Reducing relapse and suicide in bipolar disorder: practical clinical approaches to identifying risk, reducing harm and engaging service users in planning and delivery of care the PARADES (Psychoeducation, Anxiety, Relapse, Advance Directive Evaluation and Suicidality) programme

Previous research in psychoeducation

Group PEd for BD is a manualised, evidence-based intervention. As relatively large groups of 10–18 people can undertake treatment together, group PEd is an efficient and easy-to-set-up option for MHS to deliver psychological treatment to improve longer-term outcomes. Although therapists need to be trained to run the groups, the training is less intensive than that for other psychological approaches such as individual CBT. 60 The first RCTs of group PEd versus group PS showed the clinical effectiveness and cost-effectiveness of group PEd for all types of bipolar relapse, with better outcomes maintained for 5 years. 33,61,62 However, these trials were carried out in a specialist BD service in Spain, which bears little resemblance to the UK NHS. Additionally, the relapse rates in the control arm of this Spanish trial were significantly higher than expected, with 92% of the sample who attended support groups experiencing a further bipolar episode within 2 years. This level of relapse was cited as an outlier in a recent meta-analysis of eight RCTs of PEd,32 which showed, in contrast, an unweighted mean of 70% relapse in treatment as usual (TAU) or attention control studies over follow-up periods ranging from 52 to 104 weeks. 32 Overall, in this meta-analysis, group PEd was effective in relation to all relapses [odds ratio (OR) 1.98, 95% confidence interval (CI) 1.09 to 3.58; p = 0.024; I² = 54%; seven studies, 493 participants], almost significant for mania relapse (OR 1.68, 95% CI 0.99 to 2.85; p = 0.06; I² = 55%; eight studies, 562 participants), but not significant for depression relapse (OR 1.39, 95% CI 0.78 to 2.48; p = 0.26; I² = 63%; eight studies, 562 participants).

In another meta-analysis of RCTs for development of the NICE clinical guideline, the evidence base was reviewed for group psychological treatments overall, including CBT, mindfulness therapy, social cognition and interaction training, and dialectical behaviour therapy. 57,63 Up to 26 weeks, group psychological interventions showed a trend towards reducing overall relapse [relative risk (RR) 0.48, 95% CI 0.22 to 1.04; I² = 59%; eight studies, 532 participants], reducing mania relapses (RR 0.48, 95% CI 0.28 to 0.82; I² = 0%; six studies, 564 participants) and reducing depression relapses (RR 0.39, 95% CI 0.19 to 0.78; I² = 0%; seven studies, 616 participants), compared with TAU or attentional controls. By 52 to 104 weeks’ follow-up, the benefits of group psychological treatments were confined to a reduction in depression relapses (RR 0.62, 95% CI 0.45 to 0.88; I² = 44%; five studies, 333 participants), with no effects on mania relapses (RR 0.97, 95% CI 0.60 to 1.57; I² = 69%; five studies, 328 participants) or on overall relapse (RR 0.86, 95% CI 0.61 to 1.20; I² = 81%; five studies, 395 participants).

Limitations of research to date

As observed by NICE, the evidence base for group psychological treatments in preventing relapses for BD is inconsistent and imprecise. 63 The vast majority of RCTs were small (with around 100 participants or fewer), single-centre trials, sometimes in specialist settings and often with only 12-month follow-up, and the meta-analysis shows high heterogeneity. Some trials were not pre-registered on trial databases, few reported functional, quality-of-life and economic outcomes, and there was evidence of selective reporting of outcomes. Furthermore, most RCTs did not control for the clustering effect of group treatments in their analysis, thereby overestimating the effectiveness of their treatments. 64 Furthermore, none of the trials of group PEd to date has occurred within a UK setting. In addition, there are concerns in relation to RCTs of psychological interventions that they are often carried out by the developers of the treatment approach, and that there is lack of blinding and of appropriate control groups that control for expectancy of effect as well as frequency, length and duration of treatment. 65 Thus, there is a need for a large multicentre RCT of the clinical effectiveness and cost-effectiveness of group PEd carried out in non-specialist service settings in the UK that addresses all of these methodological issues.

In both the UK and other countries, people with BD often have the opportunity to access support groups, such as those run by Bipolar UK, as well as receiving medication and follow-up from health professionals. People with BD value such support,66 and PS groups run by expert patients improve self-efficacy and are effective for many health conditions. 67 The nature of the group PS in the first trials of PEd in Barcelona is unclear. 61,62 NICE57 could not find any RCT of support groups run by peers versus TAU for BD.

Aims of the current study

The primary aim of this RCT was to determine whether or not the content of a group PEd programme, as outlined by Colom et al. 60,68,69 but contextualised to UK practice, was clinically effective and cost-effective compared with group PS, the most frequently available form of group psychological treatment in the UK, with over 120 groups run by Bipolar UK (www.bipolaruk.org/find-a-support-group, accessed 9 October 2017). To ensure that groups were acceptable to participants and that participants benefited from both professional expertise and lived experience,70 both group interventions were co-facilitated by two health professionals and a trained SU. To ensure generalisability, health professionals currently working in NHS services were recruited to run the groups, and the inclusion and exclusion criteria for participants were designed to be as inclusive as possible.

To ensure that group PS was an effective control, both groups received the same number of sessions of the same duration; each set of sessions involved two health professionals and a SUF and started off with the same number of participants with BD, and each meeting was held at the same place but on a different day that was not disclosed to the blinded assessor of outcome. Each group intervention had a manual and had the same aim of treatment, which was to use the resources of the group to develop a plan to prevent further bipolar episodes and improve overall function. Both groups were supervised by senior clinical academics and both groups were closed. Therefore, although this group PS was similar to the group PS offered by Bipolar UK, in that the participants determined the content of the group, it was also a long-term supervised closed group with a more specific aim, so in many important ways it was quite different from the usual group PS offered in the UK. We also addressed many of the issues highlighted by NICE58,63 as key problems with previous group psychological treatment and PEd RCTs for BD by ensuring that our trial was multicentre and pre-registered, and had relapse, symptom, function, quality-of-life and economic outcomes (reported in the next chapter). The trial was carried out in routine secondary care mental health settings. (A protocol paper has already been published. 71)

Aims

• To determine whether or not it was feasible to run group PEd for BD across many different settings in the NHS (urban and more rural, teaching and non-teaching hospital, different parts of England) to explore its generalisability.

• To determine whether or not group PEd is clinically effective compared with group PS in increasing time to first bipolar relapse.

• To determine whether or not group PEd is clinically effective compared with group PS in relation to secondary outcomes of time to the next mania and depression relapse, severity of mania and depression symptoms, function, quality of life, health status and costs.

• To identify the barriers to, and potential solutions of barriers to, the implementation of effective group PEd.

Design

A single-blind RCT compared the effectiveness of:

• 21 weekly bipolar group PEd sessions delivered by two health professionals (nurse, psychiatrist, psychologist or occupational therapist) and a SUF, plus TAU

• 21 weekly unstructured bipolar group PS sessions delivered by two health professionals and a SUF plus TAU71 (see Table 3).

The trial was located in two centres in England (East Midlands and North West), with four clinical sites in each centre. Overall, 11 groups of PEd were compared with 11 groups of PS, with some sites running two waves of groups. The intervention was taken to different locations in each regional centre to improve access for those people who wanted to participate but for whom travelling distance was a practical barrier to engagement.

Consecutively eligible patients were individually randomised to either intervention, with stratification by clinical site and minimisation in terms of number of previous bipolar episodes (1–7, 8–19, ≥ 20). The latter was to control for the effect of rate of bipolar episodes: relapse is up to three times more likely in those with more than 20 episodes than in those with fewer than seven bipolar episodes. 72 There is also increasing evidence that psychological interventions may be particularly effective in people with fewer bipolar episodes or admissions to hospital,73,74 whereas one psychological intervention for unipolar depression, mindfulness CBT, may be more effective in people with three or more previous depression episodes than in those with fewer episodes. 75

Qualitative interviews were used to investigate the experiences of the group participants in order to identify facilitators of, and barriers to, the widespread delivery of the intervention across the NHS.

Setting and target population

The recruitment strategy was deliberately broad to ensure that targets were met and that the sample reflected a diversity of people with BD. Community mental health team bases at a number of NHS trusts located in two distinct geographical centres (East Midlands and North West) were encouraged to invite potentially relevant participants. The study was also promoted at a primary care level, with local family doctors being asked to display posters about the trial, and through SU-run local BD networks and the general media, allowing people to self-refer. The target population was patients with bipolar 1 or bipolar 2 affective disorder at increased risk of further relapse (defined as having had an episode in the last 24 months), as preventing relapse was the key aim of the intervention.

Baseline and outcome measures

At baseline interview, the SCID was used to assess sociodemographic features, the presence of axis 1 comorbid psychopathology,76,77 the presence of borderline or antisocial personality disorder78 and the number of previous bipolar episodes. 79 Participants were also asked if they had any preference as to which group they were allocated to.

Interventions

Randomisation and treatment allocation

Randomisation was conducted by a clinical trials unit that was given the participant information by the trial manager or by the trial administrator working under the direct supervision of the trial manager. Randomisation was undertaken only once a group of 20 participants in a wave had been identified at one geographical site close to where both groups would be held, as 10 was the minimum number of participants that would constitute a viable group size. The maximum number allocated to a group was 36. The allocation of participants to treatment was based on stochastic minimisation using number of previous episodes banded as 1–7, 8–19 and ≥ 20, and clinical site. The clinical trials unit conveyed the allocation to the trial manager, who then informed both the participants and the therapists conducting the group. The research assistants (RAs) conducting the interview assessments were blind to group allocation. All information regarding attendance at the groups was collected by the therapists so that RAs remained blind. Any instances of unblindings were recorded by the trial office. On the rare occasion that a RA was unblinded, an alternative RA at that site collected the remainder of that participant’s follow-up data. At baseline, before attending either group, participants were asked whether they had a preference for PEd or PS.

Ethics and research governance approval

The trial received national multicentre research ethics approval (09/H0408/33) and research governance approval and permission at each participating health service delivery organisation. The University of Nottingham sponsored and indemnified the trial. Each participant gave written informed consent to take part in the trial and, separately, gave written informed consent to take part in qualitative interviews, exploring in depth their experience of the trial. The trial was overseen by an independent Trial Steering Committee (TSC) and a Data Monitoring and Ethics Committee organised by the research investigators for the purpose. The members of the committees were drawn from outside the institutions in which the research team worked and were not involved in other currently funded research with the investigators to ensure their independence from the research team.

Sample size

When patients receive treatment in groups, interactions between patients may lead to correlation of outcomes of patients in the same group,89 sometimes referred to as cluster. As with cluster randomised trials, sample size calculation needs to consider the possibility of intracluster correlation. 64 As there are no previous trials of group interventions for BD that have considered clustering, we considered empirical evidence regarding the magnitude of clustering from cluster randomised trials. A previous trial90 found a very variable clustering effect, ranging from 0 to 0.27 with a median of 0, but this was from a small sample and so would be imprecisely estimated. We therefore assumed a small, but not zero, intracluster correlation coefficient (ICC) for group therapy of 5% for the purpose of sample size calculation. Based on the outcomes from the two previous Barcelona protocol PEd RCTs,61,62 a differential treatment effect of 0.22 was estimated (60% recurrence in the control group and 38% in the PEd group at 12-month follow-up, as one of these studies reported only 12-month follow-up). The power for 80% probability of detecting a difference at 0.05 level using two-tailed testing required 82 patients per arm. Assuming a mean group size of 18 and an ICC equal to 0.05, the design effect will be 1.85, giving a sample size of 152 in each arm. An 85% follow-up rate was assumed, giving a target of 360 SUs randomised, with 10 therapy groups planned in each treatment arm to achieve 80% power. During the initial waves of the study, the therapy group size was smaller than planned (approximately 14 participants per group owing to a slightly slower rate of accrual of each wave). In consultation with the Programme Steering Committee, the number of waves was increased from 10 to 11. Owing to the smaller therapy group size, with a reduced target sample size of 308, the study retained 80% power under the same assumptions about the ICC and follow-up rate.

Statistical methods

The quantitative trial outcome data included time-to-event data and longitudinal quantitative data that could be considered continuous.

Qualitative methods

A nested qualitative study was conducted to explore the experiences of group participants and identify facilitators of, and barriers to, the widespread delivery of the intervention. A maximum variance sample of SUs who had been randomised to either the PEd or PS groups was recruited, including participants who had taken part in the full programme and those who had dropped out at different stages. We sought to ensure that the final qualitative sample had a varied range of clinical and demographic variables (sex, age, number of previous bipolar episodes, geographical location of group), thereby likely to include the range of experiences from participants in the trial.

Service users were approached following the completion of the programme and invited to take part in a single semistructured interview at a time and location of their choice. A researcher then interviewed consenting participants. The interviews followed a topic guide that provided a flexible framework for questioning and explored a number of areas including participants’ experiences and views of the groups, their expectations and motivations for participating, group content and process, key points of learning or interest, and any issues or challenges experienced. The topic guide was developed following a literature search and discussion with the PARADES SU research group and amended following pilot interviews with members of this group.

The interviewing researcher used a combination of open questions to elicit free responses, and focused questions to probe and prompt as necessary. The interviews were digitally audio-recorded and transcribed verbatim by a professional transcriber. The interviewer checked the quality and accuracy of the transcripts against the original audio-recording. At this stage, any identifying information (e.g. names and places) was removed.

The analysis proceeded in parallel with the interviews. Coding was informed by accumulating data and continuing thematic analysis. 91,92 An inductive and emergent approach was taken to the analysis, guided by the study’s core questions about participants’ experiences of the groups. The interviewing researcher open-coded the interview transcripts as they became available using an inductive approach, and researchers from different backgrounds (health and clinical psychology, psychiatry and SU) discussed the developing coding scheme. All team members read and coded a selection of transcripts, and at least two members of the team read all of the transcripts. Hence, all authors undertook the interpretation and coding of data and agreed the themes through discussion. Using a multidisciplinary team is a recognised method for increasing the trustworthiness of the analysis. 93 The thematic categories identified in the initial interviews were then tested or explored in subsequent interviews in which disconfirmatory evidence was sought. Themes were continuously compared against the data using a constant comparative approach. 94 The final sample size was determined by thematic saturation: the point at which new data are no longer contributing to the findings owing to participants’ repetition of themes and comments. 95 At this point, data generation was terminated.

Of the 304 participants in the trial, 68 were identified for interview, of whom 37 consented and completed an interview (18 from the PEd arm and 19 from the PS arm). Of the 31 participants who were not interviewed, all initially agreed to be contacted, were in receipt of the participant information sheet and had the opportunity to discuss the interview with the interviewing researcher. Seven participants decided at this point not to take part, 20 were unable to be contacted and four did not attend their arranged interview. Twenty-eight interviews were completed face to face and eight interviews were completed by telephone. Owing to an unavoidable interruption, one participant was interviewed in two parts, initially by telephone and subsequently face to face. The mean duration of SU interviews was 47.2 minutes (range 29–103 minutes). Interviews were also conducted with health professional facilitators (n = 14/15) and SUFs (n = 9/10) to explore their experiences of the groups; however, these are not reported here.

Recruitment into the trial

Table 3 and Figure 1 show the flow of participants through the trial, with six waves of recruitment in the North West (at four different geographical sites) and five in the East Midlands (at four different geographical sites). The additional wave in the North West was required to meet the total sample size of 308 participants estimated by our power calculation; we recruited 304 participants, four participants short of the revised target, between June 2009 and January 2012. In each wave the total numbers of participants were individually randomised so that they had an equal chance of attending PEd or PS, and the procedure for recruitment, treatment and assessment were the same in both the North West and the East Midlands. Recruitment was carried out in two inner-city sites, three rural sites and three urban sites.

FIGURE 1.

Flow chart of participants into the study from all sites and waves combined.

Three hundred and four (84.7%) participants were recruited into the study. In addition, 27 people who otherwise met the inclusion criteria of the study refused to give consent and six further people were unable to commit to the groups for practical reasons (e.g. time off work); 22 people were either not contactable or did not meet inclusion criteria on assessment. Two hundred and forty-three (79.9%) people completed the first follow-up point after completing treatment; and 204 (67.1%) completed interview assessments at the 96-week follow-up. Three participants were missing baseline assessments, although they had an assessment of eligibility and consented to take part. A further four subjects had baseline assessments taken after randomisation, so these data are not included in the summary statistics and analyses. Data on all bipolar relapses, mania and depression relapses were completed using primary care and secondary care records.

Four participants died during the follow-up period of the study from natural causes that were deemed by an independent TSC and Data Monitoring and Ethics Committee to be unrelated to the interventions or procedures in the trial.

Baseline characteristics of treatment groups

Table 4 shows the baseline characteristics by treatment group. The study recruited a predominantly white British sample that contained more women than men, and a wide age range from 21 to 73 years. There were some baseline differences between the groups, with the PEd group being, on average, slightly younger [PEd mean age = 44.2 years, standard deviation (SD) = 11.1 years; PS mean age = 46.5 years, SD = 11.4 years] and more likely to be married and including fewer people with bipolar 1 disorder. Only 24 (7.9%) participants were not white British; 10 were black, seven were of mixed race, three were South Asian, three were white non-British and one was Chinese. Ethnicity was unavailable for two participants. Also of note was that the majority of participants had children. Just over one-quarter of participants were employed. Forty per cent of the sample expressed no preference for either treatment group, but more people expressed a preference for PEd over PS, although all were willing to be randomised to either group. Table 4 also shows that the groups did not differ on psychiatric comorbidity, with around 70% reporting psychotic symptoms in their lifetime, 40% describing a current anxiety disorder, just over 40% reporting an alcohol abuse or dependence problem in their lifetime and just under 10% with a borderline or antisocial personality disorder. Nearly three-quarters of the participants in both groups were taking mood-stabilising medication, although the proportion doing so was slightly higher in the PS group than in the PEd group. Similar numbers from both groups were receiving antipsychotic (54–55%) and antidepressant (46–47%) medication.

Tables 5–7 summarise the trial secondary outcome measures at baseline and follow-up assessments. These show that, symptomatically and functionally, there were no clinically important differences between the groups at baseline. Symptomatically, participants scored on average just below the clinical threshold for observed depression symptoms on the HAM-D, slightly above the minimum clinical threshold for depression and anxiety symptoms on self-rated measures, with more severe anxiety than depression, and minimal mania symptoms at baseline. Consistent with the inclusion criteria, although a few participants scored above the symptomatic threshold score for depression for 1 week, the duration of symptoms meant that they did not meet the criteria for a major depressive episode. No participants met the criteria for a mania-type episode at baseline. On average, participants showed no more than slight impairment in social and occupational functioning overall (SOFAS and overall SAS) and across all domains (performance, interpersonal, friction and dependency).

Attendance at the groups

The distribution of the number of sessions attended by PEd and PS group is presented in Table 8. The proportions who did not attend any group sessions of PEd and PS were similar (13% and 15%, respectively). The median number of sessions attended was 14 in the PEd group, compared with nine in the PS group. Overall, rates of attendance were higher for PEd groups than for the PS groups (Mann–Whitney U test z = 2.23; p = 0.026). With evidence of overdispersion, a negative binomial model for count data was used to investigate selected baseline factors that could potentially influence attendance. The following variables were included in the model (p-values given in brackets): sex (p = 0.888); number of previous episodes (1–7 vs. ≥ 20, p = 0.761; 8–19 vs. ≥ 20, p = 0.926); bipolar 1 disorder versus bipolar 2 disorder status (p = 0.289); group preference (none vs. prefer PEd; p = 0.779; prefer PS vs. prefer PEd p = 0.517); arm (p = 0.223); and wave (p-values ranged from 0.415 to 0.974). As shown from the above, none of these differences was significant.

Response rates and missing data

Logistic models with a random effect for wave and arm (nested within wave) to take account of therapy group clustering effects were fitted to examine factors potentially predictive of missing primary depression outcome data at weeks 32 (over the treatment period) and 96 (over the whole follow-up period). The same covariates (i.e. sex, number of previous episodes, arm, baseline mania score and baseline depression score) were used for each model. None of these factors was predictive of non-response on the SCID LIFE at weeks 32 and 96. Sex, number of previous episodes and baseline score were specified as covariates in the statistical analysis plan for the LME analyses to increase the precision of estimates. No follow-up SCID LIFE data were available for a total of 14 (9%) participants in the PEd arm, compared with 21 (14%) in the PS arm.

Reliability and quality assurance

Ten unblindings occurred because the participant divulged their group allocation to their RA (three in PEd group and seven in the PS group). In seven cases this occurred during follow-up interview. One participant wrote their group allocation on self-completion paperwork, one telephoned the RA to change their assessment date and divulged their group allocation, and one RA overheard a conversation from the therapy team that unblinded them. When unblinding occurred during an interview, an alternative RA rated the SCID LIFE codings to avoid bias. For all seven participants, an unblinded RA was assigned for their remaining follow-ups.

An inter-rater reliability (IRR) study was carried out of the weekly SCID LIFE assessment in which scores of 1–6 were assigned for each of mania and depression. An ICC was determined for the depression and mania scores separately and percentile CIs were determined using a bootstrap resampling by subject. Six raters made 292 assessments of 17 subjects, with a mean of 17.2 assessments of each subject and 2.35 assessments per time point. The ICC was 0.71 (95% CI 0.64 to 0.85) for depression scores and 0.57 (95% CI 0.40 to 0.83) for mania scores, indicating a moderate level of inter-rater agreement. 96 There were 24 bipolar relapses in the IRR exercise (18 with depression-type and six with mania-type relapses). For the combined depression and mania outcomes, the ICC was 0.56 (95% CI 0.22 to 0.89).

Episode recurrence

Of the 153 participants in the PEd arm, 89 (58%) had a recurrence of an episode of depression or mania type during the 96-week follow-up. In the PS arm, 98 (65%) had a recurrent episode. Of the first recurrences, 75% in the PEd arm and 63% in the PS arm were depressive episodes. When the Cox models with and without robust standard errors were compared, there was no evidence of a clustering effect (as the ratio of the square of the standard errors of the models with and without a cluster term was less than one), and so the latter are presented. A Kaplan–Meier plot for time to first recurrence in the two groups is shown in Figure 2. The estimated median time to first bipolar relapse was 67.1 (95% CI 37.3 to 90.9) weeks in the PEd group, compared with 48.0 (95% CI 30.6 to 65.9) weeks in the PS group, a difference of 19.1 weeks. When the Cox model was fitted, the adjusted HR was 0.83 (95% CI 0.62 to 1.11; LRT p = 0.217). Male participants tended to show a greater improvement than female participants in terms of time to relapse (HR 0.74; p = 0.052). There was also a non-significant (p = 0.094) trend for greater beneficial effects among SUs, with few bipolar episodes prior to joining the trial (8–19 vs. ≥ 20, HR 0.74; p = 0.073; 1–7 vs. ≥ 20 HR, 0.67; p = 0.105). The moderating effect of episode pre trial is considered below.

FIGURE 2.

Kaplan–Meier estimates of time to first depression- or mania-type bipolar episode.

In a sensitivity analysis we additionally controlled for baseline differences between the groups in terms of the proportion of bipolar 2 versus bipolar 1, marital status and taking mood stabilisers (lithium, sodium valproate, carbamazepine and/or lamotrigine). When these covariates were included together, the intervention effect on relapse remained non-significant (LRT p = 0.294).

The proportion of participants with a first mania-type episode was lower in the PEd group than in the PS group [n = 39 (25%) vs. n = 53 (35%)]. As so few participants relapsed with mania over 2 years in both groups, we present the 15th centile time to relapse rather than the median time to relapse, which was 44.6 (95% CI 27.3 to 73.6) weeks after baseline in the PEd group and 21.7 (95% CI 13.1 to 39.3) weeks after baseline in the PS group. Figure 3 shows the Kaplan–Meier estimates for the two groups. There was evidence that the time to first relapse was delayed in the PEd arm compared with the PS arm (HR 0.66, 95% CI 0.44 to 1.002; LRT p = 0.049). When bipolar 1, marital status and the proportion taking mood stabilisers were added in a sensitivity analysis, the HR increased to 0.74 (95% CI 0.48 to 1.13; LRT p = 0.165); bipolar 1 and taking mood stabilisers increased the HR, whereas married/living as married decreased the risk of relapse, but all three were non-significant.

FIGURE 3.

Kaplan–Meier estimates of time to first mania-type bipolar episode.

A total of 80 (52%) subjects in the PEd arm had a first depressive episode, compared with 82 (54%) in the PS arm. As so few participants relapsed with depression, we report the 25th centile time to relapse, which was 22.0 (95% CI 14.9 to 29.3) weeks after baseline in the PEd group and 19.4 (95% CI 14.6 to 27.6) weeks after baseline in the PS group. The Kaplan–Meier estimates comparing the two groups are shown in Figure 4. There was no evidence of a difference between the groups, with an estimated HR of 0.96 (95% CI 0.70 to 1.31; LRT p = 0.784). Adjusting for bipolar 1, marital status and the proportion taking mood stabilisers, covariates had little impact on the treatment effect (LRT p = 0.682).

FIGURE 4.

Kaplan–Meier estimates of time to first depression-type bipolar episode.

Post-treatment benefit

Theoretically, PEd may be more effective after the treatment period than PS because it explicitly develops a staying well plan for each participant, whereas both PEd and PS provide information and mutual support during the treatment period when the groups meet. PEd aims to teach participants in the group specific evidence-based methods to recognise and respond to the early stages of mania and depression relapse compared with PS with likely impact in the follow-up period after treatment is completed. Therefore, we decided to run an analysis separating the effects of the interventions during the treatment period and then those during the follow-up period. On inspection of the data on time to first relapse, we decided to carry out this analysis on time to the next bipolar episode and time to the next mania episode but not on time to the next depression episode, as overall it was clear that there were no differences between the two groups. For each wave we calculated the duration in weeks between the earliest randomisation date and the final date of the last session (i.e. session 21). The maximum was 26 weeks. We therefore used ≥ 27 weeks post randomisation as the start point for ‘off treatment’. A Kaplan–Meier plot and Cox analysis for the primary outcome (time to first depressive or mania relapse) were performed on data < 27 weeks. Excluding one subject who was censored at week 25, the data for the remaining 303 participants were then analysed from 27 weeks onwards (the ‘post-treatment’ period) up to 96 weeks. All failure (when a participant has a first relapse in the specified period) and censor times were rescaled by subtracting 27 weeks, so that the time origin was set to zero. The same covariates as for the main analysis were used for both Cox models. This was repeated for the mania outcome.

In the ‘on-treatment’ period, a total of 51 (33%) in the PEd arm had a first bipolar (mania-type or depression) relapse, compared with 60 (40%) in the PS arm. There was no evidence of a difference between the groups, with an estimated HR of 0.81 (95% CI 0.56 to 1.19; LRT p = 0.286). In the ‘off-treatment’ period, a total of 75 (49%) in the PEd arm had a ‘first’ bipolar relapse, compared with 81 (54%) on the PS arm. There was no evidence of a difference between the groups, with an estimated HR of 0.91 (95% CI 0.66 to 1.26; LRT p = 0.578).

Considering mania alone, during the in-treatment period, a total of 14 (9%) in the PEd arm had a first relapse, compared with 27 (18%) in the PS arm. There was evidence of a benefit for the PEd group, with an estimated HR of 0.49 (95% CI 0.26 to 0.94; LRT p = 0.028). In the off-treatment period, a total of 30 (20%) in the PEd arm had a first mania relapse, compared with 38 (25%) in the PS arm. There was no evidence of delaying relapse in the PEd arm, with an estimated HR of 0.75 (95% CI 0.47 to 1.22; LRT p = 0.246), suggesting that the differences between the groups largely emerged during the treatment period and did not continue to widen during the follow-up period.

Symptom and functioning outcomes

Table 5 gives summary statistics for the secondary outcome interview-measured symptom measures. Overall, 205 participants (67.4%) completed these measures by 96 weeks, with increasing attrition over follow-up from baseline. The mean depression and mania scores were based on each participant’s mean score over the previous 16 weeks (last 4 weeks for baseline). As shown in Table 5, the mean scores on SCID LIFE depression, mania, MAS, HAM-D and HADS depression and Hospital Anxiety and Depression Scale – anxiety subscale (HADS-A) were mostly flat over time. However, there was a small increase in the SCID LIFE depression score from 1.84 at 16 weeks to a maximum of 2.18 at week 32 when the group finished, and a fall back to 1.89 in the PEd arm. These changes were not seen in the PS group. However, all of these changes are too small to be of clinical importance and the increase was not mirrored in the HAM-D during the same time period.

Table 9 shows the LME analyses for observer-rated depression and mania symptom outcomes, with no statistically significant interaction between treatment arm and time, indicating that there was no significant difference between the two treatment groups in rate of improvement over time. There was little, if any, evidence of variation between therapy groups, with the ICC for therapy group within wave being either zero or negligible.

Tables 6 and 9 show the effects of PEd and PS on interview-rated function (SOFAS and SAS) over time with analysis using the LMEs. Overall, 200 (65.8%) participants completed these measures by 96 weeks with increasing attrition over time. Participants in the PEd group improved faster on the SAS interpersonal domain than those in the PS group (difference in gradient –0.017, 95% CI –0.030 to –0.004; p = 0.012). The estimated difference in means between PEd and PS was –0.037 (95% CI –0.156 to 0.082; p = 0.544). On fitting models without the treatment by time interaction to the remaining SOFAS and SAS (overall, friction and dependency) outcomes, there was no evidence of a difference between the PEd- and PS-adjusted means or in gradients. There were also no significant changes in function over time irrespective of treatment allocation.

Tables 7 and 10 show the results for the self-rated clinical outcomes. Unfortunately, only 141 (46.7%) and 103 (33.9%) of participants completed the secondary outcome measures by 32 and 96 weeks, respectively. In the analysis using the LMEs, there was a non-significant trend for a time by treatment group interaction (LRT p = 0.084) on HADS-A scores. From inspection of Table 7, this would appear to reflect the fact that levels of anxiety in the immediate post-treatment period were higher in the PS group than in the PE group, but equalised at the 96-week time point. Other secondary outcome measures, namely HADS depression score and scores on the SF-12 mental and physical component scales, did not differ between treatment arms. The effect of treatment on HADS-A was small, with a rise of just under 1 point in anxiety from baseline to 32 and 64 weeks in the PS group, and an even smaller drop in HADS-A at these time points in the PEd group, but in both groups these changes returned approximately to baseline scores by the final follow-up point. Therefore, these effects are probably not clinically important. There was rather more evidence of variation between therapy group, with the ICC for therapy group ranging from 0 to 0.048.

Treatment moderators

We examined two prespecified treatment moderators: number of previous episodes (1–7, 8–19 and ≥ 20 episodes minimised in the randomisation as part of the design) and participant treatment preference at baseline. On adding two interaction terms for the number of previous bipolar episodes (1–7 and 8–19 relative to the ≥ 20 reference group) by treatment arm to the primary outcome model, the overall interaction was significant [χ² = 6.80, degrees of freedom (df) = 2; p = 0.034]. The subgroup-specific HRs of the treatment effect were 0.28 (95% CI 0.12 to 0.68) for 1–7 previous episodes, 0.86 (95% CI 0.50 to 1.49) for 8–19 previous episodes and 1.01 (95% CI 0.70 to 1.46) for ≥ 20 previous episodes. Figure 5 shows the Kaplan–Meier for the three episode groups split by arm; those in the PEd group with 1–7 episodes had the longest delay in time to episode, whereas those with the same number of episodes in the PS arm had the shortest.

FIGURE 5.

Kaplan–Meier estimates of time to first mania-type or depressive bipolar episode by group and number of previous bipolar episodes.

Similar plots for time to first mania- and depressive-type episodes are shown in Figures 6 and 7. Figure 7 shows that in the sample with 1–7 episodes, time to depression relapse was significantly delayed in the PEd group compared with the PS group (HR 0.24, 95% CI 0.08 to 0.68; p = 0.008), with five (22%) participants in the PEd group experiencing a depression relapse by 96 weeks, compared with 11 (61%) in the PS group. There were no other statistically significant interactions between number of episodes and time to depression relapse and none at all between number of episodes and time to mania-type relapse.

FIGURE 6.

Kaplan–Meier estimates of time to first mania-type bipolar episode by group and number of previous bipolar episodes.

FIGURE 7.

Kaplan–Meier estimates of time to first depressive bipolar episode by group and number of previous bipolar episodes.

The interaction between treatment and treatment preference (no preference or prefer PS vs. the reference category of ‘would prefer PEd’) was not statistically significant (χ²= 1.95, df = 2; p = 0.378) when added to the primary outcome model.

Knowledge is power

Participants described having gained greater understanding about BD, the different forms it could take and ways of managing it, both in general and as it applied to them. Participants felt that this was an important outcome in its own right and that accessing information was a good reason for taking part in the trial. In particular, participants described how useful it had been to learn how the condition could affect them and their day-to-day living. Spending time reflecting on previous events (and particularly, for the PEd group, life charting; recording patterns of life events and mood experiences over time) helped them to understand their condition in different and more helpful ways. This included understanding the biological and genetic aspects of the illness, the function of medication and the inter-relationships of symptoms. Participants valued identifying triggers and early warning signs that were particular to them. Some described how this had led to their feeling more confident in managing future problems and talked about ways in which they had made use of the action plans in everyday life:

On holidays in the past I’ve had to come home because I’ve been so bad. This is way before the study. And now I can go away and think well if anything happens I can work on it, rather than get scared and just sort of abandon ship, and come to my home and curl up in bed . . . I have a little card which is in my handwriting about the early warning signs and strategies, just in my wallet, so I take that with me every day everywhere I go and it’s not there to show off it’s, well nobody sees it do they, well actually I have shown it to quite a few people and it’s a constant reminder to me.

SU053 PEd

Although the PS intervention had not been designed to cover the prescribed topics of the PEd, participants from the support groups also described that information exchange and learning was prevalent within their groups. Thus, many of the same topics (e.g. medication and early warning signs) were discussed in the PS and PEd groups:

We learnt probably the same if not more of what they were just teaching just from our experiences from the other people with their experience of the mental health sector and erm . . . using the services and also just their own personal experience how they get out of scrapes and how they, how they bring themselves out of depression or stuff.

SU102 PS

However, some participants commented that these groups could feel unstructured and lacking in focus, which meant they were more vulnerable to dominant group members:

I mean by the fifth week there was only about five or six of us left, so probably out of that group, probably only one or two you know that, erm . . . there was a couple of other people that I really wanted to listen to and I couldn’t because this other person took the whole hour-and-a-half up and I suppose that upset me.

SU153 PS

The newly acquired knowledge and understanding was described as being useful when interacting with others, such as carers. The materials were seen as useful resources to share and also helped individuals to be more confident to start conversations and explain the disorder to friends and family. Participants also talked about how the expertise they had acquired made them more confident about understanding and using medical ‘jargon’ and interacting with health-care professionals. Ultimately, having information that was tailored to them enabled individuals to feel that they had greater control over managing their condition in ways other than taking medication, and this was seen as empowering:

I’ve come down, and like halved my dose since . . . since I was on the programme. And I think that was contrary to the previous . . . you know, the line, if you like, for treating bipolar. It’s always definitely just drugs and that’s it and I think psychological therapy has a massive impact and I think having that information about what you can do to help yourself and to help manage your illness is power to you really.

SU052 PEd

People like me

Participants described having often felt isolated. Having BD had had an impact on their relationships as well as their ability to work, and people often felt unable to share their experiences and thoughts about the condition with friends, family members and colleagues. Many described having withdrawn as a way of coping or having lost confidence in social interaction, which had led some to become increasingly removed from other people. Many did not have regular contact with other people with BD, and meeting others who had similar experiences was raised as an important reason to take part and stay involved in the trial. Being able to associate with a group and share experiences with people they knew would understand was seen as comforting:

You feel very lonely when you go through this sort of experience, so if, you know, there is other people, you know, that have had bad times like yourself it helps to give you a bit of relief that you are not on your own.

SU004 PE

For some it was a very reassuring experience to learn that they were not unusual in the struggles they had encountered and to learn about how others managed their symptoms. It also allowed them to normalise some of their experiences, understand them and realise, therefore, that they were not abnormal. For some, attending the group was the first time that they had knowingly met someone with BD:

I think the first time to sit down in a room full of people who have bipolar and listen to what they are saying and how their condition affects them. You are thinking ‘wow, I am not alone . . . [laughs] it’s not just me’. So yes it was very beneficial at the time.

SU058 PE

Some participants described feeling wary of meeting people with BD and feeling fearful of being viewed as part of this group. Meeting and coming to know individuals helped them reconcile their own stigmatising views of people with BD and hence be more accepting of their own diagnosis:

If you are actually diagnosed and you are put with a group of people who are happy to talk about it, and their feelings, how they cope and actually they are normal human beings. You know, we had the good, the bad and the ugly there, so we had everyone, very glamorous young lady, and . . . [participant’s name] was very hippy. And a very big fat old man who was, well, there was just everybody and you didn’t, I didn’t feel alone. I felt for the first time I had somebody who really understood me, who didn’t judge me and, erm, and I wasn’t odd, so that is vitally important.

SU102 PS

Participants described feeling inspired by these encounters. It was encouraging to meet others who were able to function well and had recovered from severe episodes, particularly for individuals who had been diagnosed more recently:

I found that quite inspiring to know there were people who had been sectioned, had been full-blown woooo, and now have recovered to the degree that they can hold down some sort of work.

SU001 PEd

However, at the same time, this could be challenging:

Initially with the course, I’ll be really honest, I found it quite upsetting. And my friend said to me ‘I thought if anything it would comfort you. That you’d met other people that were in a similar situation, or actually worse’ . . . I found it really upset me to see all these nice normal people, suffering. And I know how it feels . . . I’d be alright in the group, and then I’d come home and get really upset, having heard people’s stories, and what they’d been through, and how relationships had been damaged, and they’d turned to drugs, and things like that. It just really . . . I sort of wanted to make it better and you can’t.

SU001 PEd

Taking part took effort in terms of both the amount of time participants had to dedicate to the 21-week programme and the emotional demands of thinking about and sharing dark times. In particular, some found it very challenging to create their life chart and talk about how they had felt and what they had done during episodes of illness. Although most felt safe and supported to do so, some individuals found the focus on difficult times too demanding and struggled to manage the feelings they were left with at the end of the session:

I left that group feeling worse . . . there was nothing positive that came along with it to me, it was negative, negative, negative.

SU153 PS (dropped out after four sessions)

As the intervention lasted 21 weeks, it is not surprising that some individuals experienced life events or illness during this period, and the groups could feel too demanding at these times:

I just couldn’t sit down for long enough to take in the subject matter. It wasn’t the subject matter or the way it was presented because it was presented very well and in an orderly and sensitive manner. What it was my own illness basically prevented me from attending all the sessions.

SU057 PEd

The relationships formed during the course of the groups were, for many, extremely important and even lasting; it was evident that this support function was just as important for participants in the PEd groups as for those in the PS groups:

I would say that what’s happened for us afterwards, for those that have retained the connection . . . it’s been almost as important as the course itself like . . . you’ve had that knowledge that you bring to each other, the relationships that you’ve established . . . [it’s] more important to staying stable than . . . that pile of hand-outs . . . these relationships hopefully are sustaining.

SU051 PEd

A novel feature of both interventions was that SUs co-facilitated the sessions. Participants found this extremely valuable and felt understood in a way that might not have been possible otherwise:

[SUF] was the only one that could have understood a couple of the symptoms I have . . . She had exactly the same symptom, and she went to the doctor loads of times and the doctor just brushed it off like ‘that is not a symptom of bipolar disorder’. And [health professional facilitator] had never heard of it, you know . . . and I said [SUF] I get exactly the same thing and I have been trying to explain it too and I can never describe it . . . and so without [SUF] there I would never really have been able to explore that or talk about that, or understand that in any way.

SU154 PS

They appreciated that at times SUFs might also feel challenged by the demands of the group and the role:

I think it was a hard role for her because, like me talking this morning, I’ve found it very hard now because it brings memories back, not about the good times, but about the bad times. I’m sure she had some, well she [said], you know, ‘I’m not too good this week’ and we were all very appreciative that she was able to share that and it made us understand she knows where we are coming from.

SU053 PEd

It appeared that what made the groups useful and engaging was the combination of gaining knowledge and techniques about BD alongside sharing experiences and forming part of a group with other SUs:

I’m not going too far when I say it has been life-changing, it really has. And I don’t think it would have been that massive a life-changing experience if it was just a set of lectures in a lecture theatre with no participation from many of the people who were listening . . . I think it’s that combination of having the right information given to you so that you can understand what’s happening, but also teach you some tools and techniques to be able to manage it properly as well as having all of the support of the people who have actually been through it.

SU052 PEd

Summary of findings

The data show that both group PEd and group PS are feasible to deliver in routine NHS care across eight different inner-city urban and rural clinical sites in two regions of England. PEd was associated with improved attendance rates compared with PS, with the qualitative data indicating that participants who dropped out of PS did not like the lack of structure of the groups and the tendency of some individuals to dominate the sessions. Overall, the PEd and PS groups did not differ in terms of the primary outcome variable, namely time to the next bipolar relapse. However, there was evidence of a large effect favouring PEd over PS in delaying the time to relapse (HR 0.28, 95% CI 0.12 to 0.68) in the 13% (39/304) of participants with fewer than eight previous bipolar episodes. This result should be treated with caution, as only 13% of the sample had fewer than eight previous episodes. The level of attendance at groups, sex, recruitment centre and preference for one group over another at baseline did not significantly influence time to next bipolar relapse or differences in this outcome between the groups.

Among the secondary outcome variables, two favoured PEd over PS and none favoured PS over PEd. There was a significant but marginal delay in time to next mania episode and evidence of improved interpersonal function. However, the delay in time to next mania episode occurred during the treatment period with no further increases in effectiveness in the follow-up period after treatment, and was attenuated once baseline differences in the proportion with BD 1 to BD 2, marital status and treatment with mood-stabiliser medication were considered. Although there is little doubt that mood-stabilising drugs, such as lithium, sodium valproate and carbamazepine, reduce the risk of relapse with mania,57 it is less clear whether marital status or having BD 1 or BD 2 would necessarily change time to the next mania relapse (although BD 1 and BD 2 status would certainly be expected to affect the severity and duration of mania-type episodes once they had occurred). Therefore, controlling for these post hoc baseline differences may underestimate the relative effectiveness of PEd versus PS. There were no significant differences in time to the next depression relapse, although the proportion who relapsed with depression was more than halved in people with one to seven previous bipolar episodes compared with those with eight or more previous episodes. There were also no improvements in mania or depression symptoms, function (apart from interpersonal function) and self-rated physical health. Qualitative data indicated that, in both PEd and PS, participants valued gaining knowledge about BD and feeling less socially isolated by discussing their condition with other people who had BD. They also gained the confidence and knowledge to communicate with their friends, family and health-care professionals so that they could express their wishes and perspectives related to their condition and how it affected their life. People with fewer bipolar episodes and a more recent diagnosis of BD felt encouraged by meeting people with many previous episodes who were functioning well and leading productive lives. In terms of adverse effects, in both groups some people found the content and the reflection on the past distressing, particularly those who felt isolated or were experiencing life difficulties at that time.

Overall, there are no differences in effectiveness between PEd and PS groups except for those people who had fewer bipolar episodes. If only one treatment was to be offered, there would be some advantages in offering PEd over PS in terms of increased acceptability, some better clinical outcomes and better outcomes in people with few previous bipolar episodes. However, on the basis of this RCT, compared with an active control group, there is little evidence that PEd is more clinically effective than PS, except in people who have had up to seven previous bipolar episodes.

Strengths and limitations

The study achieved its recruitment target, with a shortfall of only four participants. It was designed to take clustering effects into account, thereby increasing generalisability. There was no evidence of a clustering effect of therapy group for the primary outcome. The study has, therefore, greater power than planned, as the sample size calculation assumed an ICC for group therapy of 5%. However, we aimed to detect a large differential treatment effect (0.22) found by Colom et al. 61,62 that was considered to be an outlier in a recent meta-analysis of RCTs of individual and group PEd versus TAU or group PS. 32 It is possible that a small differential treatment effect was not detected. Given that the study was designed as a superiority trial rather than a non-inferiority trial, and that there is no TAU group, we cannot conclude that group PEd and group PS are equally effective in BD on the basis of this RCT alone.

There was a high follow-up rate at the end of the treatment period, as reported in other group and individual psychological treatment trials for the long-term management of BD,73,80,97 and an acceptable follow-up for all interviewer-rated measures of 67% at 96 weeks. Few studies have reported interview follow-up rates at 96 weeks, but the follow-up rate in the current study at 12 and 18 months is similar to that in previously reported UK studies of psychological treatments using similar methods. 72,79 There was complete follow-up for the primary outcome variable. The IRR for measuring mania-type and depression symptoms scores was only moderate, but we believe that this is an acceptable level of agreement given that these were ordinal judgements over six levels of severity rather than binary assessments. Randomisation was carried out by a clinical trials unit, and there were only 10 instances of unblinding. Unusually for a psychological treatment study, there was very close matching of the PEd and PS groups for therapists, the existence of manuals and duration of treatment, so it was possible to blind assessors of outcome. Unblindings were rare, and in all cases independent unblinded assessment of outcome was achieved. Assessment at baseline occurred within a few weeks before the groups started, with the result that none of the participants started the groups while in a bipolar episode. The trial was registered and the protocol for the trial and its analysis has been published. 71

Unlike some previous studies of treatment in BD using time to next bipolar episode as the primary outcome, the study minimised at randomisation for the number of previous bipolar episodes that the participant had experienced. This is important, as time to next bipolar episode can increase threefold in people with ≥ 20 previous episodes, compared with those with ≤ seven episodes. 72 Not controlling for the natural history of illness is one of the most common failings of RCTs in mental health. 97 A further strength of the study is that the effects on function were measured in addition to symptoms and relapse; improvements in function are prioritised in the personal recovery of people with BD. 98 The analysis of the continuous data used a linear mixed modelling approach allowing for clustering by therapy group, unlike almost all previous group PEd trials in BD. We also explored the effects of missing non-response over time and baseline differences in the groups that might be clinically important. Overall, the study had high internal validity, and many methodological strengths compared with previous RCTs of group PEd and other psychological treatment in BD; it is also the largest RCT of its type.

The study recruited efficiently, suggesting that the participants welcomed the possibility of taking part in a study offering two types of group psychological interventions, including 85% of the sample who were contacted about taking part in the study following referral. Only 8% of participants withdrew their consent once they had received a full explanation of the study, and only 2% withdrew because they were unable to attend the groups. Overall, 22% of the sample never started or attended only one session of psychological treatment in this study, in comparison with 60% not attending or attending only one session of psychological treatment from the Improving Access to Psychological Treatment services routinely delivering NHS psychological treatment to adults with depression and anxiety in England. 99

The study successfully recruited from two regions in England, from inner-city, urban and rural sites, and from direct referral by primary care and secondary care and through self-referral. Participants were, on average, aged in their mid-forties, typical of samples in pragmatic RCTs in BD in the UK. 72,79,100 Similar to other samples recruiting a combination of participants with either BD 1 and BD 2, there was a slight excess of female participants in the sample. Also typical of previous studies in this field was that only around 40% of participants were currently married or cohabiting. The sample was predominantly white British and slightly above the average for England according to the 2011 census. 101 In particular, there were only three (1.0%) participants of South Asian origin. Similar to most previous studies of psychological treatments with a mixed BD 1 and BD 2 sample, 80% of participants had BD 1 and 80% of these had a lifetime history of psychotic symptoms. The current sample was similar to that recruited for another large UK study of psychological treatment in BD by Scott et al. 79 in the lifetime prevalence of anxiety disorder, alcohol and drug abuse or dependence and borderline or antisocial personality disorder. Overall, the representativeness of the sites and the participants, together with the study’s wide inclusion and exclusion criteria for comorbidity, suggests that it has high external validity and generalisability to NHS practice in England, except in relation to people of non-white British origin.

The main limitation of the study in terms of making inferences to routine clinical care and to other studies is the design, as there was no TAU group who were not receiving psychological treatment. This trial was always intended to be a rigorous examination of the clinical effectiveness of running group PEd using the structured Barcelona approach adapted to a UK NHS (public service) context compared with another well-matched and commonly available group psychological treatment alternative, rather than an examination of the clinical effectiveness of a group psychological approach to BD compared with not offering any structured psychological treatment. In the current study, the control group was an attentional control receiving the most widely available alternative group psychological treatment in the UK, namely PS. The control group was designed to control for everything except the content and structure of the intervention; it had the same aim of treatment, to develop a stay well plan, as well as the same duration of treatment sessions, the same number of treatment sessions, the use of therapists and the use of a manual. In contrast, the support group control for group PEd in the first two 21-session group PEd RCTs reported33,61,62 was not given the same purpose, but it did control for meeting as a group for the same number of sessions, the same duration of sessions and the number of therapists. In these studies, the control group did not necessarily discuss BD at all during their group sessions, and therefore the authors did not control for an expectancy of beneficial effect on outcome. 102

Another important limitation in psychological treatment studies is the experience level of the therapists offering the treatment. In both treatment arms, we utilised at least one therapist who had experience of delivering psychological treatment and running group interventions, and at least two therapists who had a great deal of experience of managing BD. The SUF was usually relatively new to the role. All therapists and SUFs were regularly supervised by experts who had extensive experience in conducting RCTs of psychological interventions in BD, including those involving PEd and psychological support. However, none of the therapists was experienced at delivering these interventions, and this relative inexperience could be a criticism of the study. On the other hand, a strength of the study was that none of the research team had developed these interventions and were not advocates for them, so they were independent researchers who were also experienced in running multicentre RCTs of psychological interventions in BD.

Our analysis of the content of the PS group shows that it had a clear focus on the task, namely to develop a stay well plan for participants. In terms of content, there was a considerable overlap, with participants in both groups almost always receiving sessions on causes and triggers of BD, mania, depression, prescribed drugs and alternative therapies, regularity of habits, stress control techniques and problem-solving strategies. In roughly half of the PS groups, the participants also discussed a method of preventing relapse into mania and depression episodes that has some evidence of effectiveness,34,72 but in one session only, rather than five sessions, as for PEd. Thus, the PS control group covered some of the content that is thought to be effective in reducing bipolar relapse,103 albeit not necessarily in a structured or intensive format . Qualitative data from the participants suggest that their experience in both groups was powerful in relation to information-giving and receiving support from people with similar experiences. There was no evidence of contamination between the groups of participants in the group PS who did not come together in any forum, or between facilitators, who allowed the participants to decide the content of their sessions and played a passive role in running the groups. Evidence in support of some effectiveness of the PS interventions is that the relapse rate for both PEd and TAU groups reported by Bond and Anderson32 was 60–70% overall at 1 year; in the current study, the relapse rate in the PS group was only 53% at 12 months and 65% at 24 months. Caution should therefore be applied before assuming that the rates of relapse in this study in the PS group are the same as would be found for TAU in future meta-analysis including this RCT. The results in the current RCT might underestimate the effectiveness of group PEd compared with TAU in NHS practice.

The PS group also differed in some other important ways from PS groups that are available routinely in NHS and non-NHS settings in the UK. PS groups are usually run by peers with BD rather than by health professionals. They often have an open format, so that new participants can join at any session, and sometimes are potentially long term and do not have any fixed number of sessions that a participant attends. They often do not have any fixed agenda or aim other than to broadly educate and support the people with BD who participate. Therefore, the results in this study may not be generalisable to other PS groups, particularly if these groups are open to new participants, do not have the same aim of developing a stay well plan and are open ended, and if the facilitators are not provided with peer and group supervision.

A limitation of the study may be the relatively poor attendance at the treatment groups. This, however, is, however, likely to be a reflection of what might happen in routine clinical care, given the pragmatic design of the RCT. Although attendance rates at sessions did not account for differences in outcome between the groups, it might have compromised the continuity in learning for those participants who attended intermittently and for the rest of the group as a whole in both treatment arms.

Finally, the study recruited people with only low levels of depression and mania symptoms at baseline, so floor effects limited a determination of whether or not group PEd may be effective against symptoms of mania or depression compared with PS groups. However, the study was designed primarily to test the effectiveness of group PEd for the prevention of future bipolar relapse and therefore deliberately did not recruit participants who were in a bipolar episode. Those in the sample did show mild anxiety, and around 40% of the sample had had anxiety disorders in the last month and over half had them in their lifetime. There was a trend towards a benefit of group PEd versus group PS on self-rated anxiety symptoms in the immediate post-treatment period, but this assessment was compromised by the low return rate of the HADS. The low return rate by participants who were given the option of returning questionnaires by post is a substantial limitation when considering self-rated outcomes in the current study.

Comparison with other psychological long-term treatment trials in bipolar disorder

Compared with the first studies of 21-session group PEd,33,61,62 the current RCT had a larger sample size and a larger number of groups, so it improved the generalisability of our findings to practice. Our trial also employed a control of using a credible alternative psychological treatment with a similar aim to remain well. We found that group PEd was more acceptable in terms of sessions attended and from qualitative interviews with participants from both treatment groups. However, group PEd was not more effective than PS in terms of time to the next bipolar episode, except for people with only one to seven previous bipolar episodes. We were able to replicate a more recent finding that group PEd was effective in improving interpersonal function compared with TAU. 104 Although there were some positive effects of group PEd on time to next mania episode, interpersonal function and quality of life, they do not support the effectiveness of PEd versus all other types of control treatment on overall relapse, as reported by Bond and Anderson,32 and provide only weak support for its effectiveness against mania relapse. The results are consistent with the lack of effectiveness of group psychological treatments versus control treatments against bipolar relapse of all types found by NICE, but unlike NICE we could find no evidence of effectiveness against depression episodes. 57,63 However, the latter meta-analysis included studies of group CBT, mindfulness CBT, dialectical behaviour therapy and social cognition and interaction training, not just PEd.

Given that sample size and other methodological limitations are unlikely to explain our results, there are a number of other possible reasons why the current RCT did not show the effectiveness of PEd compared with a control treatment. First, there is a tendency for large multicentre trials to show smaller effect sizes of treatments than smaller single-centre trials. Second, the PS group offered structure, provided information, promoted self-efficacy and provided interpersonal support over 26 weeks in addition to covering some approaches demonstrated to be effective against depression, such as problem-solving. 103 Thus, the PS group may have been effective in its own right. Bond and Anderson32 found that interventions offering > 20 hours of face-to-face group PEd were effective against depression relapse and those offering fewer hours were not effective. In the current study, only 24 (15.7%) participants in the PEd group and 15 (9.9%) participants in the PS group attended 20 or 21 group sessions, so it is possible that there were too few in either group receiving enough therapy time to show an effect on depression relapse. Overall, the proportion of participants experiencing a depression relapse in either group was comparable to the proportion relapsing in TAU or control groups in other RCTs of PEd; in contrast, the proportion experiencing mania relapse was low in both groups, but especially the PEd group, compared with other RCTs of PEd. 32 Group PEd is an intervention in which later sessions of treatment build on learning from earlier sessions, so it is possible that lower levels of attendance at key points might explain its overall lack of effectiveness against depression relapses versus PS.

However, another possible explanation is the case mix of participants in the groups in terms of both the course of their illness and sex. In the case of participants with one to seven previous bipolar episodes, the HR (0.28) for overall bipolar relapse was substantially lower in the PEd group than in the PS, with little or no difference between the groups with more bipolar episodes. Furthermore, the rate of relapse into depression in the PEd group was less than half that of the PS group. However, only 13% of the total sample in the current study had one to seven previous bipolar episodes. Other studies may have had a different case mix, with more participants in an earlier stage of their illness. RCTs of group PEd either vary in how they establish and report the number of previous bipolar episodes or have not reported this baseline variable. Colom et al. 61,62 reported a substantial effect of group PEd versus group PS, and the average number of previous bipolar episodes in the two treatment groups was between 8 and 11, suggesting that a higher proportion of their sample than of our study had few previous bipolar episodes. However, an assessment of the number of previous bipolar episodes is subject to recall bias, and little weight can be given to the precise number of previous bipolar episodes from study to study using different assessment methods; it is possible only to distinguish between those who have had relatively few bipolar relapses and those who have had many bipolar relapses. It is also unclear whether or not the number of previous bipolar episodes is a proxy measure for some other feature of BD that determines if people might benefit from PEd versus PS. Large effect sizes have been reported in people with BD of recent onset in RCTs of psychological interventions and service interventions incorporating group PEd compared with TAU. 73,74 A long-standing theory that has some empirical support is that psychosocial factors may be more associated with relapse earlier in the course of BD (a phenomenon known as behavioural sensitisation)105 than later. Our own qualitative data suggest that people with fewer previous bipolar episodes felt encouraged by meeting people with BD who functioned well, including the SUFs, although this was true of both groups and does not, therefore, explain why PEd was more effective than PS in people with fewer previous bipolar episodes.

The effects of individual and group PEd on function are mixed, although all except the smallest studies measuring function show a benefit of PEd on overall function,34,72,106 overall function in treatment completers107 or interpersonal function. 104 The benefits in this study with group PEd were seen only in relation to interpersonal function and not in overall function or in the performance, friction and dependency domains of the SAS. The participants in this study reported only mild impairments in these domains at the start of the study, although their marital and employment history suggested substantial damage to these aspects of their lives in the past. Therefore, floor effects on function are likely to be operating. The effects on interpersonal function are not necessarily a secondary benefit of PEd on time to bipolar relapse, given its effectiveness only in a subgroup compared with PS. Torrent et al. 104 proposed that forming new relationships and receiving repeated small group exercises in group PEd might also be plausible explanations for improvements in interpersonal function. In the current study, PS was characterised by poorer attendance and a more discursive approach to the content of the group.

There did not seem to be any differences across the treatment centres, suggesting that different types of mental health professional (MHP) can be utilised for this intervention (nurses and psychiatrists were used in the East Midlands, but nurses, occupational therapists and psychologists were used in the North West). Another important difference between group PEd in the current study and group PEd in other studies61,62 is that we utilised a SUF in the groups alongside health professionals. Some qualitative work has investigated this role, but feedback from the participants in both treatment arms in our qualitative study suggests that participants found the SUF role useful, a finding that resonated with a previous study exploring the feasibility of group PEd in another part of the UK. 70 The role of the SUF may be particularly welcome in the context of MHS in the UK, compared with some other countries. SUs now play an active role helping other SUs, particularly in the recovery phases of their mental health problems, so their role in groups promoting longer-term recovery seems culturally appropriate in the UK.

Implications for research

Overall, the current trial did not show that PEd was more clinically effective than PS when delivered in a very similar format and had the same expectations of benefit. The trial provides some limited support that the content of group PEd might provide greater clinical benefit over that of group PS in terms of acceptability, time to first bipolar episode in those with few previous bipolar episodes, time to first mania relapse and improvement in interpersonal function and quality of life. Ideally, another RCT would provide a precise estimate of the clinical and economic benefits of group PEd versus TAU, although this may become increasingly impractical in England and Wales given that psychological interventions for BD are a key NICE recommendation for implementation in the NHS. 57 The data suggest that there are important avenues to pursue both in terms of the stage of BD that a person is in (i.e. early or later in the course of BD) and related to the delivery of optimal psychological treatment such as the added benefit of an initial support group followed by group PEd coupled with optimal medication management. 108 A recent meta-analysis raises the question whether or not treatment lasting ≥ 20 completed sessions is required to improve depression outcomes. 32 It is unlikely that all people with BD who might benefit from group PEd or PEd in general can attend groups, so there are already attempts to deliver both individual PEd109 and group PEd110 through the internet. Further research is needed to refine and more thoroughly test such internet and video conferencing approaches to PEd and group PEd, as well as work on individual techniques to improve clinical outcomes further.

Implications for clinical practice

The National Institute for Health and Care Excellence has already recommended individual, family and group psychological interventions, including group PEd, in BD for implementation in the NHS. 57 Arguably, group PEd is one of the most efficient interventions to deliver because up to 18 participants can be treated at the same time by two health professionals and one SUF. Health professionals and SUFs can be trained quickly to deliver group PEd, provided that they have quite extensive clinical or personal experience of BD and knowledge of group processes, as well as training, access to manuals and supervision by health professionals with experience in both psychological treatment and BD. An account of the importance of these aspects of providing group PEd has been reported. 70 If NHS trusts are now looking to implement either group PEd or group PS, then group PEd appears more acceptable, and there are some clinical benefits, especially for people early in the course of their BD. Our qualitative data suggest that both group PEd and group PS help people to gain the knowledge required to self-manage their BD with more confidence, to discuss optimal treatment with health professionals and to feel less socially isolated. As the content of the groups could be distressing, particularly for those who are isolated or going through life stress, some consideration should be given to providing support after sessions if participants indicate that they have been distressed as a result. There is some evidence of sustained clinical improvement over 5 or 6 years at a reduced overall cost if group PEd is combined with specialist BD expertise in prescribing and ongoing care. 108,111 However, the evidence from the current trial suggests that there is still some doubt about the effectiveness of group PEd compared with providing closed group PS; if there are benefits from group PEd, these may be largely related to the nature of the support and information-sharing offered, rather than to the specific content of the PEd curriculum itself. As there was no TAU group (a group not receiving a group psychological intervention), the trial cannot draw any conclusions about the clinical effectiveness of group PEd versus TAU. Overall, the trial does not provide evidence to reverse recent NICE recommendations of group psychological treatments to prevent relapse, but the benefits of offering group PEd over the group PS offered in the trial are modest.

Approach

The framework of cost-effectiveness analysis was used to compare the relative costs and outcomes of the bipolar group PEd intervention with those of the bipolar group PS control. The primary measure of health benefit was the QALY. The primary economic outcome measure was the incremental direct cost per QALY gained. The direct costs of care and QALYs were estimated for each trial participant from baseline to the end of scheduled follow-up at 96 weeks. The perspective of the evaluation was that recommended by NICE and included health and social care agencies (direct costs of care) and patients (health benefits); these are the key components of a societal perspective. The target population for the economic evaluation (within-trial and economic model analyses) was adults with a primary diagnosis of BD 1 or BD 2 who were at increased risk of further relapse (an episode in the last 24 months). The within-trial analysis used an intent-to-treat approach and included all participants randomised to start therapy in both trial groups. Costs and outcomes were estimated over a 96-week time horizon and were not discounted.

Within-trial analyses

Economic analyses

Descriptive and regression analyses were used to identify key elements of service use and cost. Descriptive and regression analyses were also used to explore the potential impact of baseline participant and service characteristics on the costs and QALYs in order to identify covariates for the cost-effectiveness analysis.

The primary economic analysis was adjusted for the baseline covariates identified as potential predictors of future costs and outcomes. Average and incremental costs and QALYs were estimated, as was the incremental cost-effectiveness ratio (ICER).

The primary measure for the economic analysis was the ICER. Accordingly, no statistical tests of differences in mean costs or outcomes were conducted. The ICER was estimated as:

The estimates of incremental costs and outcomes from the regression were bootstrapped to simulate 10,000 pairs of net costs and net outcomes of the PEd group. These simulated pairs of net costs and net outcomes were used to generate cost-effectiveness acceptability curves, as is recommended by NICE for health technology appraisals. 119 The simulated data were also used to estimate the probability that the PEd intervention is cost-effective compared with PS. This takes a Bayesian approach to estimating the likelihood that the intervention is cost-effective and avoids hypothesis testing and the risk of a type II error.

The cost-effectiveness acceptability approach described above revalues effects or benefits in monetary terms. However, in the UK there is no universally agreed monetary value for the types of health benefit measures (such as QALYs or relapse-free years) typically used in cost-effectiveness analyses. An approach used in health care is to ask the question: what is the maximum amount decision-makers are willing to pay to gain one unit of health benefit? The simulated net utility values were revalued using a range of maximum willingness-to-pay values (WTPVs) from £1 to £50,000 to gain one unit of health benefit. This was based on the range of WTPVs historically implied by NICE decisions. 121

The data for the cost-effectiveness acceptability curve were derived by first revaluing each of the 10,000 net health benefit estimates from the bootstrap simulation by a single WTPV. This was repeated for each WTPV within the range used. A net benefit (NB) statistic for each pair of simulated net costs and net outcomes for each WTPV can then be calculated as:

where HB = net health benefit and C = net cost.

This calculation was repeated for each WTPV. Cost-effectiveness acceptability curves plotted the proportion of bootstrapped simulations when the NB of PEd is greater than zero for each WTPV. 122–125

Sensitivity and subgroup analyses were used to assess the robustness of the results to changes in the range of costs included in the total cost measure, the measure of health benefit and imputation method. The subgroup analyses explored whether or not the cost-effectiveness of PEd varied according to the type of BD, the number of previous episodes, the number of sessions attended and whether or not the participant completed the 96-week SCID assessment.

Decision tree and probabilistic simulation model

The clinical and economic trial was designed as a pragmatic evaluation. The economic within-trial sensitivity and subgroup analyses explore aspects of the trial design and participants that could affect generalisability. Economic models can be used to explore the cost-effectiveness of interventions over different time periods, settings and participants. The within-trial sensitivity and subgroup analyses explored the key issues discussed in Chapter 2a and that may occur when implementing PEd in routine practice.

A consistent issue raised by existing literature is that there is limited evidence about the relative effectiveness, costs or health benefits of psychological therapies,. As discussed in Chapter 2a, the existing clinical evidence is uncertain because of issues with the evaluation designs and participant samples used. 32,57 In addition, there was a lack of primary data on service use and costs, and measures of health benefits that could be used to estimate QALYs. This limits the extent to which further economic modelling could add to the information provided by the within-trial analyses about the relative cost-effectiveness of PEd compared with PS.

A key problem identified in the updated NICE guideline CG185 was that there was insufficient evidence to model the relative cost-effectiveness of psychological therapies compared with TAU. 57 This issue was also discussed in Chapter 2a of this report.

Accordingly, a simple economic model and a threshold analysis were used to explore the reduction in the probability of relapse and/or time to relapse associated with PEd that would be required for it to be cost-effective when compared with TAU. This approach provides additional information for settings or patient groups in which access to psychological therapies such as that provided by PS is limited. The key benefits of this approach are that it controls for heterogeneity in study design, participant samples, the range of service use measured and unit costs used to value resource use. However, the approach used means that the scope and relevance of the economic model results are limited to the participants, setting and time horizon of the clinical trial. The overall aim of these modelling analyses was to explore the circumstances in which PEd may be cost-effective. Key questions for the model based analyses were:

• What is the relative reduction in the probability of relapse in PEd for it to be cost-effective compared with TAU?

• What is the relative reduction in the time to relapse in PEd for it to be cost-effective compared with TAU?

• What is the combined relative reduction in the probability of relapse and time to relapse in PEd for it to be cost-effective compared with TAU?

A simple economic model that combined an initial decision tree to map the early trial pathways and a Markov model to map relapse was constructed. The TAU arm of the model was defined as no psychological treatment.

The primary economic outcome measure for the model analyses was the incremental direct cost per QALY gained. The direct costs of care and QALYs were estimated from the start of treatment to 96 weeks, to mirror the trial scheduled follow-up. The perspective of the evaluation was that of health and social care agencies and patients. The target population for the economic model analyses was people with BD 1 or BD 2 at increased risk of further relapse (an episode in the last 24 months). Data from the trial sample of participants were used to represent this population.

The initial decision tree structure was based on the care pathways used in the trial design. The Markov section of the model was based on the main aim of the PEd and PS interventions, namely to prevent and/or delay the first relapse following treatment. The model structure is shown in Figure 8. The decision tree part of the model describes the distibution of participants in terms of attending the group sessions. The Markov section explores the probability and time to first relapse in the follow-up period. The time horizon is 96 weeks, split into three cycles of 32 weeks. An analysis of the trial data indicated that there were differences in the average costs per assessment period (32 weeks) between those participants who had a relapse and those who did not. This applied overall and by assessment. However, there did not appear to be differences in the average cost per assessment within those people who had a relapse, or for those people who did not have a relapse. This also applied to the utility data.

FIGURE 8.

Economic model structure.

The following data were estimated from the PEd arm of the trial data set:

• probability of attending group sessions

• probability of first relapse in the follow-up period

• the costs of the intervention

• the cost per week of relapse/no relapse, estimated as the average cost per assessment divided by 32 weeks (the planned length of the period between assessments)

• the average utility of relapse/no relapse

• the time to relapse and time following relapse (weeks).

It was assumed that the cost per week of relapse/no relapse and the utility associated with the relapse and no relapse states would be the same for both PEd and TAU. All that differed between the two management strategies was the probability of relapse and time to relapse.

The probabilities of relapse and time to relapse for PEd were calculated from the predicted data generated for the clinical analysis. The utility and cost parameters for the relapse and no relapse states were estimated from the multiple imputation data generated for the economic evaluation. All data were entered into the model as distributions. Beta distributions for integers were used for the probability data. Gamma distributions were used for the cost, utility and time data. The cost and time distributions were constrained to a minimum value of zero.

Monte Carlo simulation in TreeAge software (10,000 iterations; TreeAge Software, Inc., Williamstown, MA, USA) was used to estimate the net costs, QALYs and ICER of PEd. A probabilistic sensitivity analysis was used to estimate the likelihood that PEd was cost-effective compared with TAU and to generate cost-effectiveness acceptability curves.

Baseline clinical and demographic characteristics

The main sociodemographic characteristics of participants are reported in Table 4. Additional sociodemographic characteristics of participants were used as covariates for the economic analyses, and these are shown in Appendix 1, Table 77. Overall, there were no apparent differences between the groups on these additional characteristics, although fewer people in the PEd group (33%) than in the PS group (45%) lived alone.

The key baseline characteristics identified (by regression analysis and/or literature) as important for the analysis of costs and QALYs were:

• which wave (PS or PEd group) participants were enrolled in

• the number of previous episodes of BD

• type of bipolar diagnosis (1 or 2)

• the HADS-A and HADS depression scores

• whether or not the participant was white British

• whether or not the participant was currently working

• whether or not the participant had further or higher education

• sex and age

• whether or not the participant lived alone.

These variables were used as covariates for all of the primary and secondary analyses of costs, QALYs and cost-effectiveness. In addition, the baseline SF-6D utility score and the EQ-5D thermometer were used as covariates in the analyses of QALY data and baseline costs were used in the analyses of cost data. Utility values were not correlated with the baseline SCID mania scores (Pearson’s coefficient: –0.004; p = 0.954).

Baseline utility scores were correlated with the following baseline clinical measures:

• SOFAS (Pearson’s coefficient = 0.302; p = 0.000; n = 243)

• SAS (Pearson’s coefficient = –0.163; p = 0.010; n = 246)

• HADS-A (Pearson’s coefficient = –0.474; p = 0.000; n = 245)

• HADS depression (Pearson’s coefficient = –0.481; p = 0.000; n = 245)

• SCID depression (Pearson’s coefficient = –0.345; p = 0.000; n = 258).

In addition, utility values at the 96-week assessment were correlated with whether or not the participant had a relapse (Pearson’s coefficient = –0.298; p = 0.003; n = 99) and time to relapse (Pearson’s coefficient = 0.322; p = 0.001; n = 99). QALYs, which combine survival with utility, were similarly correlated with whether or not the participant relapsed (Pearson’s coefficient = –0.234; p = 0.048; n = 72) and time to relapse (Pearson’s coefficient = 0.227; p = 0.055; n = 72). Both utility values and QALYs were correlated with the 96-week clinical measures used as secondary outcomes (p < 0.02).

The trial participants had lower utility scores compared with the population norms for the UK (mean 0.68, 95% CI 0.65 to 0.72; n = 258) versus (95% CI 0.85 to 0.91126) for people of a similar age band (UK population norms, 35–54 years; trial participants, mean age 45 years; 95% CI 44 to 46 years). Overall, the baseline EQ-5D utility values indicate that this sample of participants was of similar health to the sample of participants enrolled in the Cost Utility of the Latest Antipsychotic drugs in Schizophrenia Study (CUtLASS) trials,116,127 who had severe schizophrenia and required a change in medication (mean 0.63, 95% CI 0.60 to 0.66; n = 362) (Linda M Davies, University of Manchester, 2015, unpublished analysis).

Descriptive analysis

Descriptive data are presented below for participants with complete cost and EQ-5D data. Tables reporting available data are summarised in Appendices 2 (utility and QALY data) and 3 (service use and costs) for interested readers.

Primary analysis

As described in Methods, the primary analysis used multiple imputation to account for missing observations and missing follow-up data. The imputed mean costs and QALYs for each assessment are reported in Table 21. A regression analysis (using the Stata mi estimate command to account for multiple imputation data) was used to estimate the net costs and QALYs and to control for any impact of baseline covariates that may affect the cost and QALY values.

TABLE 21 - Mean costs and QALYs of group PEd and group PS and p-value for the difference, multiple imputation data

Twenty-nine participants did not complete the service use or EQ-5D at baseline or any of the follow-up assessments. It was not possible to impute utility/QALY or cost data for these people.

Adjusted for study wave; employment status; sex; number of previous bipolar episodes; level of education; bipolar type (1 or 2); ethnicity; living arrangement; age; HADS-A score (baseline); HADS depression score (baseline); SF-6D utility (baseline, costs not adjusted for this variable); EQ-5D thermometer (baseline, costs not adjusted for this variable); pre-baseline health-care cost (QALYs not adjusted for this variable).

Follow-up period	Unadjusted, mean (95% CI)		p-value for difference (PEd vs. PS)
	PEd (n = 139/153a)	PS (n = 136/151a)	Unadjusted	Adjustedb
Baseline–week 32
Days of follow-up	227 (212 to 242)	219 (203 to 235)	0.48	0.68
QALYs	0.42 (0.38 to 0.46)	0.39 (0.35 to 0.43)	0.31	0.55
Costs (£)	2924 (2298 to 3549)	2420 (1925 to 2914)	0.17	0.15
Weeks 32–64
Days of follow-up	225 (218 to 231)	227 (220 to 234)	0.68	0.73
QALYs	0.40 (0.36 to 0.43)	0.38 (0.34 to 0.42)	0.50	0.78
Costs (£)	2379 (1432 to 3326)	2383 (1556 to 3210)	0.99	0.87
Weeks 64–96
Days of follow-up	224 (215 to 234)	219 (209 to 230)	0.46	0.51
QALYs	0.38 (0.35 to 0.42)	0.37 (0.33 to 0.41)	0.59	0.96
Costs (£)	2722 (1646 to 3799)	2590 (1689 to 3491)	0.85	0.59
Baseline–week 96
Days of follow-up	676 (658 to 694)	665 (649 to 681)	0.36	0.53
QALYs	1.20 (1.12 to 1.28)	1.14 (1.04 to 1.23)	0.33	0.67
Costs (£)	8025 (6001 to 10,049)	7393 (5857 to 8929)	0.61	0.42

Table 22 reports the net costs and QALYS for the two groups using the imputed data. As with the previous descriptive analyses, these results are characterised by relatively wide 95% CIs. This suggests that there are no statistically significant differences in costs or QALYs between the PEd and PS groups (after controlling for the effect of any baseline characteristics of participants or trial implementation, and treating costs and QALYs as independent of each other).

TABLE 22 - Net costs and QALYs of group PEd compared with group PS, multiple imputation data, adjusted for baseline characteristicsa

Adjusted for study wave; employment status; sex; number of previous bipolar episodes; level of education; bipolar type (1 or 2), ethnicity; living arrangement; age; HADS-A score (baseline); HADS depression score (baseline); SF-6D utility (baseline, costs not adjusted for this variable); EQ-5D thermometer (baseline, costs not adjusted for this variable); pre-baseline health-care cost (QALYs not adjusted for this variable).

Regression analysis with/without bootstrap simulation	Net cost (£) (standard error) [95% CI]	Net QALY (standard error) [95% CI]	Net cost (£) per QALY gained	Probability that PEd is cost-effective if willing to pay £30,000 to gain 1 QALY
No bootstrap (unadjusted)	638 (1262) [–1848 to 3124]	0.061 (0.062) [–0.064 to 0.185]	10,484	Not estimated
No bootstrap (adjusted)	1042 (1276) [–1473 to 3557]	0.025 (0.058) [–0.092 to 0.141]	41,680	Not estimated
Bootstrap	1098 (431) [252 to 1943]	0.023 (0.014) [0.001 to 0.056]	47,739	0.35

The net cost and QALY data from the regressions were simulated (bootstrapped) 10,000 times to generate 10,000 pairs of net costs and QALYs. The results are reported in the last row of Table 22. In contrast to the results from the regression analysis alone, these simulated data suggest that PEd is associated with an overall net cost and gain in QALYs compared with the PS intervention. Given the similar costs of the two interventions, any additional cost incurred by the PEd participants indicates higher use of health and social care services. The point estimate of the net cost per QALY gained is relatively high at £47,739 (£1098/0.023 QALY), which is above the commonly reported NICE threshold of £20,000–30,000 per QALY gained. 119 There is some uncertainty associated with this cost per QALY estimate that reflects the variance in the costs and QALYs as well as uncertainty about the willingness-to-pay threshold.

This uncertainty is illustrated in Figure 9, which presents a scatterplot of the 10,000 pairs of net cost and QALY data in the form of a cost-effectiveness plane. Figure 9 illustrates that the net costs are mostly scattered above the horizontal axis, suggesting that, overall, PEd is likely to cost more than PS. The net QALY points on the scatterplot are mainly to the right of the y-axis, suggesting that PEd is more likely than PS to produce a health benefit in terms of QALYs.

FIGURE 9.

Cost-effectiveness plane of net cost and QALY pairs: primary analysis.

Figure 10 shows the cost-effectiveness acceptability plane. This indicates the likelihood or probability that the net cost per QALY gained by PEd (within the 10,000 simulated net cost and QALY pairs) is less than or equal to the amount (WTPV) that decision-makers would be willing to pay to gain 1 QALY. Compared with PS, the PEd intervention is likely to be cost-effective in ≥ 50% of cases if decision-makers are willing to pay ≥ £37,500 to gain 1 QALY.

FIGURE 10.

Cost-effectiveness acceptability curve: primary analysis.

Sensitivity analyses

The results of all of the sensitivity analyses are presented in Table 23. However, Figures 11 and 12 report the cost-effectiveness acceptability curves for only those sensitivity analyses in which the conclusion differed from that of the primary analysis. All of the sensitivity and subgroup analyses report the bootstrapped net costs and QALYs that were generated from regression analyses to control for the impact of baseline characteristics. It is important to note that these analyses are exploratory and that the sample size available for each analysis may not be sufficient to identify important differences. In addition, the number of these exploratory analyses increases the chance of finding a difference between the exploratory analysis and the primary analysis because of the number of analyses, rather than because an important difference exists. These sensitivity analyses used an intention-to-treat approach. They were used to assess whether or not there was uncertainty as a result of the choices made about the design of the economic evaluation (e.g. choice of health benefit measure, range of costs included) and analysis methods (e.g. method of dealing with missing observations and missing follow-up).

TABLE 23 - Net costs and QALYs of group PEd compared with group PS, sensitivity analyses, bootstrapped data, adjusted for baseline characteristicsa

Adjusted for study wave; employment status; sex; number of previous bipolar episodes; level of education; bipolar type (1 or 2); ethnicity; living arrangement; age; HADS-A score (baseline); HADS depression score (baseline); SF-6D utility (baseline, costs not adjusted for this variable); EQ-5D thermometer (baseline, costs not adjusted for this variable); pre-baseline health-care cost (QALYs not adjusted for this variable).

Sensitivity analysis	Net cost (£) (95% CI)	Net QALY (95% CI)	Net cost per QALY gained (£)	Probability PEd cost-effective if willing to pay £30,000 to gain 1 QALY
Primary analysis
	1098 (252 to 1943)	0.023 (0.001 to 0.056)	47,739	0.35
Type of imputation
Complete-case analysis (n = 55)	–588 (–4734 to 3558)	0.049 (–0.177 to 0.274)	PEd dominates, QALY gain, net saving	0.83
Missing QALY data imputed by linear interpolation (n = 287)	1122 (272 to 1971)	0.024 (0.004 to 0.045)	46,750	0.23
Alternative measures of health benefit
Relapse-free years (n = 287)	1098 (252 to 1943)	0.131 (0.08 to 0.182)	8382	Probability PEd cost-effective if willing to pay £30,000 to gain 1 relapse-free year: 0.99
Relapse avoided (n = 287)	1098 (252 to 1943)	0.102 (0.68 to 136)	10,765	Probability PEd cost-effective if willing to pay £30,000 to avoid one relapse: 0.99
QALY estimated from SF-6D utility values (n = 287)	1098 (252 to 1943)	0.013 (0.000 to 0.027)	84,461	0.07
Alternative cost measures
Costs medications, exclude costs ‘other’ medicines (n = 289)	1008 (162 to 1854)	0.028 (0.001 to 0.056)	25,200	0.41
Costs exclude costs of PEd/PS (n = 287)	1124 (275 to 1974)	0.028 (0.001 to 0.056)	40,142	0.34

FIGURE 11.

Cost-effectiveness acceptability curves, willingness to pay to gain 1 QALY, sensitivity analyses: complete cases, relapse-free years, relapse avoided.

FIGURE 12.

Cost-effectiveness acceptability curves, willingness to pay to gain 1 QALY, sensitivity analyses: length of follow-up.

Figures 11 and 12 summarise the cost-effectiveness acceptability curves for only the sensitivity analyses that indicate a different conclusion from the primary analysis. PEd is more likely to be cost-effective than PS in the following scenarios:

• for the complete-case analysis (that does not include multiple imputation of missing data)

• if participants are followed up for 64 weeks rather than for 96 weeks

• when the clinically based outcomes of relapse avoided and relapse-free years are used as the measure of health benefit instead of QALYs.

Table 24 reports the estimated net costs, QALYs and ICER of PEd for each of the within-trial subgroup analyses conducted, along with the probability that PEd is cost-effective compared with PS. Figures 13 and 14 summarise the cost-effectiveness acceptability curves for those subgroup analyses that indicate a different conclusion (compared with the primary analysis) about the likely cost-effectiveness of PEd.

TABLE 24 - Net costs and QALYs of group PEd compared with group PS, subgroup analyses, bootstrapped data, adjusted for baseline characteristicsa

Adjusted for study wave; employment status; sex; number of previous bipolar episodes; level of education; bipolar type (1 or 2); ethnicity; living arrangement; age; HADS-A score (baseline); HADS depression score (baseline); SF-6D utility (baseline, costs not adjusted for this variable); EQ-5D thermometer (baseline, costs not adjusted for this variable); pre-baseline health-care cost (QALYs not adjusted for this variable).

Subgroup analysis	Net cost (£) (95% CI)	Net QALY (95% CI)	Net cost per QALY gained (£)	Probability that PEd is cost-effective if willing to pay £30,000 to gain 1 QALY
Primary analysis
	1098 (252 to 1943)	0.023 (0.001 to 0.056)	47,739	0.35
Number of PEd or PS sessions attended
< 10 (n = 112)	149 (–555 to 852)	0.019 (–0.025 to 0.064)	7842	0.72
≥ 10 (n = 175)	1266 (24 to 2507)	0.023 (–0.012 to 0.057)	55,043	0.25
≥ 16 (n = 118)	–844 (–1997 to 308)	0.080 (0.036 to 0.124)	PEd dominates	0.99
1–15 (n = 138)	1598 (896 to 2301)	0.001 (–0.040 to 0.043)	1,598,000	0.02
0–15 (n = 169)	955 (332 to 1577)	–0.003 (–0.040 to 0.034)	PS dominates	0.07
0 (n = 31)	–649 (–2114 to 814)	0.056 (–0.049 to 0.161)	PEd dominates	0.92
Number of previous episodes
≥ 8 (n = 249)	1238 (270 to 2206)	0.030 (0.001 to 0.060)	41,267	0.34
1–7 (n = 38)	1267 (–451 to 2985)	–0.056 (–0.132 to 0.019)	PS dominates	0.04
Number of previous episodes, number of sessions attended
≥ 8 episodes, < 16 sessions (n = 147)	1026 (385 to 1667)	0.012 (–0.028 to 0.052)	85,500	0.21
≥ 8 episodes, 16+ sessions (n = 103)	–1109 (2504 to 287)	0.043 (–0.003 to 0.089)	PEd dominates	0.99
1–7 episodes, 16+ sessions (n = 15)	–713 (–1898 to 472)	0.266 (–0.201 to 0.733)	PEd dominates	0.99
1–7 episodes, <16 sessions (n = 23)	6271 (3716 to 8826)	–0.133 (–0.221 to –0.044)	PS dominates	0.00
Participant completed SCID at 96-week follow-up
SCID score (n = 212)	983 (–25 to 1990)	0.038 (0.009 to 0.067)	25,868	0.61
No SCID score (n = 75)	470 (–273 to 1213)	–0.017 (–0.003 to 0.049)	PS dominates	0.18
Type 1 or 2 bipolar diagnosis
Type 1 (n = 229)	1183 (177 to 2189)	0.006 (–0.025 to 0.038)	197,167	0.009
Type 2 (n = 58)	375 (–107 to 857)	0.129 (0.077 to 0.181)	2907	1.00

FIGURE 13.

Cost-effectiveness acceptability curves, willingness to pay to gain 1 QALY, subgroup analyses: number of sessions attended.

FIGURE 14.

Cost-effectiveness acceptability curves, willingness to pay to gain 1 QALY, subgroup analyses: completed 96-week SCID assessment, BD 2 diagnosis.

The subgroup analyses exploring the impact of number of sessions attended indicate that PEd was more likely to be cost-effective than PS for the following:

• participants who attended ≤ 9 sessions

• participants who attended ≥ 16 sessions

• participants who attended no sessions; however, numbers for this analysis are small, and the group of people who attended no sessions and had follow-up assessments may be atypical.

The number of previous bipolar episodes at baseline did not change the conclusions of the primary analysis. Defining subgroups according to both number of sessions and number of previous episodes indicates that PEd was likely to be cost-effective in people who had ≥ 16 sessions regardless of the number of previous episodes (1–7 previous bipolar episodes or ≥ 8 previous bipolar episodes).

PEd was also likely to be cost-effective compared with PS in:

• the sample of participants who completed the 96-week SCID assessment

• those participants with a BD 2 diagnosis.

The final set of subgroup analyses explored whether or not the results of the analyses using the primary clinical outcome (relapse-free years) varied by different groups of the trial participants. Table 25 shows the results of these analyses. These indicated that PEd was not likely to be a cost-effective way to delay relapse if participants did not attend any sessions (PS or PEd) or for the group of people who had eight or more previous bipolar episodes and attended < 16 sessions.

TABLE 25 - Net costs and relapse-free years of group PEd compared with group PS, subgroup analyses, bootstrapped data, adjusted for baseline characteristics

Subgroup analysis	Net cost (£) (95% CI)	Net relapse-free year (95% CI)	Net cost per relapse-free year gained (£)	Probability that PEd is cost-effective if willing to pay £30,000 to gain 1 relapse-free year
Primary analysis
	1098 (252 to 1943)	0.131 (0.080 to 0.182)	8382	0.90
Number of PEd or PS sessions attended
< 10 (n = 112)	149 (–555 to 852)	0.086 (–0.0003 to 0.168)	1733	0.95
≥ 10 (n = 175)	1266 (24 to 2507)	0.137 (0.072 to 0.203)	9240	0.99
≥ 16 (n = 118)	–844 (–1997 to 308)	0.200 (0.115 to 0.280)	PEd dominates	1.00
1–15 (n = 138)	1598 (896 to 2301)	0.063 (–0.011 to 0.136)	25,365	0.61
0–15 (n = 169)	955 (332 to 1577)	0.076 (–0.010 to 0.141)	12,566	0.87
0 (n = 31)	–649 (–2114 to 814)	–0.082 (–0.196 to 0.031)	7915 saving per relapse-free year lost	0.19
Number of previous episodes
≥ 8 (n = 249)	1238 (270 to 2206)	0.053 (0.001 to 0.106)	23,358	0.61
1–7 (n = 38)	1267 (–451 to 2985)	0.841 (0.689 to 0.993)	1507	1.00
Number of previous episodes, number of sessions attended
≥ 8 episodes, < 16 sessions (n = 147)	1026 (385 to 1667)	–0.014 (–0.082 to 0.055)	PS dominates	0.09
≥ 8 episodes, ≥ 16 sessions (n = 103)	–1109 (2504 to 287)	0.099 (0.011 to 0.186)	PEd dominates	1.00
1–7 episodes, ≥ 16 sessions (n = 15)	–713 (–1898 to 472)	1.290 (0.876 to 1.700)	PEd dominates	0.99
1–7 episodes, < 16 sessions (n = 23)	6271 (3716 to 8826)	0.314 (–0.101 to –0.527)	19,971	0.83
Participant completed SCID at 96-week follow-up
SCID score (n = 212)	983 (–25 to 1990)	0.072 (0.012 to 0.131)	13,653	0.89
No SCID score (n = 75)	470 (–273 to 1213)	0.379 (0.282 to 0.476)	51	1.00
Type 1 or 2 bipolar diagnosis
Type 1 (n = 229)	1183 (177 to 2189)	0.123 (0.067 to 0.179)	9617	0.99
Type 2 (n = 58)	375 (–107 to 857)	0.187 (0.062 to 0.311)	2005	1.00

Summary of findings

Overall, the within-trial analyses of mean costs and QALYs suggested uncertainty about whether or not there were any differences between the PEd and PS groups. However, the primary analysis of net costs, QALYs and cost per QALY gained suggested a net additional cost for the PEd intervention of £1098 (95% CI £252 to £1943) and a small QALY gain of 0.023 (95% CI 0.001 to 0.056). This gives an incremental cost of £47,739 to gain 1 QALY. The additional cost of PEd indicates higher service use by the PEd participants. There was some indication that costs were higher in the PEd group in the period prior to baseline, which could mean that there were differences in the underlying need and access to health-care services. It is not clear what impact this may have on the analyses. However, as previous costs tend to predict future costs, costs for the 6 months before the baseline assessment were included as a covariate to control for baseline differences and to take any possible imbalances between the PEd and PS groups into account.

A further conclusion from the primary, within-trial analysis was that the PEd intervention, compared with PS, was likely to be cost-effective only if decision-makers are willing to pay at least £37,500 to gain 1 QALY. If decision-makers are not prepared to pay this amount, then PEd is unlikely to be more cost-effective than the equally intensive PS used for the control arm in this trial. At a willingness-to-pay threshold of £30,000, PS has a higher probability of being cost-effective (65%) than PEd (35%). As discussed in Chapter 2a, it is unclear whether or not people are able to access similar PS in routine clinical practice.

The results of the economic model analyses suggest that PEd could be cost-effective compared with TAU if PEd affects only the probability of relapse and reduces the probability of relapse by 15–25%. The probability of relapse for PEd used in the model was 0.58. This appears feasible when compared with the overall rates of relapse found in a recent systematic review of PEd. 32 If PEd affects both probability of relapse and the time to relapse, it would need to decrease the probability of relapse by 10–20% and increase the time to relapse by 10–20% to be cost-effective (depending on how much the decision-maker is willing to pay to gain 1 QALY). This translates into a probability of relapse in TAU of 0.64–0.70 and a time to relapse of 24–27 weeks.

However, the results from the economic model analyses indicated that the 95th percentiles for the net costs and net QALYs were wide and crossed zero, indicating a high level of uncertainty.

Strengths and limitations

The within-trial economic analysis was subject to the same strengths and limitations discussed for the clinical evaluation of effectiveness in terms of trial design, trial sample and length of follow-up.

Descriptive and regression analyses were used to explore the potential impact of baseline participant and service characteristics on the costs and QALYs in order to identify covariates for the cost-effectiveness analysis. The key covariates identified were similar to those identified in other economic evaluations of interventions for people with complex physical or mental health problems. 116,117,128–130

Additional issues that are relevant for the economic analysis include the range of costs considered, the measure of health benefit used for the primary analysis, and a low level of participants with complete cost and utility data.

The costs included in the economic evaluation were limited to the direct costs of providing health and social care, in accordance with the perspective taken. This will underestimate the total costs of BD for the trial participants. If there are differences in the effectiveness of the PEd or PS interventions that lead to differences in employment between the two groups, then this will also affect the net costs associated with PEd. If employment is higher in the PS group, then the net cost of the PEd group will be underestimated, and may reduce the likelihood that PEd is cost-effective yet further.

The measure of health benefit, the QALY, was used for the primary analysis and most of the sensitivity and subgroup analyses. The EQ-5D, a generic measure, was used to collect data about health status, estimate utility values and combine these with data about survival to generate QALYs. This is in line with the NICE recommendations for economic evaluations. 119 It also enables comparison between different disorders. This is relevant for policy-makers and commissioners, who have to consider the distribution of limited budgets between different health-care services. However, a generic health status measure may not be sufficiently sensitive to identify important clinical changes in participants’ mental health or bipolar status and so could underestimate the benefits of PEd.

The descriptive analyses indicated that the EQ-5D health status measure and associated utility index correlated well with the clinical measures used in the trial. In addition, QALY estimates were associated with the primary relapse outcome measures. This gives an indication that the EQ-5D and QALYs are sensitive to changes in the health of trial participants.

There was a large number of missing data, in large part owing to missing mid-follow-up observations for people who completed follow-up. This adds uncertainty to the results of the primary and sensitivity analyses of the costs and outcomes. Multiple imputation was used to impute missing values. There was evidence that the missing data owing to either missing observations or incomplete follow-up were not missing at random, which is a key threat to the validity of the multiple imputation procedure. In addition, only 55 out of 304 participants had complete cost and QALY data for all four assessments, although 54% had complete data at two or more assessments and 34% had complete data at three or more assessments.

The regression models used to impute missing data were based on initial descriptive analyses and regression analyses to identify key baseline and follow-up variables that were associated with either cost or outcomes. These were included in the chained equations to iteratively impute missing data. This goes some way to controlling for the influence of observed variables on the missing data (e.g. the influence of deteriorating mental health/relapse on the use of services, health status and willingness/ability to engage with data collection at follow-up assessments). However, the influence of unobserved variables could not be taken into account, and it is unclear the extent to which this was important.

The results of the sensitivity analyses indicated some uncertainty in the data due to the method of dealing with missing data. The analysis that used complete-case data suggested a lower net cost per QALY gained and a higher probability that PEd was cost-effective compared with PS. However, the complete case could overestimate the cost-effectiveness of PEd if the group of participants with complete data had better health, were more likely to engage with the intervention or were more likely to respond to the intervention than those with missing data. This would suggest that the analysis using multiple imputation may be conservative and underestimate the relative cost-effectiveness of PEd. The results when QALYs were imputed by linear interpolation were similar to those found in the primary analysis.

The results were also sensitive to the length of follow-up. PEd was more likely to be cost-effective at the 64-week assessment than at the 32- or 96-week assessment. This may indicate that the effect of the intervention increased between weeks 32 and 64. Alternatively, it may also indicate that missing data due to incomplete 96-week follow-up had an influence on the overall net costs and QALYs that was not accounted for by the multiple imputation process to deal with missing data.

A range of sensitivity and subgroup analyses were conducted to explore whether or not aspects of the trial and economic evaluation design may affect the likelihood that PEd is cost-effective. These analyses also explored whether or not there were specific groups of participants who may benefit more from PEd. This approach can help to identify issues for policy and further research. However, it is important to note that these analyses are exploratory, and the sample size available for each analysis may not be sufficient to identify important differences. Additionally, the number of these exploratory analyses increases the chance of finding a difference between the exploratory analysis and the primary analysis because of the number of analyses, rather than because an important difference exists.

There was limited evidence from the published literature with which to explore whether or not PEd would be likely to be cost-effective when implemented in routine practice in different settings and patient groups. This limits the extent to which any additional modelling analyses could add to the results and conclusions of the within-trial sensitivity analyses, many of which confirmed the conclusion of the primary analysis that PEd may not be cost-effective using the willingness-to-pay thresholds indicated by NICE decisions and recent commentaries. 131–136 However, a key issue noted in Chapter 2a and by NICE guidelines is the lack of evidence about the cost-effectiveness of psychological therapies compared with TAU. Accordingly, an economic model was used to explore what PEd would need to achieve in terms of effectiveness to be cost-effective compared with TAU.

There was limited evidence from the published literature to inform the structure of a full economic model that compares PEd with TAU or estimate the costs and QALYs associated with events in the TAU arm of the model. An assessment of the likely cost-effectiveness of PEd compared with TAU is important for commissioners and service providers because of the intensive nature of the control intervention (PS) in our study, which may not reflect what people receive in reality.

A simple model explored the relative reductions in the probability of relapse and increases in time to relapse that need to be achieved by PEd for it to be cost-effective compared with TAU. The advantage of the modelling approach used is that it controls for heterogeneity and uncertainty arising from using data from several sources, collected in different ways, for different samples of study participants. The simple model was focused on the relative impact of the PEd intervention on the probability of relapse and the time to relapse, compared with TAU (defined as no PEd). However, these advantages also mean that the scope of the model is constrained by the trial design, the length of follow-up and the people with BD recruited into the main RCT. Additionally, using cost and utility data from the trial also means that these are subject to the same limitations and uncertainty as missing data.

Other issues

Current evidence and commentary indicates that NICE WTPVs are < £15,000 per QALY. 113,115–120 The actual willingness to pay depends on the level of uncertainty in the evidence and the corresponding confidence that an intervention with a cost per QALY within the WTPV will occur in practice.

The net cost per QALY gained by PEd, compared with PS, is relatively high, at £47,739, which is above the commonly reported NICE threshold of £20,000–30,000 per QALY gained, based on past decisions. Whether or not NICE supports a technology also depends on the level of uncertainty associated with the cost and QALY results supporting an intervention. A range of thresholds that decision-makers are willing to pay to gain 1 QALY are reported and debated in the literature. 131–136 However, if, as seems likely, commissioners and service providers wish to focus on relapse-free years, then PEd is much more likely to be cost-effective than PS, with less uncertainty. Whether or not the NICE thresholds for cost per relapse-free year gained would be as high as those for cost per QALY gained is unclear and would depend, in part, on whether or not the relapse-free year includes all the key aspects of health status included in the QALY.

The economic model analyses indicate that PEd may be cost-effective compared with TAU if it achieves reductions in the probability of relapse and increases in the time to relapse. This is an important consideration for commissioners and service providers in areas in which TAU does not include the type of intensive group PS used as the control in our trial. Whether or not these reductions are plausible will vary from setting to setting, depending on what TAU comprises.

The quantitative and qualitative evaluations of PEd reported in Chapter 2a suggest that there may be additional reasons for implementing PEd rather than PS, which may counteract the high cost per QALY gained.

Implications for research

The exploratory sensitivity and subgroup analyses indicated a number of areas for further research.

The results were sensitive to the length of follow-up. PEd was more likely to be cost-effective at the 64-week assessment than at the 32- or 96-week assessment. Further work to identify the trajectory of costs and QALYs during the intervention phase and over the longer term is needed. This will help understanding of whether the change found in our study reflects what would happen in routine practice or whether it is an artefact produced by missing observations and incomplete follow-up.

A subgroup analysis of the economic data indicated that PEd might be cost-effective compared with PS for people with type 2 BD, those who attended ≥ 16 sessions and those who completed the 96-week SCID assessment. The subgroup analyses also indicate that the number of sessions attended affected the relative cost-effectiveness of PEd. There appeared to be differences in the utility values for those people who completed no sessions or only one session of PEd or PS compared with those who completed two or more sessions. This may indicate that those people who did not engage with the PEd or PS interventions had better overall health, which may have influenced their decisions about whether or not to participate in the group interventions. However, those people who did not complete the 96-week SCID assessment had lower utility values at baseline and follow-up than those who did complete the 96-week SCID assessment. This may indicate that these participants had lower overall health from the start of their involvement in the trial, which may have influenced whether or not they completed follow-up.

These factors suggest the need for careful identification of people who are likely to engage with and benefit from the intervention. Further work is needed to understand the influence of a patient’s overall health on decisions to participate in group PEd or PS. Combined with the process evaluation conducted as part of our study, the additional research would help to target the intervention at participants able and willing to engage in the group intervention and/or improve access to group interventions.

The baseline EQ-5D data also suggest that the participants in this trial had worse overall health than the general population (compared with population norms for the UK). The data also suggest that the trial participants had similar levels of health to those enrolled in the CUtLASS trials of antipsychotic medications for people with severe schizophrenia who required a change in medication. 116,117 However, little is known about the health of people with BD. Further work to measure and value overall health would help in the design and effective implementation of current and future interventions.

The economic model identified that PEd could be cost-effective compared with TAU. However, there was little economic evidence to inform either the structure of the model or the estimate of the parameters. The results of the economic model analyses also indicated a high level of uncertainty. This, combined with the conclusions of recent reviews, indicates that further primary data collection and evaluation is required to identify whether PEd (or PS) is effective and cost-effective in comparison with TAU.

Prevalence of anxiety in bipolar disorder

Anxiety disorders are relatively common in the general population, with 12-month prevalence estimates of 2–18% and lifetime risk of 20–29%. 137–142 However, prevalence rates for comorbid anxiety disorders are substantially higher in those with BD, with 12-month prevalence rates of 32–53%11,12 and lifetime risk of 60–90%. 5,143 In addition, there is evidence that anxiety disorders constitute the most common comordibity in BD, being more frequent than alcohol and substance misuse or personality disorder. 144 The most commonly used diagnostic scheme in bipolar research, DSM-IV,76 identifies seven types of anxiety disorder, namely agoraphobia, generalised anxiety disorder (GAD), obsessive–compulsive disorder (OCD), panic disorder, panic disorder with agoraphobia, post-traumatic stress disorder (PTSD), social phobia and specific phobia. According to Merikangas et al. ,5 the most common types of anxiety disorder comorbid with BD are social phobia, specific phobia and GAD.

Impact of anxiety on clinical course and outcome

Comorbid anxiety is an important issue in BD because of the distress it causes patients and, specifically, the negative impact it can have on the course and outcome of the condition. Feske et al. 17 found that higher scores on a composite measure of anxiety were associated with longer time to remission and worse medication side effects in individuals with an acute BD 1 episode. Otto et al. 12 followed 1000 participants with BD over 12 months. The presence of a comorbid anxiety disorder at recruitment was associated with longer time to recovery, higher relapse risk and poorer psychosocial function across the follow-up period. This pattern of poorer psychosocial functioning in individuals with comorbid anxiety disorders was replicated in a study of 1600 participants with BD followed up over 36 months. 143 Anxiety is also associated with higher rates of suicidal ideation and attempted suicide in those with BD. 145,146

Available interventions for anxiety in bipolar disorder

Given the potential importance of anxiety in relation to clinical outcomes in BD, there has been increasing interest in how to improve anxiety symptoms. As NICE indicates, there is clear evidence for the benefits of time-limited interventions for anxiety disorders as primary diagnoses. 145–149 To date, there have been no specific psychological interventions targeted at anxiety in the context of BD. Provencher et al. 150 have explored both drug and psychological therapy outcomes studies including individuals with BD and comorbid anxiety. The majority of studies reported were not specifically designed to address anxiety as such. The exceptions were a number of single cases and small case series studies of psychological treatment of anxiety in BD, the findings of which suggest a reduction of anxiety symptoms. The main conclusion of the paper was the urgent need for a RCT to evaluate the benefits of psychotherapy for anxiety in the context of BD. Since this publication, a further paper has presented three case series in which Barlow’s unified therapy protocol for anxiety disorders was applied to three individuals with anxiety disorders comorbid with BD. 151 This indicated that, again, psychological therapy focused on anxiety was feasible, with promising outcomes in particular for two of the three participants. Again, the authors highlighted the importance of a RCT to properly evaluate the benefits of psychotherapy for anxiety in BD.

Limitations in current research

The main focus of research to date has concerned the prevalence and impact of anxiety disorders on BD. However, many individuals who do not meet full diagnostic criteria for anxiety comorbidity still experience high levels of anxiety symptoms. 18 Furthermore, both current anxiety disorder and higher levels of current anxiety symptoms (irrespective of anxiety diagnosis) are associated with worse illness course. 152 It is, therefore, a limitation of current literature that it has tended to focus more on diagnostic anxiety categories rather than experience of anxiety symptoms. The research described below therefore focuses on the latter to avoid excluding participants with BD with significant anxiety experiences that do not neatly fit with DSM-IV criteria. 76 Although there are now a number of preliminary non-RCT studies that have delivered psychological therapy for anxiety in BD, none of these adopted an integrated approach in which both bipolar and anxiety experiences were addressed. Additionally, the therapies so far proposed seem to have been developed in the absence of a detailed understanding of the experience of anxiety from the perspective of people living with BD or of their views of the type of treatment that they would prefer. In line with these issues, the research reported here places the experiences and preferences of SUs with anxiety and BD as central to the development of a novel psychological intervention and its evaluation in a RCT.

Aims

The research reported here has the following aims.

• To understand more about the impact anxiety has on the lives of people with BD.

• To understand what psychological help is required and by whom in relation to anxiety in people with BD.

• To develop and evaluate the feasibility and acceptability of CBT-informed time-limited therapy for people with anxiety and BD.

Overview of methods

Methods

Both phase 1 and phase 2 of this WS were approved by the NHS North West Research Ethics Committee reference (10/H1015/5) on 22 January 2010.

Sociodemographic and clinical data

The sociodemographic and clinical characteristics of the participants are shown in Table 30. Participants were, on average, 45.75 years old (SD 10.97 years; range 25–62 years) and mostly of white British ethnicity and > 60% were female. Despite the majority of participants having had experience of tertiary education, only 24% were in paid employment. Two-thirds of participants were either divorced or single, and three-quarters were parents.

All but one participant had a diagnosis of bipolar 1 disorder and time since diagnosis averaged > 10 years (mean age at bipolar diagnosis 34.14 years, SD 9.83 years). HAM-D score (mean 9.81, SD 7 .43) and MAS score (mean 1.57, SD 2.27) were consistent with participants being out of acute mood episode at interview. HADS-A scores (mean 11.19, SD 4.59) indicated that > 80% of participants had a current experience of at least mild anxiety. Consistent with this, most participants met the SCID criteria for at least one anxiety disorder (mean age at first mood disorder diagnosis 25.7 years, SD 10.65 years), the most common being GAD and social phobia. As nine participants had more than one anxiety disorder, the current anxiety disorders list in Table 30 totals more than 21.

Qualitative interviews

Participants were happy to discuss both their experiences of anxiety and their perspectives on relevant treatment. The quotations below all refer to responses directly related to experiences of anxiety, except where highlighted. The primary themes identified are (1) benefits and limitations of existing approaches, (2) therapeutic experiences and aspirations and (3) cognitive and behavioural approaches to managing anxiety. Each is described in turn and supported with illustrative quotations. Participant numbers and HADS-A score at the time of the interview are provided.

Methods

Focus group structure and format

At the beginning of each group, participants were introduced to the facilitators and informed of the purpose of the groups, which was to obtain their views on refining the structure, content and format of a planned, integrated, individual intervention for anxiety in BD. The group was told that the refinement of the treatment manual for the therapy would run in parallel with the focus groups, so the first group would take place as the manual was being drafted and the final group would discuss a summary of the near-final version of the manual, based on the previous groups. In addition to information from the focus groups, manual development was informed by a literature review, expert guidance and NICE guidelines for both anxiety and BD, as described in more detail below.

The topics covered in the focus groups included SU perspectives on:

• what content should be included in an individual intervention for anxiety in BD

• how an individual intervention of this type should be delivered

• what type and format of supporting resources should be offered in the intervention

• what potential barriers are to accessing this type of intervention, and how they might be overcome

• the ways in which an intervention of the type described might be helpful

• the potential strengths and weaknesses of the final therapy manual reviewed in session 3 based on input from sessions 1 and 2.

Results

These results provide a summary of the primary themes emerging from the focus groups. The participant number (P) and focus group number (FG) are included with quotations throughout. SU participants readily engaged with the focus group format and talked in detail about their views of the proposed approach and the contrast between this and previous experiences of therapy. There was a general view that offering an approach that was focused on integrating help with anxiety with people’s experience of bipolar mood experience was both novel and appropriate. Participants reported their frustrations at not being able to access this type of care in the past or, if they had managed to access it, the limitations of what had been offered.

Methods

This study was reviewed and approved by the UK NHS Ethics Committee process (REC reference 10/H1015/83). It was preregistered with International Standard Randomised Controlled Trial Number (ISRCTN) 84288072, and a detailed research protocol has been published. 159

Primary outcomes

Feasibility

Figure 18 presents the recruitment and retention rates for participants in AIBD and TAU arms in accordance with Consolidated Standards of Reporting Trials (CONSORT) criteria. 173 These figures and the additional information on participation rates for the AIBD intervention were used to inform assessments of feasibility. Of 122 people screened for eligibility, 14 declined, 9 were uncontactable and 27 did not meet the study’s inclusion criteria. Thus, only 11% of those offered the opportunity to participate in the study declined. Of the participants recruited, 32 (44%) were recruited by clinician referral and 40 (56%) were recruited via self-referral, with no evidence for differential eligibility or refusal rates in the AIBD versus TAU arms. The two arms did differ in the proportions coming from each referral source, with two-thirds of AIBD participants coming from self-referral and two-thirds of TAU participants coming from NHS referrals (Table 31).

FIGURE 18.

Recruitment and retention. Reproduced with permission from Jones et al. 157

TABLE 31 - Baseline sociodemographic and clinical characteristics of clinical sample phase 3

CMHT, Community Mental Health Team; GCSE, General Certificate of Secondary Education.

Adapted with permission from Jones et al. 157

Characteristic	Treatment arm, n (%)
	AIBD (N = 37)	TAU (N = 35)
Sex
Male	13 (35.1)	10 (28.6)
Female	24 (64.9)	25 (71.4)
Number of previous episodes
≤ 7	4 (10.8)	3 (8.6)
8–19	7 (18.9)	6 (17.1)
≥ 20	26 (70.3)	28 (74.3)
HAM-A score
≤ 17	35 (94.6)	29 (82.9)
18–24	1 (2.7)	4 (11.4)
> 24	1 (2.7)	2 (5.7)
HADS-A score
8–10 (mild)	11 (29.7)	9 (25.7)
11–15 (moderate)	16 (43.2)	15 (42.9)
16–21 (severe)	10 (27.0)	8 (22.9)
Missing	0 (0.0)	3 (8.6)
Current anxiety disorders
Social phobia	12 (32.4)	9 (25.7)
Agoraphobia	6 (16.2)	4 (11.4)
GAD	19 (51.4)	18 (51.4)
PTSD	10 (27.0)	7 (20.0)
Panic ± agoraphobia	9 (24.3)	6 (17.1)
OCD	5 (13.5)	6 (17.1)
Specific phobia	9 (24.3)	6 (17.1)
Number of current anxiety disorders
0	5 (13.5)	6 (17.1)
1	10 (27.0)	15 (42.9)
2	12 (32.4)	6 (17.1)
3	6 (16.2)	4 (11.4)
4	3 (8.1)	3 (8.6)
5	0 (0.0)	1 (2.9)
6	1 (2.7)	0 (0.0)
Bipolar status
BD 1	31 (83.8)	31 (88.6)
BD 2	6 (16.2)	4 (11.4)
Medication at baseline
Antidepressants	20 (54.1)	11 (31.4)
Mood stabilisers	23 (62.2)	26 (74.3)
Antipsychotics	18 (48.6)	17 (48.6)
NHS service use 6 months pre baseline
Inpatient stay days	4.27 (16.63)	0.91 (4.74)
CMHT appointments	7.19 (10.95)	6.17 (7.49)
Outpatient appointments	2.97 (4.81)	1.63 (3.48)
GP appointments	4.73 (4.17)	3.69 (3.15)
Source of referral
NHS	12 (37.5)	20 (62.5)
Self	25 (62.5)	15 (37.5)
Ethnicity
White British	33 (89.2)	32 (91.4)
Other white	0 (0.0)	1 (2.9)
Asian other	0 (0.0)	1 (2.9)
British Asian	2 (5.4)	1 (2.9)
Not stated	2 (5.4)	0 (0.0)
Marital status
Married or cohabiting	17 (45.9)	7 (20.0)
Divorced/annulled/separated	9 (24.3)	13 (37.1)
Never married	11 (29.7)	15 (42.9)
Number of children
0	15 (40.5)	15 (42.9)
1	7 (18.9)	4 (11.4)
2	8 (21.6)	5 (15.3)
3	7 (18.9)	7 (20.0)
4	0 (0.0)	3 (8.6)
≥ 5	0 (0.0)	1 (2.9)
Education
Years 7–11 (no GCSE)	3 (8.1)	0 (0.0)
GCSEs or equivalent	6 (16.2)	8 (22.9)
Further or higher education not completed	7 (18.9)	4 (11.5)
Further or higher education completed	16 (43.2)	16 (45.7)
Postgraduate education not completed	0 (0.0)	1 (2.9)
Postgraduate education completed	5 (13.5)	6 (17.1)
Working
No	21 (56.8)	23 (65.7)
Yes	16 (43.2)	12 (34.3)
Type of work
Employed full-time	5 (13.5)	4 (11.4)
Employed part-time	7 (18.9)	4 (11.4)
Voluntary	3 (8.1)	3 (8.6)
Self-employed	1 (2.7)	1 (2.9)
Unemployed	6 (16.2)	8 (22.9)
Sickness/disability benefit	11 (29.7)	12 (34.3)
Retired	3 (8.1)	2 (5.7)
Student	1 (2.7)	1 (2.9)

Seventy-two participants were randomised, fully meeting the prespecified recruitment target and representing 59% of those screened for eligibility. These participants were recruited over a period of 11 months, indicating an average recruitment rate of 6.5 per month. Retention was 88% (n = 64) to end of therapy follow-up (16 weeks), 75% (n = 54) to 32-week follow-up, 74% (n = 53) to 48- and 64-month follow-ups and 72% (n = 52) to the final 80-week follow-up. Two of the participants lost to follow-up died during the final follow-up period for reasons that were deemed by TSC/Data Monitoring Committee to be unrelated to the trial. Thus, retention rates across the trial were consistent with the feasibility target of 75% ± 10% up to 20 months as prespecified. 159

As for any large ongoing study, the main aim was to collect primary outcome data. As this was a pilot study, we were able to collect data on completion rates for the secondary self-report measures (Table 32). These will assist with developing subsequent studies by indicating the willingness of participants to complete such measures.

TABLE 32 - Completion of secondary measures by trial arm and assessment time point

n and % presented of those with primary outcome data.

Reproduced with permission from Jones et al. 157

Treatment arm	Time point, n (SD)
	Baselinea (0 weeks)	16 weeks	48 weeks	80 weeks
AIBD	36 (97.3)	28 (90.3)	20 (74.1)	20 (76.9)
TAU	34 (97.1)	27 (84.4)	22 (84.6)	22 (84.6)

As is evident from Figure 18, there is no evidence of resentful demoralisation in this trial, with retention rates at least as high in TAU as in the active intervention across assessment points.

Unblindings were reported in seven AIBD and four TAU participants (15.3% of the RCT total). Reasons for unblindings were participants in AIBD mentioning therapist by name to a RA, RAs being asked to amend therapy appointments, care co-ordinators’ accidental disclosure during risk assessment and therapist accidental disclosure. Unblindings in TAU arose from participant disclosure at follow-up interview, and knowledge that the TAU group was invited to undertake a secondary interview for another Spectrum Centre for Mental Health Research study (Everyday Momentary Observations of Thoughts and Emotions). For all 11 participants we were successfully able to reallocate an alternative blinded RA to carry out future follow-up interviews.

Participants

The mean age of the participants in both groups was > 40 years (AIBD: 45.5 years, SD 10.7 years; TAU: 42.9 years, SD 16.6 years) and most were female (see Table 2; n = 49, 68%). Although all participants had a HADS score of > 8, consistent with anxiety cases,85,174 the scores at baseline on the HAM-A were very variable, ranging from 0 to 33 (normal range to severe), and on average scoring in the upper levels of the normal range (AIBD: 12.84, SD 3.26; TAU: 12.97, SD 3.75). An exploration of anxiety disorders in the recruited participants indicated that 85% met the diagnostic criteria for at least one anxiety disorder, GAD and social phobia being the most common in both arms. More participants in the AIBD arm met the criteria for two or more anxiety disorders (59% vs. 40%). Consistent with this, numerical rates of social phobia, agoraphobia, PTSD, panic ± agoraphobia and specific phobia were all higher in the AIBD arm, although these differences were not statistically significant. The AIBD arm also reported higher rates of service use across all areas than did the TAU arm.

Participants predominantly had a chronic course of BD, with > 70% of participants having had more than 20 episodes of mania or depression at baseline. The majority of people had a diagnosis of BD 1 (n = 62, 86%). The majority were either divorced or never married (n = 48, 67%) and most were parents (n = 42, 58%). Although the majority of participants had at least commenced further education (n = 55, 76%), only just over one-third were in employment (n = 28, 39%). Of those employed, the highest proportion was in part-time work (n = 11 of 28, 39%). Of those not in work, the majority were in receipt of sickness/disability benefits (n = 23 of 44, 52%). Comparing the treatment arm with the control arm, there were no substantial differences in age, sex, number of previous episodes, bipolar status, ethnicity, number of children or education. Twice as many participants in the intervention arm were married or cohabiting, and approximately 10% more intervention group participants were in work. Two-thirds of referrals in the intervention arm were self-referrals into the study, compared with only one-third of self-referrals in the control arm. The remainder were referred by the NHS from a mix of clinicians, clinical studies approach to clients at clinics and invitation letter from GPs. Self-referrals were in response to adverts in the local press and radio, through Pendulum (the magazine of Bipolar UK) and participants’ local support groups, and from newsletters from Spectrum Centre for Mental Health Research at Lancaster University.

Treatment delivered and treatment fidelity

Participants offered AIBD were offered up to 10 sessions. The mean number of sessions attended was 7.7 (SD 2.8), with only two participants not attending any sessions. The remainder attended between 1 and 10 sessions, with 84% (n = 31) attending more than three sessions and 59% (n = 22) attending at least nine sessions.

Adherence to the therapy protocol and CTS-R was assessed independently for 26 randomly selected therapy recordings across early, mid and late sessions (sessions 1 and 2, sessions 4 and 5, and sessions 7–10, respectively). Adherence to the treatment protocol was 92% across the appropriate therapeutic stages (initial, formulation, intermediate, later and throughout). The mean score for CTS-R was 47.1 (SD 5.03), and all recordings met threshold criteria for competence. 175 Mean CTS-R showed little change across sessions: early, 47.5 (SD 5.74); mid, 47.4 (SD 5.24); and late, 46.1 (SD 3.76).

Client ratings of therapy

Therapists asked each client who received AIBD to rate the therapy on two 0–10 scales. The first was to rate how useful they thought their experience of the therapy had been from 0 (not at all) to 10 (extremely); the second was to rate whether or not they would be likely to recommend the therapy to anyone else with similar problems from 0 (definitely not) to 10 (definitely yes). These data were available for 21 participants. They were missing for nine participants and not collected from the remainder due to dropout, failure to attend or, in one case, death of the participant (for reasons unconnected to the trial).

Ratings for their own experience of therapy averaged a score of 9.14 (range 8–10). The average rating for the likelihood of recommending the therapy to others was 9.26 (range 8–10). Thus, for those who completed the ratings, no participants provided anything less than a high (≥ 8) rating for the helpfulness of the therapy and the extent to which they were likely to recommend it to someone else.

Therapeutic alliance

Therapeutic alliance was assessed using the client short-form 12-item version of the WAI, which was completed by participants after sessions 3 and 10 and scored out of a total of 28 (four items each using a seven-point Likert scale, with two items reverse coded). At session 3, agreement on the goals of treatment were similar between all three therapists whose clients returned fully completed WAI questionnaires (score range 23.3–26.7, n = 23/37); agreement on how to achieve these goals (task agreement) ranged from 22.7 to 24.6; and the development of a bond between participant and therapist ranged from 22.8 to 23.3. These scores summed provided overall scores out of 84, ranging from 68.8 to 74.3.

By session 10, 16 participants had fully completed the measure. One of our therapists had left by this point, and their participants had been assigned to one of the two remaining therapists, which might have had an impact on the alliance scores. Agreement on the goals of treatment ranged from 23.7 to 24.4; agreement on how to achieve these goals (task agreement) ranged from 23.7 to 24.0; and, as expected over a period of time, the bond score between participant and therapist had increased to range from 23.3 to 25.2. These scores summed provided overall scores of 70.8–73.6 out of 84.

Qualitative interview findings regarding acceptability

A number of key themes emerged from a preliminary analysis of in-depth qualitative interviews with 17 participants who had received AIBD. Participants indicated that they valued the intervention and contrasted this with other previous forms of support they had received. This had led many to commit wholeheartedly to the therapy:

And it is the only therapy I have had in all my . . . I have had anxiety since I was what, about 15 and that is the only thing that worked for me was that CBT therapy.

AN003

. . . the [therapy was] a real gift, and so I took it with you know both hands, six toes and my teeth because it was a gift.

AN011

Participants also recognised value in treating anxiety and BD together and contrasted this with previous experiences of having had these problems addressed separately, which they perceived as leading to a more superficial approach to both difficulties:

Normally people do them separately and trying to put them together when you are ill is just . . . not easy at all . . . if you have got them separate it’s like skirting round each issue, but putting them together and showing a person that deal with them all, and will go slowly over everything so you know what to expect, it is so much better, definitely.

AN007

The AIBD intervention, at 10 sessions over 4 months, was relatively brief compared with other forms of structured psychological therapy for significant mental health problems. For some participants, this was helpful in providing a clear structure with clear personally identified targets to work towards:

. . . it was 10 sessions, and that were it, but the 10 sessions meant that the goals that were laid out at session one, those were the goals that were worked at, and those were the goals that were achieved by the end and that is a far better way of working.

AN001

However, others felt that they would have preferred the therapy to continue for longer while they felt that they were benefiting from it, although it was not evident that specific elements were omitted from the therapy:

I remember thinking I didn’t want them to come to an end. But I couldn’t have told you, or couldn’t have probably pinpointed aspects that I needed more to cover I think I was just . . . gaining so much that I kind of felt maybe there was more to gain as well you know if I had carried on.

AN008

Another finding identified related to participant perceptions of outcomes following therapy. Participants identified that the coping strategies they had learnt in AIBD helped them escape the social isolation and functional limitations that anxiety had historically imposed on them. It was also noted that people still experienced challenges in the course of their BD, but that they felt more confident in dealing with these, including by using the self-management strategies learnt during therapy:

. . . it has been crippling for me over the years. I have had a number of breakdowns, and each time it’s always been the anxiety that has kept me prisoner in my own home, it has stopped me from socialising, and progressing so this time, I have healed better and with coping strategies that have allowed me to do things, a lot quicker than before.

AN008

. . . prior to becoming part of this study I thought my bipolar disorder owned me . . . and was in control of me and I had no control over it . . . and through this study, now I own it and I control it . . . I had wrote my life off . . . I had consigned myself to a life on benefits and you know I have got this mental illness and I am on major drugs and antipsychotic you know it doesn’t make me feel good . . . and I had just resigned myself to it, even though you know in the past I have had relatively good success in whatever endeavours I have undertaken . . . through this study and through the work that I did with [research assistant] and [therapist] it give me, renewed belief and . . . made me take control of my life again.

AN001

Secondary outcomes

Outcomes are described below with respect to 48-week follow-up, as submitted with the original application, and also 80 weeks (extended to capture mood relapse more accurately).

Self- and observer-reported outcome measures

Key clinical outcomes for the AIBD trial were self- and observer-rated anxiety symptoms, subsyndromal mood symptoms, recovery (Tables 33 and 34) and time to next mood episode (Figures 19–21).

TABLE 33 - Summary of continuous clinical outcome measures by time in AIBD and TAU groups

Adapted with permission from Jones et al. 157

Measure	Week	Treatment arm
		AIBD						TAU
		Mean	SD	Median	Minimum	Maximum	n	Mean	SD	Median	Minimum	Maximum	n
STAI-State	0	47.1	11.2	47	25	70	36	48.0	15.1	48	20	79	35
	16	44.3	14.3	48	22	67	30	47.7	11.0	49	23	69	31
	48	40.8	12.2	38	21	60	23	44.5	14.2	43	21	70	24
	80	48.7	15.0	51	23	69	24	41.7	15.1	41	21	68	24
STAI-Trait	0	53.7	8.2	54	32	66	37	53.4	11.3	53	30	71	35
	16	51.5	11.4	54	30	72	30	52.6	9.9	51	35	77	31
	48	47.9	12.6	48	29	72	23	50.3	10.9	51	32	80	24
	80	49.6	11.2	50	34	70	24	46.3	13.1	43	23	79	23
HAM-D	0	8.7	5.1	8	0	21	37	8.0	5.5	7	0	23	35
	16	10.1	7.8	8	0	27	31	10.5	7.7	8	1	34	31
	32	9.2	7.5	7	1	27	25	8.9	7.6	8	0	29	27
	48	8.7	7.9	5	0	25	24	7.9	6.3	8	0	22	25
	64	9.8	7.0	9	0	25	26	7.4	8.6	6	0	30	23
	80	9.1	7.0	9	0	26	27	8.7	8.0	7	0	31	26
HAM-A	0	9.1	6.1	8	0	33	37	9.8	7.8	7	0	31	35
	16	10.3	9.2	9	0	30	31	11.0	7.7	9	1	31	31
	32	9.7	8.1	9	0	26	25	10.8	9.9	8	0	37	27
	48	10.2	9.1	7	0	34	24	10.8	8.8	8	0	37	25
	64	11.6	8.1	11	0	28	26	11.1	11.4	7	0	38	23
	80	12.3	9.1	10	0	31	27	11.3	10.6	9	0	45	26
MAS	0	2.2	2.4	2	0	8	37	2.5	2.9	2	0	11	35
	16	2.1	2.5	1	0	10	31	2.2	2.1	1	0	8	31
	32	1.5	2.9	1	0	14	25	2.6	4.1	1	0	18	27
	48	3.4	5.0	1	0	15	24	2.0	3.0	1	0	12	25
	64	2.9	3.3	3	0	16	26	3.7	4.9	2	0	20	23
	80	2.4	3.7	1	0	16	27	4.8	5.8	3	0	20	26
BRQ	0	2107	374	2097	1109	2901	36	2161	405	2090	1525	2909	34
	16	2248	411	2158	1458	3066	31	2146	401	2221	1272	2828	27
	48	2326	374	2210	1723	3073	21	2351	480	2379	1296	3252	23
	80	2318	379	2225	1733	2967	20	2398	497	2428	1450	3290	23

TABLE 34 - Repeated measures analyses

Differences are between AIBD and TAU adjusting for sex, number of previous bipolar episodes, baseline HAM-A subscale, the baseline score, and the interaction between the baseline score and discrete time.

Model 1 includes a time by treatment interaction.

Model 2 omits the interaction.

Adapted with permission from Jones et al. 157

Outcome measure/model	Week	Treatment effecta	95% CI	p-value
STAI-State
Model 1b	16	–2.39	–7.45 to 2.67	0.355
	48	–2.50	–8.11 to 3.11	0.382
	80	8.49	1.58 to 15.41	0.016
				0.001
STAI-Trait
Model 1	16	–0.98	–5.16 to 3.20	0.645
	48	–1.37	–6.64 to 3.90	0.611
	80	3.70	–1.75 to 9.15	0.184
				0.072
Model 2c		–0.40	–4.38 to 3.57	0.842
HAM-A
Model 1	16	–0.09	–3.33 to 3.16	0.959
	62	–0.30	–4.00 to 3.39	0.872
	48	0.71	–3.38 to 4.80	0.734
	64	1.43	–3.06 to 5.92	0.533
	80	1.88	–2.59 to 6.35	0.409
				0.798
Model 2		0.25	–2.44 to 2.93	0.858
HAM-D
Model 1	16	–0.22	–3.74 to 3.30	0.901
	32	1.28	–2.45 to 5.00	0.501
	48	1.68	–2.17 to 5.53	0.393
	64	2.53	–1.55 to 6.61	0.225
	80	0.91	–2.68 to 4.49	0.620
				0.861
Model 2		0.98	–1.46 to 3.42	0.431
MAS
Model 1	16	0.05	–1.04 to 1.14	0.928
	32	–0.63	–2.25 to 1.00	0.450
	48	1.84	–0.26 to 3.94	0.086
	64	–0.89	–3.20 to 1.42	0.449
	80	–2.14	–4.64 to 0.36	0.094
				0.032
BRQ
Model 1	16	138.40	–33.06 to 309.80	0.114
	48	–22.90	–199.90 to 154.10	0.800
	80	–86.80	–269.10 to 95.80	0.352
				0.063
Model 2		47.40	–93.10 to 187.90	0.508

FIGURE 19.

Kaplan–Meier estimates of time to first depression- or mania-type bipolar episode. Reproduced with permission from Jones et al. 157

FIGURE 20.

Kaplan–Meier estimates of time to first depression-type bipolar episode. Reproduced with permission from Jones et al. 157

FIGURE 21.

Kaplan–Meier estimates of time to first mania-type bipolar episode.

State–Trait Anxiety Inventory-State scores at baseline for both arms were elevated compared with age-matched norms,166 consistent with the intention to target an anxiety subgroup (norm means 35.88–36.03; SD 10.52–11.07). The mean score fell within the highest 20% of scores (AIBD 87th percentile; TAU 88th percentile) found among age-matched general population controls166 and were similar to the normative score for a diagnosed anxiety group (without BD) in the STAI manual (mean 49.02; SD 11.62). STAI-Trait scores were higher than those for STAI-State and substantially elevated compared with age-matched norms (mean 35.03–35.06; SD 8.88–9.31), with comparable SDs. Both arms fell within the highest 15% of scores (both 93rd percentile) of age-matched general population controls and were substantially higher than normative scores for a diagnosed anxiety group (mean 48.08; SD 10.65). These elevated scores are consistent with a successful HADS screening procedure to include individuals with clinical levels of anxiety symptoms. For both STAI-State and STAI-Trait, the trajectory of the reduction in STAI score from baseline to the 48-week follow-up was clearly steeper in the AIBD group than in the TAU group. This was the planned end point in the original application. However, this apparent gain did not persist to the 80-week follow-up.

Observer-rated anxiety assessment was conducted using the HAM-A. At baseline, both group means on the HAM-A were very low, consistent with a very mild level of anxiety (mild anxiety > 8; Matza et al. 176). Levels of observer-rated anxiety remained low throughout the follow-up period in both arms, never exceeding the mild anxiety threshold.

Observer-rated depression scores were determined using the HAM-D. 82 According to cut-offs determined by Zimmerman et al. ,177 participants were at the low end of mild depression at baseline in both arms (mild depression 8–16) and remained so throughout the trial.

Observer-rated mania scores were determined using the MAS. Bech178 defined absence of mania as a score of ≤ 6. Scores for both arms of the trial were extremely low throughout the study, consistent with participants being recruited during euthymia. The only exception was in the TAU group at weeks 64 and 80, when MAS increased above 4, although this remains within the ‘no mania’ range.

Self-reported recovery scores were measured using the BRQ. At baseline, recovery scores were substantially below those reported in the development paper for this measure98 (mean 2357.7, SD 414.0). The BRQ scores increased during therapy in the AIBD arm, in contrast to the TAU arm, but by 48 and 80 weeks the mean scores in both arms were comparable to those in the development sample previously noted.

In terms of relapse overall, 37% of the TAU group and 43% of the AIBD group had relapsed by the 48-week follow-up, and at the end of the final follow-up (80 weeks) the corresponding figures were 49% and 59% respectively. When relapses were compared by type of episode, it was found that the proportion who had experienced a depressive relapse was similar in the two groups, both at 48 weeks (31% of the TAU group and 32% of the AIBD group) and at 80 weeks (40% of the TAU group and 43% of the AIBD group). Likewise, at 48 weeks, 17% of the TAU group and 24% of the AIBD group had experienced a manic relapse, and by 80 weeks 31% of the TAU group and 32% of the AIBD group had experienced a manic relapse.

As expected for a feasibility study, the estimates of difference between the groups were very imprecise. As the CIs crossed 1 for the Cox model HRs, these were all non-significant. For the repeated measures analysis of the STAI-State outcome, the treatment by discrete time interaction was significant, with the estimated mean scores lower at weeks 16 and 48, but 8.49 units higher at week 80 in the AIBD arm than in the TAU arm.

Phase 1

Semistructured interviews were conducted with 21 individuals with experience of BD and anxiety. The primary themes that emerged from the data highlighted:

1. ‘Benefits and challenges of existing approaches’, in particular the limitations of trying to address anxiety with medication alone and the lack of information and integrated therapy approaches to anxiety and bipolar in the experience of participants.

2. ‘Therapeutic experiences and aspirations’ indicated the importance of clinicians recognising how serious anxiety problems are for some people and the need for the appropriate treatment at the right time. In terms of psychological therapy, participants wanted an approach that was sensitive to their individual experiences and a therapist who was genuinely collaborative rather than didactic. Supplementing therapy sessions with information in an accessible booklet format was also highlighted.

3. ‘Self-management strategies for anxiety’. Participants mainly described approaches they had discovered over time rather than through formal therapy. The primary approaches identified were behavioural approaches (such as relaxation or distraction). Although participants understood that cognitive factors were potentially important, most had not developed strategies to address unhelpful thoughts in the absence of structured psychological therapy.

These results highlight that individuals living with anxiety alongside their BD regard this as a serious problem, find current therapeutic options inadequate and would like access to structured psychological care to address this to supplement their own self-management strategies. Consistent with Frank et al.,18 even participants who did not have a current formal anxiety diagnosis were experiencing a significant impact from their anxiety symptoms. Current NICE guidelines1 indicate that individuals with BD should be offered anxiety intervention when required, based on NICE anxiety guidelines. 147,148 However, as the participants indicated in their interviews, treating anxiety without regard to the influence of BD is unsatisfactory in a number ways; these relate to the timing of therapy, the understanding of how anxiety and bipolar symptoms interact, and the primary importance of flexible collaborative approaches to engaging clients.

In terms of psychological therapy, participants wanted an approach in which there was due regard paid to the impact of anxiety on their lives, a focus on formulating individuals unique experiences and a therapy style that was collaborative and non-didactic. In contrast with findings that highlight ambivalence towards treatment in individuals with BD,179 participants expressed a clear interest in help of this type supplemented by additional materials in booklet form (or similar accessible format). Liebert180 has recently argued that the apparent ambivalence in individuals diagnosed with BD is commonly a response to the person feeling as though their actual experiences are undervalued and to inadequate and/or inappropriate treatment. Thus, in the current study, participants were clear about the core elements of an intervention that they would feel motivated to engage with. The individualised approach suggested by participants is consistent with other research into BD which, for instance, highlighted the potential benefits of a novel intervention to enhance personal recovery in BD which engaged directly with key personal goals rather than clinician-presumed therapy targets. 98,181

The majority of participants had not received psychological therapy targeting their anxiety symptoms. Despite or perhaps because of this, a range of self-management strategies was described. Behavioural strategies such as relaxation were reported to be quite effective when instigated in the early stages of an anxiety episode. Consistent with these reports, NICE guidance for GAD149 continues to recommend the use of guided relaxation as an effective intervention. However, in contrast, very few participants were able to use more cognitive strategies despite understanding that patterns of thinking were linked to their anxiety episodes. As cognitive approaches are also recognised as important in NICE guidance for anxiety disorder generally,1,147,148 this gap underscores the importance of offering people access to psychological help that can help them to identify and implement effective cognitive strategies to help manage their anxiety.

Phase 2

A series of three focus groups were conducted to elicit SU views on a planned integrated individual intervention for anxiety in the context of BD. Across the groups participants discussed the following.

1. Content: support with respect to both anxiety and BD, including individualised information, problem-solving and relaxation techniques, capitalising on current successful coping and tackling rather than avoiding problematic situations. The importance of cognitive as well as behavioural strategies was noted, and participants felt that it would be helpful to have an intervention with clear structure and monitoring of progress through this.

2. Focus: participants indicated that knowing something about the therapist as a person was important for engagement, as was instilling hope that progress could be made. Conversely, it was indicated that it would be helpful if the therapy was about developing skills that could be used beyond the intervention period, rather than there being a rush to resolve all anxiety issues within therapy. Such skills should build on client assets as well as aiming to remedy any deficits that the person was finding problematic.

3. Format: participants wanted the location and timings of therapy as well as therapy goals to be individually negotiated with each client. It was also noted that clients were likely to engage honestly with a therapist who makes an effort to communicate in ‘normal’ rather than ‘expert’ language and uses humour where this is appropriate. Likewise, participants wanted a therapeutic environment that felt safe so that new skills could be practised in a supportive environment in which immediate success was neither required nor expected. There was general agreement that an individual approach was preferable for this intervention, but that this should not exclude family or the care co-ordinator from specific sessions that the client wished them to attend.

4. Support materials: participants indicated that it would be helpful to have material that provided an overview of therapy along with worksheets to summarise each session. Although there was some disagreement about whether or not materials should be held on a website or in booklet format, there was clear consensus that materials should be well presented and include coping focused accounts from other individuals with experience of BD and anxiety.

5. Barriers and facilitators: a key issue here was accessibility. Based on prior experience of long waiting lists, participants highlighted the importance of a time-limited approach that could be delivered by a range of MHPs to increase the likelihood that those who needed the intervention could access it in a timely manner.

The information from phases 1 and 2 was broadly consistent in highlighting that individuals with BD and anxiety wanted access to a flexible, individualised, coping-focused psychological intervention that provided them with both cognitive and behavioural strategies to address their anxiety issues in the context of their BD. These views overlap with the limited research evidence to date, which indicates that the most promising psychological interventions for anxiety in BD are likely to be cognitive behavioural. 150,182 In addition, consistent with the views expressed by participants on the limitations of current treatments, there is a dearth of empirical evidence concerning the effectiveness of benzodiazepenes, mood stabilisers or antipsychotics in addressing anxiety in the context of BD. 182 These findings, therefore, converge on supporting the rationale for the evaluation of the new therapy in phase 3.

Phase 3

Strengths and limitations of the research

The qualitative research in phase 1 of this WS highlighted the importance of developing an integrated psychological therapy approach for people with BD and comorbid anxiety. Although the sample size was reasonable for a thematic analysis, the study reflects the particular views of a subgroup of individuals from north-west England. Thus, although the findings were relevant to both treatment development in this WS and informing consideration of other approaches more widely, it would be sensible to explore these issues in more ethnically diverse groups and individuals earlier in their course of BD. Phase 2 had similar strengths and weaknesses in providing rich and detailed therapy development information but from a limited number of patients and clinicians in a particular region of the UK. The work conducted employed established focus group principles,154 and therefore such limitations are generally characteristic of this type of investigation. It seems that the information provided in phases 1 and 2 led to the development of an intervention (AIBD) that was feasible and acceptable based on data collected in phase 3. As a feasibility and acceptability study, the sample size was relatively small and not powered to detect statistically significant differences in clinical effects of the therapy. As such, estimates of clinical outcomes were imprecise with large standard errors. The sample was, however, adequate to provide evidence for the key feasibility and acceptability outcomes and to provide formative information about potential adaptations for a larger trial. A further definitive trial of AIBD would need to be conducted using a trial of sufficient power to reliably detect clinically significant differences in clinical outcomes.

Clinical and research implications

Information from phases 1 and 2 indicates that there is demand for improved psychological therapy for people with BD and anxiety and that this needs to be delivered with due recognition of the bipolar experiences people have had. This is relevant not only to further development of AIBD, but also to the consideration of treatment planning in NHS services.

Overall, the results from phase 3 indicate that the trial was successful in demonstrating feasibility and acceptability in that our selection and recruitment procedures worked, people took up the opportunity for therapy when this was offered to them, and retention to the trial was good. Secondary clinical data present a mixed picture, with most promising outcomes in self-reported anxiety ratings. This suggests that a future larger trial should explore anxiety outcomes in more detail. The impact on relapse and mood was expected to be less direct given the focus of the therapy on anxiety issues. It is therefore likely that a future, more definitive, trial would focus on a briefer follow-up to capture anxiety changes, with the need for lengthy follow-up required to monitor time to relapse. As most RCTs of CBT for primary anxiety disorder report a maximum follow-up period of 48 months, it is envisaged that this would be the appropriate duration for a future AIBD trial. 186 Although the intervention was generally well received by participants, some interviewed expressed a desire for more sessions to help generalisation of therapy. For a future trial, it would therefore be sensible to consider the potential to extend AIBD to 12 to15 sessions, in line with NICE guidance for high-intensity interventions for primary anxiety disorder. 149 If successful, such a definitive trial would provide the first evidence-based integrated therapy for anxiety in BD. This would be a timely addition to the therapeutic options for BD given the recognition of the importance of anxiety in recently revised NICE guidelines for BD57 and the prospective national roll-out of the Improving Access to Psychological Therapies Severe Mental Illness programme.

Prevalence of substance misuse, including alcohol, in bipolar disorder

Substance use is common in BD, with lifetime prevalence estimates ranging from 50% to 60% for BD 1. 5,187 Substance use rates are significantly higher than in the general population16 and those with other axis 1 psychiatric disorders. 188,189 For example, Schaffer et al. 190 reported past-year problematic substance use disorder (SUD) of 29% for BD and of 14.3% for major depressive disorders in a population-based sample. This high comorbidity in BD is clinically relevant because individuals with BD who use substances tend to haveworse social and neurocognitive functioning,191–193 a more severe clinical course194,195 and poorer treatment outcomes. 196,197

Alcohol is the most commonly abused substance among individuals with comorbid BD and SUD,187 with prevalence rates of 30–35% reported for alcohol use disorder in BD. 198 Comorbid alcohol misuse is associated with more severe mood disturbance199 and high levels of subsyndromal symptom,200 which are themselves associated with increased relapse risk and greater risks of suicide and violence. 201 Comorbid cannabis misuse is also common, occurring with a lifetime prevalence of 20–25% in BD. 13 Cannabis misuse is associated with earlier onset of BD, longer duration of mood episodes, greater illness severity, worse treatment adherence, greater disability and higher risk of suicide. 202–205

Even moderate alcohol consumption appears to be associated with higher risk of relapse and higher current mood symptoms. 199 Higher rates of impulsivity,206–209 suicide,191,200,210 criminality211 and violence201 are also observed for those with BD and comorbid alcohol misuse.

Available interventions for substance use in bipolar disorder

Despite the high prevalence of substance misuse in BD, and in particular comorbid cannabis and alcohol problems,5,13 there have been few developments in integrated therapy. To date there have been three published RCTs addressing this issue. Schmitz et al. 212 compared 12 weeks of individual CBT with a medication-monitoring control intervention and observed no differences between groups on substance use outcomes during treatment, but some improvement in mood. No data were collected at follow-up to explore the persistence of the mood effect. Weiss et al. 38,213 conducted two RCT evaluations of their integrated therapy compared with group counselling. Integrated group therapy delivered over 12–20 weeks led to improvements in substance abuse outcomes following treatment at 3 months’ post-treatment follow-up, particularly with respect to alcohol use. However, the findings with respect to mood were more mixed, with one trial indicating a worsening of depressive symptoms following treatment and the other indicating a reduction in risk of mood episodes and improved overall clinical outcome.

There have been no studies to date that have integrated MI with CBT approaches. This is an important omission, as ambivalence to treatment is common in BD. 180,214 There is clear evidence for the benefits of MI in addressing alcohol and drug issues in a range of client groups,215,216 as well as its particular impact of addressing treatment ambivalence. 217 By not addressing this issue, previous trials have focused on clients who are already motivated to change their substance use, therefore potentially excluding those with higher levels of clinical need. The work described in this stream therefore reports on the development of a therapy approach that integrated MI approaches to engage clients and achieve commitment to change with CBT approaches to provide tools for change and relapse prevention (MI-CBT) intended to be both inclusive and integrated, not excluding potential clients due to ambivalence.

Aims

The research reported here has the following aims:

• to understand more about the experience and impacts of cannabis and alcohol use in BD

• to develop an integrated therapy approach in partnership with SUs to address alcohol misuse (including in individuals who are currently ambivalent about their use of alcohol)

• to evaluate the feasibility and acceptability of MI-CBT for alcohol use in BD with respect to recruitment, retention and feedback from participants.

Phase 1

Q-Sort methodology was used to explore experiences and consequences of substance use in BD. This drew on a review of the literature, a qualitative evaluation conducted previously51 and analysis of client therapy tapes from our pilot work in preparation for this stream218 to develop the Q-concourse. Fifty individuals reporting either alcohol or cannabis use problems sorted the concourse statements in accordance with Q-methodology, first for experiences and second for consequences.

Phase 2

A series of focus group meetings were held with either SUs or health-care professionals. Focus groups were conducted in line with the principles described by Rabiee. 154 Both the SU and the professional focus groups were structured around eliciting information on prior experiences of therapy for alcohol issues, obtaining feedback on plans for the new intervention and views on format, style and timing of the intervention planned. The draft manual discussed in phase 2 was informed from a review of current literature, the phase 1 results and the therapy manual employed for our pilot work before this programme.

Phase 3

A single-blind RCT was conducted to determine the feasibility and acceptability of the integrated intervention compared with TAU. MI-CBT was delivered over 20 sessions over a period of 6 months, with participants followed up at 12 months post randomisation. Primary outcomes of the study were the feasibility and acceptability of MI-CBT (recruitment to target, retention to follow-up, absence of untoward incidents). We have also conducted preliminary analysis of alcohol and mood outcomes (frequency and severity of alcohol use and time to mood relapse).

Material and methods

Procedure

Each participant sorted first the 41 reasons statements and subsequently the 40 after-effects statements. A response grid (Figure 22) was provided on which participants could sort the statements in line with the following instructions.

FIGURE 22.

Q-Sort response matrix.

Reasons statements

Order the statements onto the response grid according to your reasons for using substance [X] by placing the statements that you feel apply to you the most on the right hand side of the grid, and the statements you feel apply to you the least on the left hand side of the grid.

Effects statements

Order the statements onto the response grid according to experiences you may have had in the 24 hours following using substance [X] by placing the statements that you feel apply to you the most on the right hand side of the grid, and the statements you feel apply to you the least on the left hand side of the grid.

For each ranking, the same procedure was applied. After constructing three sorts of statements (applies to me, does not apply to me and neutral), each participant identified the strongest ‘applies to me’ statement and placed it in the far right space on the response grid, then identified the two statements which ranked next highest in terms of relevance or importance to them, and placed these in the next column. This process was repeated until all statements from the ‘applies to me’ pile were placed on the grid. The participants were then asked to consider their ‘does not apply to me’ pile and repeat the process, starting with the statements they felt applied to them the least, placing statements on the opposite side of the response grid. They were then asked to consider any statements in their ‘neutral’ pile and place them accordingly in the centre of the grid. Following sorting, participants were given the opportunity to review their ranking before the final sort was recorded for the study.

Data analysis

An analysis of the Q-Sorts was performed using a dedicated software package (PQ Method, version 2.3, Schmolck, Munich, Germany) and employing principal components analysis with Varimax rotation. The resulting correlational matrix reveals relationships between individuals rather than between items and allows for the identification of factor exemplars to define each factor. PQ Method merges these factor exemplars using the weighted average of all the sorts that load significantly on each factor to produce a factor array. Chi-squared, t-tests and one-way analyses of variance were used to examine differences between participants loading on factors in terms of demographic (age, sex, marital status, living arrangements, education and employment status), clinical (number of previous manic/depressed episodes and current mood ratings) and substance use (MPS, SCID diagnosis, number of days use in past month and mean number of years use of MPS) characteristics.

Participant demographics

Fifty individuals meeting the DSM-IV criteria for bipolar affective disorder 1 (n = 48) or 2 (n = 2) participated in the study, of whom 58% (n = 29) reported alcohol as the MPS and 42% (n = 21) reported cannabis as the MPS. For the alcohol MPS participants, 79% met the SCID criteria for current alcohol abuse or dependence. For the cannabis MPS participants, 57% met the criteria for current cannabis dependence and 14% met the criteria for cannabis abuse. Participants reported having used their MPS for an average of 21 days in the past month (SD 8.65 days in the past month). The time participants had used the substance before the study ranged from 3 months to 30 years.

The sample comprised 28 men (56%) and 22 women (44%), and the mean age of 40.2 years (SD 12.12 years). Most participants were of white British origin (94%), were unmarried (82%) and lived alone (54%); the remainder (46%) lived with others (partners, friends, children or others). The majority had completed further education (60%), but most were currently unemployed (80%).

Reasons for use

Principal components analysis of the Q-Sorts describing reasons for use resulted in a two-factor solution with 48 participants loaded on one or other of these factors (26 factor 1; 22 factor 2), and 41% of the variance explained (Table 35). Two participants’ sorts did not load on either of the factors.

Group differences between factors

Statistical tests were employed to explore whether there were any differences between participants loading on factor 1 (mood management) and factor 2 (social enhancement) in terms of demographic, clinical and substance use variables. The majority of group 1 (mood management) were cannabis users (58%), whereas only 23% of group 2 (social enhancement) were cannabis users [χ² (1) = 5.99; p = 0.01]. There was a trend for group 1 participants to have experienced more manic episodes [χ² (2) = 6.18; p = 0.06].

After-effects of use

Principal components analysis of the Q-Sorts relating to the after-effects of substance use resulted in a three-factor solution (Table 36). Of the 50 participants who completed this sort, 48 loaded on one of these three factors (25 factor 1; 16 factor 2; 7 factor 3), and 49% of the variance was explained.