ToSCA baseline assessment where the suitability of treatment for GAD would be addressed rather than pre-empting this.

Retrospective database searches

The retrospective method was initially used to identify cases potentially meeting study criteria, who had been discharged from low-intensity PWP treatment and stepped up to be awaiting high-intensity CBT within the service. It did not prove to be very useful in terms of identifying potential participants for the trial as, of the 18 patients who had been referred for stepping up who still had a GAD score of ≥ 10, nine had depression as their main problem, two had other clinical problems they wanted to address, one had moved, one dropped out, two declined as they did not want medication and three did not respond. There were also 14 patients who had been discharged from the IAPT service despite still having a GAD score of ≥ 10, of whom six reported having other clinical presentations as their main problem, one declined when contacted about the study and seven did not respond.

Following on from this, we examined the participant flow for the whole trial recruitment period retrospectively and this indicated a significant number of people who might have been suitable to approach to discuss assessment for participation in the trial who had not been identified by the PWPs. Using the IAPTUS database, the flow of people with GAD-7 scores of ≥ 10 during the active study recruitment period was analysed at the IAPT pilot site covering the same two local London boroughs used to estimate likely recruitment flows in 2011–12 before the trial (i.e. Camden and Islington with Kingston). Approximately 1 in 4 people assessed for treatment with a GAD-7 score of ≥ 10 were seen for at least three sessions of a low-intensity treatment (the other three out of four people either went straight to high-intensity CBT or were discharged, referred on or dropped out before having two low-intensity treatment sessions). Of those who completed at least three sessions of a low-intensity intervention, 37% were being a prescribed a psychotropic medication (data missing on a further 2%), and 50% of those who were not prescribed medication improved and had GAD-7 scores of < 10 at the end of their low-intensity treatment. All of these therefore did not meet criteria for inclusion in the study.

Of those patients who were not on medication and had not recovered, a proportion dropped out of treatment before the end of their low-intensity intervention and would not have been able to be approached by their low-intensity worker about the study. However, of those who remained potentially approachable to discuss the trial, approximately only 1 in 6 had been identified to the study team as having been approached about the study by their low-intensity worker/PWP. Unfortunately, following a change in the clinical database used by the IAPT service, the completion of diagnostic coding during the recruitment period was poor (only 28% of people with GAD-7 scores of ≥ 10 had primary diagnoses coded on the system) and so we were unable to use this to calculate how many of the potentially eligible and approachable patients were considered by their IAPT low-intensity worker to have a primary diagnosis of GAD rather than other diagnoses.

Prospective database searches

The prospective identification method was used to identify cases potentially meeting the study criteria who had already had three or more sessions of PWP treatment and were still in treatment with a PWP. It was instituted on a regular basis at the Camden and Islington with Kingston site, and the algorithms and methods were disseminated to the other pilot sites. On a regular fortnightly basis, cases were extracted from the IAPTUS database that met the following criteria:

1. most recent GAD-7 score of ≥ 10

2. seen three or more times in a step 2 treatment (could include the PWP assessment session)

3. still in a step 2 treatment

4. not on a psychotropic medication.

An e-mail was sent to the PWP treating each patient identified through these searches, giving the patient’s identification details, noting that the patient was potentially eligible for ToSCA and asking the PWP to e-mail, by return, if there was some reason that they were not eligible for the study (e.g. that the patient had an agreed diagnosis or diagnoses other than GAD).

In addition, a note was placed on the front page of the patient’s electronic records in IAPTUS (on the page that first comes up when the patient’s records are entered) that the patient was potentially eligible for ToSCA.

Using the standard IAPTUS search to identify people at step 2 who had a diagnosis of GAD highlighted a problem that people had not always been given an accurate provisional diagnosis. In order to identify potentially eligible participants, this involved checking clinical records for GAD scores, session number and medication use.

Assessment of competence

A scale for measuring therapist competence in cognitive therapy, based on the original Cognitive Therapy Scale,58 is the 12-item Cognitive Therapy Scale-Revised (CTS-R). 59 This revised version improves on the original Cognitive Therapy Scale by eliminating the overlap between items, improves on the scaling system and defines items more clearly. In this trial we planned to have an independent rater listening to audio-recordings of therapy sessions using the CTS-R.

Assessment of adherence

In this context we defined therapist adherence as the extent to which the therapist stuck to the essential elements described within the treatment manual.

We aimed to collect detailed information about the content of the intervention using a Therapy Components Checklist (TCC), which was developed for use in the trial. The TCC summarises five main interventions described in the Dugas treatment manual (see Appendix 13) and deemed essential for successful treatment. Each of these areas is made up of a number of elements:

1. psychoeducation and worry work (three elements)

2. evaluation of the usefulness of worry (two elements)

3. uncertainty recognition/exposure (three elements)

4. problem-solving (three elements)

5. written exposure (five elements).

The TCC also includes sections about general therapy procedures (10 elements) and the specific materials used (nine elements). We also aimed to collect information about a number of interventions that are not permitted and, if used, to assess deviations from the protocol (e.g. use of controlled worry periods). There is one TCC completed per patient and the therapist reports which components were addressed after each therapy session. A more detailed description of these can be seen by referring to the TCC (see Appendix 14). The TCC was used in supervision, but not shared with the patient.

At the very end of a course of therapy, up to 16 sessions, the main elements of the treatment delivered are summarised in the brief End of Therapy Checklist (EoTC) (see Appendix 15) completed by the therapist. This consists of the same five main areas/interventions detailed above. The EoTC was not used in supervision or therapy, but solely for the purpose of assessing adherence. This was also developed by the CBT group for use in the trial.

Defining ‘sufficient’ adherence

The degree of adherence was rated using the EoTC as follows:

• 0 – non-adherent (between zero and two out of the five components delivered on the EoTC, i.e. < 40% of the session was spent using methods from the protocol; the rest of the time was spent on methods not in the protocol or generic techniques, such as empathic listening)

• 1 – somewhat adherent (two or three out of the five components delivered on the EoTC, i.e. 40–59% of the session was spent using methods from the protocol; the rest of the time was spent on methods not in the protocol or generic techniques)

• 2 – mostly adherent (three or four out of the five components delivered on the EoTC, i.e. 60–79% of the session was spent using methods from the protocol; the rest of the time was spent on methods not in the protocol or generic techniques)

• 3 – fully adherent (at least four out of the five components delivered on by the EoTC, i.e. ≥ 80% of the session was spent using methods from the protocol; the rest of the time was spent on methods not in the protocol or generic techniques).

Data collection

Training workshop

Training for ToSCA consisted of a 2-day workshop, on 22 and 23 January 2015, led by Professor Michel Dugas, who developed the CBT intervention being used in this trial and has conducted previous trials in GAD.

Attendees included the CI (MB), co-applicants from the project who are experienced CBT therapists (RS, MS and JC) and staff from the IAPT pilot centres located in south-east England – London (Camden, Islington, Greenwich), Surrey (Kingston), the Midlands (Coventry, Warwickshire) and the west of England (Bristol). The IAPT staff at each of these sites were very enthusiastic about the opportunity to receive training from Michel Dugas and places had to be limited to two therapists and one supervisor for each single pilot site (i.e. Greenwich and Bristol) and four therapists and two supervisors for the sites committed to a double recruitment figure (i.e. Camden and Islington with Kingston, and Coventry and Warwickshire) (see Appendix 9).

The training was delivered as a workshop, with scenarios in which attendees were encouraged to get into groups and practise using role play. It was based on the manual provided by Michel Dugas, and each therapist and supervisor was given a copy for use in the trial (see Appendix 13).

Plan to ensure therapist competence

All attendees met at the end of training on the final day to discuss implementation issues at their local sites. The plan was for each therapist who was going to be involved in ToSCA to treat two patients who had already been identified as having GAD within the IAPT service, using the manualised intervention that they had been trained in using by Michel Dugas. These patients would be used to help therapists and supervisors become familiar with the protocol, and optimise adherence and competence for the main trial. As these patients had not been recruited or consented to take part in the trial, it was not possible for the tapes or transcripts of their therapy sessions to be made available to the external supervisors. The local supervisors brought any particular issues arising from the treatment of these patients to their ‘supervision of supervision’ sessions provided by Michel Dugas and Roz Shafran. We have termed the patients treated in this way ‘practice patients’, in order to distinguish them from patients who were recruited for the internal pilot who underwent a rigorous eligibility and consent process.

It was agreed that in the practice patient stage the therapists would need to treat at least two patients and the supervisors would rate the therapist on at least one session using the CTS-R. 60 It was agreed that the therapists would need to have achieved a score of ≥ 36 to be considered competent and be permitted to treat patients in the trial. A minimum score of 36 is the standard criterion for competence within IAPT services. It was intended that competence in the delivery of CBT for the main trial would be assessed in the same way (i.e. a score of ≥ 36 according to the CTS-R). In the main trial, it was intended that an expert rater (Melissa Robichaud) would independently rate 10% of all therapy sessions using the CTS-R to establish that the protocol had been delivered competently. In the unlikely event that the therapy had not been delivered with competence or adherence to the protocol, this would have been reported.

Plan to ensure therapist adherence

Adherence to the protocol is a critical part of any trial. We therefore developed a measure of adherence to the protocol for use in the main trial. The practice patient stage enabled us to test the feasibility of this measure. As described in Assessment of adherence, the TCC is intended to be completed after every session by the therapist to record the main content of the session, and the EoTC is designed to be completed at the end of the course of therapy by the therapist. The therapists were asked to bring the TCC completed on their ‘practice patients’ to their supervision sessions so that the local supervisors could help them maintain adherence to the protocol.

We planned to also use the TCC in the main trial to assess adherence, in addition to external assessment of adherence to the protocol by an expert rater (Melissa Robichaud), who had agreed to rate 10% of the sessions selected at random to ensure that the therapists had adhered to the protocol. The intention was for the external rater to listen to 10% of the sessions selected at random and rate them for adherence and competence simultaneously. We planned to examine the relationship between therapist self-ratings of adherence and external expert ratings of adherence. If there was a strong and positive relationship between the ratings, we could conclude that therapist ratings of adherence were accurate reflections of the content of therapy, which would allow future research to dispense with the need for costly independent ratings.

Supervision structures

The IAPT therapists were supervised by local IAPT leads from the participating centres – both the supervisors and therapists were all British Association of Behavioural and Cognitive Psychotherapy-accredited therapists. The IAPT supervisors from Kingston, Camden and Islington and Bristol were externally supervised by Roz Shafran and those from Greenwich, Coventry and Warwickshire by Michel Dugas. Both external supervisors were ToSCA co-applicants.

Routine supervision of therapy in IAPT takes place at least monthly, but is flexible within this period. However, in this trial, we recommended flexibility so that if any immediate issues needed attention, the therapists could consult their IAPT supervisors. The external ToSCA supervisors (RS and MD) provided additional monthly supervision sessions (‘supervision of supervision’) by teleconference link to the IAPT supervisors, which was part of the trial process and was in addition to usual clinical practice. The external ToSCA supervisors were also accessible to the local IAPT supervisors by e-mail to answer any additional queries that arose between supervision sessions.

Flexibility in the practice patient stage was used to learn about how clarification and modifications to the CBT intervention might be required. This was done by keeping rigorous notes during supervision to help inform the project, as well as e-mail discussions between the supervisors and the trial team. It was agreed that once the ‘practice phase’ had finished, there would be no modifications to the protocol allowed.

Supervision

Direct supervision of IAPT therapists in the services took place, on average, monthly for between 60 and 90 minutes individually, in pairs or in groups of three or four, depending on the service. The expert ‘supervision of supervision’ (i.e. supervision of the IAPT supervisors by RS and MD) also took place monthly between July 2015 and January 2016 for 60 minutes. This was done over a dial-in telephone service, with times and dates agreed in advance. Michel Dugas supervised four IAPT supervisors directly and Roz Shafran supervised three. A total of 12 supervision sessions took place, six for each of the supervisors, monthly, with the exception of August 2015. Detailed notes were kept with any queries about protocols minuted and addressed at the next supervision session after consultation with the research team.

Patients treated with cognitive behavioural therapy

Of the team supervised by Roz Shafran, 7 out of the 11 therapists had cases. A total of 16 patients were seen, with a mean of 2.3 patients per therapist; the mean number of sessions was 8.8 (SD 8.5). Of these, three completed ≥ 14 sessions. In 8 out of the 16 cases the GAD checklists (TCC and EoTC) were completed. Reasons for ending therapy in the 16 patients were as follows: two clients completed therapy (one had received 16 sessions and the other was well after 10 sessions), three clients completed prematurely as they did not want to do ‘exposure’, two were unwilling to allow the recording of sessions, five had not yet completed treatment and four withdrew (one moved, one was too busy to attend and two withdrew without giving a specific reason).

Ease of supervision

Telephone supervision proved to be a useful way for the ToSCA expert supervisors to supervise the IAPT supervisors. The IAPT supervisors were able to represent the views and queries that their local therapists had raised. These are included in the summary of questions on the manual and treatment manual (see Clarification and modification to treatment protocol as a result of the practice patient stage below).

Adherence and competence

The TCC and EoTC checklist were completed or partially completed for 8 of the 16 participants seen. The supervisors considered the therapists to be adherent according to the checklist, but noticed that the most common component of the protocol that was omitted was the written exposure.

The supervisors considered the therapists to be competent on the basis of listening to excerpts from their sessions, but none had rated a full session using the CTS-R because of the time required to do so and the fact that not all of the ‘practice patients’ were being recorded as consent had not been provided for recording.

Clarification and modification to treatment protocol as a result of the practice patient stage

Confidentiality

The project aimed to use Data Safe Haven to transfer and store audio-recordings of therapy. This service provides a technical solution for storing, handling and analysing identifiable data. It has been certified to the ISO27001 information security standard and conforms to the NHS Information Governance Toolkit. This is built using a walled garden approach, in which the data are stored, processed and managed within the security of the system, avoiding the complexity of assured end point encryption. A file transfer mechanism enables information to be transferred into the walled garden simply and securely (see Chapter 3, Research governance).

Clinical Advisory Group

The CAG was made up of three service users plus the PPI co-applicant. The CAG met roughly three times per year. Ad hoc meetings were held as needed. The responsibility for organising CAG meetings was held by the PPI co-applicant, Thomas Kabir, who is employed by the McPin Foundation.

The CAG members were recruited via an open advertisement. The three people that were subsequently recruited all had personal experience of anxiety. Meetings were chaired by Thomas Kabir. Members of the wider research team were invited to attend meetings, and all meetings were minuted. The minutes of CAG meetings were shared with the research team.

A total of six CAG meetings were held between August 2014 and June 2016. As per the NIHR’s INVOLVE Briefing Notes for Researchers,62 CAG members were reimbursed for any out-of-pocket expenses. Members were offered payment for attendance at meetings.

Thomas Kabir was the PPI co-applicant for the study. He provided input into the application for funding to the HTA programme as well as the application for ethical approval. He later:

• sat on the Trial Management Group (TMG)

• attended TSC meetings as an observer

• recruited and organised CAG meetings

• acted as a link between the CAG and the wider study team

• advised on the recruitment materials for the study

• provided ad hoc advice to the study team as needed.

Reflections

A major initial focus of the CAG was on measuring the acceptability from a patient perspective of the two interventions being tested against each other: sertraline and CBT. Developing ideas about how to measure the acceptability of both a psychological therapy and a medication proved to be a particular challenge. In the end, the decision was taken to use the Client Satisfaction Questionnaire (CSQ). The CAG disagreed with this decision, as the CSQ appeared to focus on satisfaction with services rather than treatments per se. However, the TMG decided on the CSQ because of the perceived need to use one questionnaire to get participant feedback about both interventions in order to be able to use this in the analysis, and no other questionnaires were found that were able to be used to do this.

A related issue arose when dealing with the issue of what outcome measures to use in the study. CAG members felt that the adverse effects of both CBT and sertraline should be measured. This posed a problem, as it proved that there were no measures that were truly applicable to both a medication and a psychological therapy. The CAG suggested using a modified form of the Toronto Side Effects Scale, but the wider study team did not see this as an adequate solution as it was designed with drug-related side effects in mind. The CAG felt that a modified version, taking out some of the terms, could still work, but the wider study team felt it was not possible within the trial timelines to proceed with this. Additionally, there was some debate within the wider study team as to whether or not adverse events that would not be classified as serious in nature should be measured at all, and the TMG noted that the side effects of sertraline have been fully documented in other studies. CAG members held the view that although this was true, the adverse effects experienced by people actually taking part in the study needed to be considered. The idea was that both the adverse and beneficial effects experienced by study participants needed to be weighed against each other to get a true notion of the net worth of the treatments being considered.

As the study progressed, more attention was given over to recruitment issues by CAG members and the PPI co-applicant. CAG members felt that the responses made by the TMG to the recruitment issues that the trial faced were appropriate. Indeed, it was felt that the study team had done all that they could to address the recruitment problems that the trial faced. It was noted from an early stage that recruitment to the study would be challenging. The following extract from the minutes of a CAG meeting held in November 2015 summarised the CAG’s position:

A possible downside to taking part in the study is that participants will have no choice about whether they take medication or CBT. Outside of the study people can potentially access both treatments. This is clearly a barrier to participation.

Specific suggestions were made regarding how low-intensity IAPT workers could introduce the study to potential participants. One CAG member wrote a script for IAPT workers to use, which was then shared with other study team members and used in the revised version. More practical issues, such as how best to ask female participants to take a pregnancy test as part of the baseline eligibility interview, were also discussed. The recruitment strategies that were identified by the study team were fully discussed by the CAG.

Questions from the Clinical Advisory Group as a result of the study

• The CAG asked if the funder had sufficiently foreseen the potential recruitment problems with the study. The CAG felt that there would be significant numbers of people who would prefer to receive CBT rather than sertraline from the outset. This may have meant that a different study design may have been more beneficial.

• CAG members asked if the commissioning brief for the study was reviewed by people with experience of mental health problems specifically. If so, did these people identify the potential recruitment issues that may arise from the original commissioning brief?

Recommendations for future research

• It seemed that there are relatively few treatment acceptability measures in routine use. It was felt that it would be useful for the NIHR or another funder to fund research into developing a treatment acceptability measure that would be of broad use within a mental health research setting. Within the context of ToSCA, a measure that would work across different treatment modalities (a drug and a psychological therapy) was needed.

• Likewise, there are few (if any) measures that can be used to measure the adverse effects of both a drug and a psychological therapy. Again, it was felt that it would be useful to develop a measure that could be used across different kinds of interventions.

Trial Steering Committee response

These issues were fully explored at the TSC meeting on 29 September 2015, and the chair of the TSC wrote to the HTA programme informing them of the recruitment figures and of the various strategies (please see the following paragraph for further details) that the study team had been working on to address this issue. In response to this letter, the HTA programme arranged a monitoring meeting for 14 January 2016 to discuss both the reasons for the poor recruitment levels and to explore possible alternative methods that could address this issue. At the meeting, a presentation of the various strategies that were adopted by the trial team to enhance recruitment was summarised.

Health Technology Assessment monitoring meeting: strategies adopted to enhance recruitment

1. PPI input to enhance recruitment: we had worked closely with our PPI co-applicant, Thomas Kabir, and the PIs at each clinical site to enhance recruitment by offering eligible participants a balanced view of both drug and psychological treatments for GAD. We refined, with help from the PPI CAG, a user-accessible script that the PWPs would adopt to offer an unbiased account of both treatments while also tackling concerns that people had about antidepressant therapy. This was intended to be discussed with all people with a GAD-7 score of ≥ 10 on completion of step 2 IAPT treatment.

2. Engaging with PWPs at the sites to ensure that they were actively recruiting to the study. This was done by:

   1. the trial team regularly attending PWP site meetings, in which they reminded PWPs about the trial and were available to address any queries

   2. weekly telephone contacts with sites by ToSCA to field any queries on participant identification and baseline assessments

   3. 2-monthly teleconferences at which all pilot sites discussed and shared information about effective recruitment strategies.

3. Other methods used to identify eligible patients were:

   1. assisting the identification of potentially eligible people through the IAPTUS database (IAPT’s electronic clinical system), as an additional mechanism of identifying potential participants. We initially worked with the lead PWPs in Camden, Islington and Kingston on running the following searches:

      1. – retrospective searches to identify potentially eligible people missed by the PWPs during their clinical reviews at the end of their step 2 treatment

      2. – prospective searches to identify potential participants at their entry to the step 2 treatment (i.e. those with high GAD-7 scores not on antidepressants) and alerted the PWPs in advance about their possible eligibility

   2. IAPTUS prompts to PWPs of potentially eligible people: we were working with the IAPT data managers on generating a computer prompt on the IAPTUS database in order to flag potential participants, as described in the paragraph above. This, however, was not finalised and, hence, not implemented

   3. approaching people on the waiting list for step 3 (high-intensity therapy): we explored the possibility of getting eligible people to take part in the trial but, as they were on the waiting list for CBT, they were expecting to receive psychological treatments and, hence, this approach did not yield any participants

   4. opening new pilot recruitment sites: on 13 November 2015 we proposed the recruitment of more clinical sites to boost our study numbers. We had received recent expressions of interest from six sites through our contact with the CRNs and NHS trusts, as well as several over the past year that the study had been advertised on the web. When approached, four sites expressed a definite interest in taking part in the study, but after further discussion only two of these were willing to actively recruit.

These strategies were discussed at the HTA programme monitoring meeting on 14 January 2016. It was made clear that despite all the efforts that had been made by the trial team, recruitment to this study was unlikely to achieve our final target number of 360 participants and we suggested that we alter our recruitment strategy through a protocol change. Prior to the HTA programme monitoring meeting, in discussion with the TSC, we had considered two possible future protocol changes: options A and B. We requested that one of these two, or both used in succession, be adopted following our meeting with the HTA programme monitoring committee. These two options are detailed below:

1. Option A – this would have involved a drive to identify more people in primary care through a retrospective search of GP databases to identify all adult patients aged ≥ 18 years with Read codes63 for anxiety/depression and/or GAD-7 scores of ≥ 10. Letters would then be sent to those so identified containing a reply slip, consent form for screening questionnaires, GAD-7 to complete, exclusion criteria assessment checklist and a brief questionnaire asking them to express interest in participation in the study. This approach aimed to maximise the identification of patients with GAD in primary care, and to refer them on to the local IAPT services. The TSC were supportive of this change of protocol as it was consistent with the HTA programme commissioning brief. They were concerned, however, that such a strategy could overwhelm the local IAPT services, which were likely to already be under pressure and may not have been prepared to deal with more referrals. Furthermore, there could be a significant delay while people were receiving step 2 treatment before they became eligible for participation in the trial. This could then lead to high levels of dropout. Nevertheless, the advantage of such an approach was that it was tied in with the changes we had already implemented to enhance recruitment from within IAPT services.

2. Option B – the process of identifying suitable patients in option B would have been similar to that of option A. We would have run a retrospective search of GP databases to identify all adult patients aged ≥ 18 years with Read codes for anxiety/depression and/or GAD-7 scores of ≥ 10. Letters would then be sent to patients containing a reply slip, consent form for screening questionnaires, GAD-7 to complete, checklist of exclusion criteria and a brief questionnaire asking them to express an interest in taking part in the study. Suitable patients would then be assessed for their eligibility to take part in the trial in general practice and then be randomised to either the sertraline medication arm or high-intensity CBT without having to engage with the step 2 treatment as delivered by the PWPs.

Option B would have been essentially a move from the original question proposed in the HTA programme brief as it involved a change in research population. Rather than pointing people towards IAPT teams for step 2 treatment (as in option A), this option would have recruited to the trial without a step 2 intervention. This in itself is an important clinical question, but would have constituted a move from the HTA programme brief as the study would not recruit merely those who have failed to respond to a step 2 intervention as specified in the brief. The TSC initially felt it premature to consider such a departure without (1) testing the effectiveness of the planned current changes to recruitment that would adhere to the HTA programme brief, and (2) seeking the opinion of the funder, the HTA programme, as to whether or not this amendment of the protocol (and thus research question) should be held in reserve if other changes, as described above, had failed to improve recruitment.

Finally, a radical suggestion, bearing in mind that patient preferences are a major hurdle to recruitment, was to design a patient preference trial in the form of a comprehensive cohort preference trial.

The HTA programme monitoring committee was not supportive of option A as this was an approach that would not be adopted within the NHS in the future, even though the proposed protocol change would have adhered to the HTA programme brief. The committee did consider option B to be a viable strategy but, as it represented a significant deviation from the original commissioning brief, they wanted to consult with the HTA programme director in order to arrive at a final decision. At the conclusion of the HTA programme monitoring meeting the trial team was advised to persist with efforts to recruit until further notice was given by the HTA programme regarding whether or not they would allow a deviation from the original commissioning brief. If this was not considered acceptable, we were advised that the trial would be expected to close down, hence bringing the study to an end.

Low rates of identification of potential participants

It was unfortunately not possible to recruit sufficient participants to the trial of CBT versus sertraline following the recommended brief. The main barrier appears to have been at the level of identification of potential trial participants by the IAPT PWPs. There are a number of possible reasons for this, which can be summarised as follows:

1. The PWPs potentially having a bias towards psychological therapy as the treatment of choice and finding it difficult to maintain clinical equipoise when talking to the patients about the trial.

2. The PWPs being aware that having GAD meant that these patients found uncertainty very difficult and that the prospect of being randomised to a RCT for their treatment with the associated uncertainty of this was particularly difficult for them to manage; the PWPs felt uncomfortable raising something that might make their patients more anxious.

3. Some PWPs considered that the prevalence of GAD, or having this as the main clinical problem that patients wanted to address, was lower than suggested by the trial team and they were very keen to ensure that the patients they suggested the trial to definitely had GAD, despite the research team explaining that a definitive diagnosis could not be made at the routine PWP level.

4. Several PWPs raised the issue to their clinical supervisors, suggesting that a patient’s comorbid clinical or psychosocial problem was more important to address than their GAD. In such cases, the research team encouraged both the PWPs and their supervisors to consider suggesting to patients that they might be assessed for ToSCA if they had significant GAD.

5. Most PWPs had significant clinical workloads and it was difficult for them to also include the time needed to raise the possibility of being assessed for the trial with suitable patients.

Patient factors affecting recruitment once they had been identified

The reasons given by many of those patients who had been identified as not wanting to be recruited into the trial largely mirrored what the PWPs had suggested, in that by far the most significant reason given was that the patient did not want to risk being randomised to the sertraline arm of the trial, and several expressed a clear preference for wanting CBT treatment. Not wanting to be given medication appeared to be the predominant factor, although finding the uncertainty of the randomisation process difficult may also have been a factor. In retrospect, the nature of GAD – worry when there is uncertainty – should have alerted us to the likelihood that discussions about the study and the uncertainty raised by randomisation would prove challenging.

The reasons people gave for declining to participate were predominantly that they were unwilling to try antidepressant medication or that they were clear that they wanted psychological treatment. Although surveys indicate that people commonly report a preference for psychological over pharmacological treatment for common mental health problems, the number of people unwilling to consider antidepressant medication in our study was in excess of what would be expected from these surveys. One possibility is that the context of recruitment within an IAPT service may be significant. The service is focused on psychological treatments (in its name as well as in its treatment provision), and this sets a context for both patients and staff. The IAPT staff approaching patients about the study, however much the need for equipoise was discussed with them, are likely to have had an allegiance to psychological interventions and may have subtly communicated this to potential participants. Even if this did not happen, people opting to be seen in an IAPT service in the first place are likely to be those interested in psychological treatment and they may well have had less favourable views towards pharmacological treatments.

Another factor that was less of an issue than we had expected and made assumptions for was comorbid major depression – only one patient of the seven who had a baseline assessment was found to be ineligible because of major depression on the MINI questionnaire, which was much lower than we had expected with our 50% estimate, although the number of patients assessed was small. It may have been that by being very selective in the potential patients they identified, the PWPs had excluded those who were clearly significantly depressed. The other patient found to be ineligible at the baseline assessment had found it difficult to be sure whether or not GAD was the main psychological difficulty for which they wanted treatment – after discussion with the TMG we clarified the wording of this screening question to make it less potentially ambiguous.

In advance we had assumed that two-thirds of people meeting criteria for the study in other respects would either be on antidepressant medication or decline to participate. In the event, a little over one-third of potentially eligible people were already on antidepressant medication, and one in five participants approached ended up agreeing to participate in the study.

The fact that two GPs declined to support their patients in their wish to be assessed for the trial was disappointing, but it is difficult to know how we could have avoided this, given the trial design that recruited participants from IAPT who could be registered with any of the GPs linked to that service. The only way to have avoided this would have been to consider only patients whose GPs had agreed in advance to be involved in study, but the response rate to a request from the local research networks for interested GPs to state this was very low, which was not very surprising given that the chances of them having a patient selected to take part in the trial were not high. The only other way to deal with this issue would be to recruit directly from primary care, which is a strategy we considered (please see the following section) and suggested to the HTA programme at the monitoring meeting, but this was turned down by the funders because it would have been a major alteration to the commissioning brief.

Systemic factors affecting participant identification and recruitment

Linked with this is the fact that the advent of the national IAPT programme in England has meant that CBT and related evidence-based psychological treatments are now readily available and accessible as standard treatments. This means that if people have a preference for CBT they can access this relatively easily, in the same way that people have for much longer been able to easily access antidepressant medication via their GP. When access was more difficult, people might have been more prepared to consider a trial, but now if people have a preference they may well exercise this through opting for their preference. Two potential participants demonstrated this by stating a wish to have both CBT and antidepressant medication when the study was discussed with them.

An alternative treatment strategy would have been to recruit from primary care. In primary care, the full range of people with GAD would have been available to recruit, and primary care is where initial discussions naturally take place about treatment options between drug and psychological treatments. People are more likely to be in equipoise between drug and psychological treatments at this point than focused towards psychological treatments, as they are later in an IAPT service and, accordingly, are likely to be more open to accepting randomisation. They would also be less likely to already be on medication. This recruitment option was not available to the study team as the commissioned brief was, following the NICE stepped-care model, specifically to target people who had not improved following a low-intensity psychological intervention. Arguably, the restriction in the total GAD population from the brief as given would have limited the generalisability and utility of the findings even if recruitment had been successful, and recruiting from the wider primary care population would have been more useful and generalisable in the clinical sense. However, it is possible that, even if recruiting in primary care, people would have clear preferences between medication and psychological treatment and, as both are now relatively easy to access in England, they might well exercise that choice and be unwilling to accept randomisation to one or other intervention.