Peer support for discharge from inpatient to community mental health care: the ENRICH research programme

Discharge from psychiatric inpatient care

People recently discharged from psychiatric hospital often fail to continue with treatment,1 relapse and are readmitted. 2 For example, in the UK it has been reported that 36% of inpatients with psychotic disorders are readmitted within 1 year of discharge,3 while in New Zealand the readmission rate of all psychiatric inpatients has been reported at 41%. 4 A high proportion of people are readmitted shortly after discharge, with an Australian study of over 35,000 people admitted to psychiatric hospital finding that nearly a fifth of all psychiatric inpatients (half of all people readmitted) were readmitted in the first month post discharge,5 and a UK study of nearly 8000 people observing 15% readmissions within 3 months of discharge. 6 People recently discharged are also at risk of suicide and self-harm with, globally, suicide rates in the first 3 months after psychiatric discharge 100 times the suicide rate in the general population, and suicidal thoughts and behaviours 200 times the level in the general population. 7 In the UK, post-discharge suicides are most frequent in the first week after discharge, with 15% of all suicides nationally among people within 3 months of a psychiatric discharge. 8 Evidence suggests that lack of follow-up care post discharge9 and lack of continuity of care from hospital to community are predictors of early readmissions,10 with, conversely, higher levels of follow-up from community mental health services post discharge predicting lower readmission rates. 11 Systematic review evidence suggests that interventions supporting transition from inpatient to community mental health care are feasible and likely to be cost-effective. 9 Two of the studies reviewed evaluated multidisciplinary transitional discharge interventions that included peer support components. 12,13

Peer support in mental health services

Origins of peer support in mental health services are twofold. First, there is a decades-old tradition of self-help in Europe and North America, underpinned by a ‘helper therapy principle’14 that is perhaps best exemplified by the 12-step programmes of Alcoholics Anonymous and related programmes, having more recently proliferated into mental health care with the emergence of organisations such as Depression Alliance and the Hearing Voices Network. Second, peer support work in mental health grew out of reactions to negative and coercive mental health treatment15 with, by the 1970s, people building programmes focused on self-advocacy and campaigning where people focused on helping each other, acknowledging that their experiences and life stories could be a source of important knowledge. Since then peer support has gradually moved into the mainstream of mental health service provision, with state-funded and approved mental health service providers increasingly employing and training peer support workers – people with their own experiences of mental distress and of using mental health services – to role-model living well with mental illness and embody hope in the future, and to improve the ‘recovery focus’ of mental health services. 16 Peer workers (PWs) have been employed in a variety of roles in the NHS in the UK, most typically as peer healthcare assistants in inpatient settings or as peer community support workers in Community Mental Health Teams (CMHTs), with a focus on telling their ‘recovery story’ alongside enacting a more conventional support worker role. 17

Peer support for discharge

An observational pilot of a transitional discharge intervention delivered wholly by PWs has shown promise,18 while comparison group studies of community-based peer support programmes have also reported reductions in readmissions19 and longer community tenure20 compared to traditional services alone. A pilot trial of a community-based peer mentoring intervention for people with a history of recurrent psychiatric hospitalisation reported fewer readmissions for people receiving peer mentoring over a 9-month period, compared to community treatment as usual. 20 In another trial, among participants who engaged with treatment, fewer people receiving assertive community treatment (ACT) from a consumer-staffed ACT team reported being hospitalised than those receiving care from a non-consumer ACT team, although length of follow-up varied between participants and was not reported. 21

The evidence base for peer support in mental health services

While this evidence suggests that interventions employing PWs might offer a strategy for improving the outcomes of discharge, meta-analyses of trials of a range of peer support interventions have indicated little difference in outcomes for people receiving peer support, either in comparison to treatment as usual or to similar support provided by other mental health workers, once data from across trials are pooled. 22,23 Those systematic reviews have also pointed to the heterogeneity of the peer support evaluated, issues with the quality of trials, an absence of formal studies of cost-effectiveness, and a lack of reporting of how peer support is designed to bring about change in comparison to other forms of mental health support. In particular, inadequate randomisation and sequence generation processes,22 lack of blinding of assessors, risk of bias resulting from missing data, and selective or incomplete reporting of outcomes measured were identified as trial quality issues that need to be addressed. 23

Processes and principles of peer support

A wider literature, including qualitative and observational research, has indicated that organisational factors supporting the implementation of peer support into practice in mental health services might impact on its effectiveness. Issues such as clarity of job description,24,25 access to appropriate training and support,26 shared expectations of the PW role25,27 and preparation and training for the team that will be working alongside PWs28 have all been identified as important facilitators of the introduction of the PW role. It has been noted how the distinctiveness of peer support, in comparison to other forms of mental health care, can be lost in a formal environment of statutory mental health services if those organisational conditions are not met. 27,29,30

The distinctiveness of peer support has been attributed to the particular qualities of peer-to-peer relationships in contrast to the conventional clinician-patient relationships: peer-to-peer relationships are underpinned by a sense of connection between peers based on a recognition of shared experiences;31 reciprocity in the relationship whereby both parties learn from each other;32 and the validation and exchange of experiential, rather than professionally-acquired, knowledge. 33 In previous research by the team we developed a general change model for peer support in mental health services31 that underpins, theoretically, the development and implementation of both our peer support for discharge intervention and our evaluation. Elsewhere, a more recent review of peer support in mental health services concluded that a lack of attention to fidelity to the core principles underpinning peer support limits the usefulness of current peer support research to policy makers and practitioners. 34

Impact of peer support on peer workers

A parallel body of research has identified potential benefits and challenges for people working in a PW capacity. It has been suggested that peer support might enhance personal recovery for PWs, but can also be a source of stress, with mental health professionals voicing concern that the PWs they work alongside might relapse and be hospitalised because of the stresses of the role. 16 A review of qualitative research about PWs’ experiences of peer support indicated improvements in confidence, self-esteem and social contacts for PWs. 35 However, more recently, a survey of 84 PWs working in a range of mental health services in one state in the United States (US) USA found that PWs experienced difficulties including poor financial compensation, limited employment opportunities, work stress, the emotional stress of helping others and in maintaining personal wellness, with 44% of the sample reporting having experienced a relapse in their mental health while working as a PW. 36 It is important to understand and evaluate the impact of working in a peer support capacity in order to ensure that implementation of peer support into mental health services is both safe and rewarding for PWs.

Rationale

Mental health policy and workforce initiatives in higher income countries are driving the introduction of PWs into statutory mental health services despite these uncertainties in the evidence base. 22,23 As such, there is a need for high-quality trials of peer support that specify, model and evaluate the distinctive processes whereby peer support brings about change in specific contexts and settings, as well as for good health economic evaluation. This study aims to address those limitations in the current evidence base while providing clear evidence of the effectiveness and cost-effectiveness of a PW intervention to enhance discharge from inpatient to community mental health care.

Aims

The overarching aim of this programme of applied research was to enhance the experience of discharge from psychiatric inpatient care for mental health service users, preventing unnecessary readmission and reducing costs. The specific aim of the programme was to manualise, pilot and trial a peer support intervention to enhance discharge from inpatient to community mental health care, significantly reducing readmissions and the associated cost of care. The detailed research objectives of the programme were:

1. to refine an empirically and theoretically grounded model that explains how peer support interventions impact on outcomes for service users post-discharge

2. to develop and manualise a peer support intervention to enhance discharge

3. to develop an index to assess the fidelity of peer support interventions

4. to conduct a high-quality randomised controlled trial (RCT) of the intervention

5. to establish the effectiveness and cost-effectiveness of a peer support intervention to enhance discharge

6. to explore the impact on PWs of providing peer support for discharge.

This programme also aimed to deliver applied outputs that will enable mental health service providers in the NHS to effectively and cost-effectively integrate peer support into the mental health discharge pathway, including: the ENRICH training programme and intervention handbook, providing implementation guidance for managers and commissioners in the NHS, and our principles-based peer support fidelity index.

Work packages

The ENRICH programme comprised five interlinked work packages (WP) that addressed the aims listed above (see Figures 1 and 2).

FIGURE 1.

Research pathway diagram.

FIGURE 2.

Overview of the programme.

Setting

The programme took place in adult acute psychiatric inpatient and community mental health service settings in mental health NHS trusts in England. Two trusts [South West London (SWL) & St George’s Mental Health NHS Trust and East London (ELN) NHS Foundation Trust] took part in the pilot study, with five additional trusts joining for the full trial and subsequent WPs (Central and North West London NHS Foundation Trust; Sussex Partnership NHS Foundation Trust; Surrey and Borders Partnership NHS Foundation Trust; Birmingham and Solihull Mental Health NHS Foundation Trust; Bradford District Care NHS Foundation Trust). Two other trusts – Dorset HealthCare University NHS Foundation Trust and South West Yorkshire Partnership NHS Foundation Trust – were involved in WP1 of the programme but were unable to remain involved for the full trial.

Population

The primary population of the research was adult inpatients of psychiatric hospital wards (acute admission wards and their equivalents) who had had at least one previous admission in the preceding 2 years. Secondary populations included the PWs employed to implement the intervention and their supervisors (PWC).

Ethical approval

A single application for NHS research ethics approval was made for all WPs. The programme was approved by the UK National Research Ethics Service, Research Ethics Committee London – London Bridge on 10 May 2016, reference number 16/LO/0470.

Governance

The Chief Investigator, Programme Manager and trial statistician reported regularly to an independent Programme Steering Committee (PSC) and an independent Data Management and Ethics Committee (DMEC). Oversight from a patient and public perspective was provided by a Lived Experience Advisory Panel (LEAP) as well as service user representation on the PSC and DMEC.

Changes to the funded programme

Changes to the funded programme were minimal. In WP1 we split the systematic review into two outputs. We initially prioritised data extraction of components of peer support to inform development of the intervention. We returned to the review and updated searches once the WP4 trial had completed recruitment (and more researcher resource was available). This was to our advantage as a sufficient number of new trials of peer support had been published in the interim to enable us to undertake a meta-analysis.

There were no major changes to the WP4 trial that required ethical approval. Minor changes to recruitment procedures were made following the pilot trial that did require amendment to ethical approval.

As noted in Process evaluation – predictors of engagement with peer support below, because we did not observe a positive effect on primary outcome, we were not able to undertake the structured equation modelling we had intended, to explore pre-specified change mechanisms. Given that our complier average causal effect (CACE) analysis had indicated a relationship between engagement with intervention and outcome, we instead undertook an analysis of pre- and post-randomisation predictors of engagement. Similarly, in Process evaluation – peer support and change in mental health services – the qualitative component of the process evaluation – instead of exploring our pre-specified mechanisms we used our qualitative dataset to refine the change model for peer support in mental health services that underpinned the intervention (thus remaining consistent with our original objectives).

White et al.38

We produced a second review, addressing questions 2 and 3 above (outcomes and process of peer support). Work on this second review was delayed while we undertook the main trial, with searches subsequently undertaken up to 13 June 2019. The full method for the search is described in the paper. 38 The search identified 23 papers reporting 19 RCTs comparing a range of one-to-one peer support interventions with either CAU or an active control arm. Fourteen trials provided data for meta-analysis of nine different outcomes. We found that one-to-one peer support in mental health services has a small but statistically significant benefit for individual recovery [standardised mean difference (SMD) 0.22, 95% confidence interval (CI) 0.01 to 0.42; p = 0.042] and empowerment (SMD 0.23, 95% CI 0.04 to 0.42; p = 0.020). There was no effect on clinical outcomes such as symptoms or hospitalisation; the risk of being hospitalised was reduced by 14% for those receiving peer support but was non-significant [risk ratio (RR) 0.86, 95% CI 0.66 to 1.13].

In subgroup analyses we explored the extent to which processes of peer support impact change in outcome; we compared the effect on outcome of peer support provided in addition to CAU with peer support where PWs acted as ‘substitutes’ for other workers providing similar support, and compared the effect of peer support interventions where there were high levels of support for PWs with interventions where support for PWs was low. This indicated that peer support had a significantly greater effect on social network support (Qint = 4.27, p = 0.039) where peer support was delivered in addition to CAU (SMD = 0.23) compared to when it was provided as a substitute intervention (SMD = −0.30). We note in the review the continued heterogeneity in peer support interventions and in outcomes assessed in trials, and the importance of ensuring that the constructs that are assessed relate to the mechanisms and processes of the peer support that is being evaluated. 38

Marks et al.39

The intervention was developed in three sequential stages – (1) generating intervention components; (2) producing the intervention handbook; (3) piloting the intervention. Experiential knowledge was integrated through the development process, with several members of the research team identifying as service user or survivor researchers, or working as PWs. Our LEAP and a Local Advisory Group (LAG) at each study site also included people with experiences of using mental health services and peer support (see Table 3). Development closely followed the Medical Research Council complex interventions guidance. 40 The methods and results are summarised in Appendix 1, including the first stage of the systematic review identifying components of peer support interventions.

Gillard et al.42

An iterative process of developing and testing the fidelity index is described in detail in this report. A draft index was developed using expert panels including service user researchers and people doing peer support, with fidelity criteria generated for each of the peer support principles identified in WP1 and index items written as a means of testing criteria. Two rounds of testing took place in 24 mental health services providing peer support in a range of settings. Fidelity was assessed through interviews with PWs, their supervisors (PWCs) and people receiving peer support. Index items were rated by members of the research team, with responses tested for spread and internal consistency, and independently double rated for inter-rater reliability. Feedback from interviewees and service user researchers was used to refine both the content of the index and the rating procedure.

A fidelity index for one-to-one peer support in mental health services was produced with good psychometric properties. Fidelity is assessed in four principles-based domains – building trusting relationships based on shared lived experience; reciprocity and mutuality; leadership, choice and control; building strengths and making connections to community – and is separately assessed for the set-up and ongoing delivery of peer support, while an overall score can also be generated.

We conclude that the index offers the potential to improve the evidence base for peer support in mental health services, enabling future trials to assess fidelity of interventions to peer support principles, and giving service providers a means of ensuring that peer support retains its distinctive qualities as it is introduced into mental health services. We suggest that further testing of the internal structure of the index is necessary to fully establish the psychometric properties of the index, for example through a confirmatory factor analysis, and testing the construct validity for the index by exploring the relationship between fidelity scores and outcomes that might be expected to be associated with fidelity.

Gillard et al.41

The study was a parallel, two-group, individually randomised controlled superiority trial, with trial personnel (outcome assessors, data analysts) masked to allocation of participants. The study took place in the inpatient and community mental health services of seven NHS mental health trusts in England (sites). In five sites where participants were recruited from two inpatient facilities, and one site where recruitment was from three facilities, these were treated as single sites for stratification. A detailed trial protocol has been published;41 trial procedures are summarised only here.

The trial is registered with the ISRCTN clinical trial register, number ISRCTN 10043328,43 and was overseen by an independent steering committee and a data monitoring committee.

All new admissions to participating adult acute inpatient wards were screened for eligibility. Inpatients were eligible if they had at least one psychiatric admission in the preceding 2 years, were aged 18 years or older, were assessed by the ward clinical team as likely to be discharged within the next month and had capacity to give written informed consent to participate in the research. Patients were excluded if they had a diagnosis of any organic mental disorder, had a primary diagnosis of eating disorders, learning disability or drug or alcohol dependency (as recorded in clinical notes), or were assessed by the clinical team on the ward as presenting a current, substantial risk to a PW. Following completion of baseline assessments, consenting patients were randomised to treatment groups in a 1 : 1 ratio, stratified by site and diagnostic group (psychotic disorders, personality disorders, other eligible non-psychotic disorders). Measures to ensure protection of masking of assessors are detailed in the study protocol. 41

Participants in the intervention group received the peer support intervention, delivered one-to-one by a designated PW, a Discharge Information Pack and CAU at discharge. Participants were assigned a PW after allocation and prior to discharge by the site PWC who also provided support and supervision to the PW team. The peer support intervention, including the PW training programme, is specified in a handbook and described in detail in the trial protocol41 and summarised in Co-design and consensus work to develop the peer support intervention above. The Discharge Information Pack provided information about potentially useful statutory and voluntary sector services. CAU post-discharge from inpatient psychiatric care is mandated nationally in England as follow-up by community mental health services or primary care mental health team within 7 days of discharge. Within a week of discharge, a discharge summary should be sent by the inpatient team to the patient’s GP and others involved in developing their care plan, including information about why the patient was admitted and how their condition has changed during the hospital stay. A member of the CMHT or primary care mental health team to which the patient has been discharged will typically telephone or visit the patient within 1 week of discharge to make arrangements for their ongoing care. Participants in the control group received CAU and a copy of the Discharge Information Pack to control for any effect of access to information alone on outcome. We conducted the trial in sites where there was no offer of one-to-one peer support in either inpatient or community settings. In some sites group peer support was on offer, typically as peer-facilitated support groups in inpatient settings that people could ‘drop in’ to on an ad hoc basis. It is possible that participants in either peer support or control groups accessed group peer support. Stratification of randomisation by site was used in part to address this.

Data were collected at baseline (T0), and at 4 months (T1) and 12 months (T2) post-discharge from the index admission. Mental health service use data were extracted from the site EPR at baseline and 12 months. The primary outcome for the trial was psychiatric inpatient readmission in the 12 months post-discharge, including both formal and informal admissions. Secondary outcomes were number of voluntary admissions, involuntary admissions and total number of admissions, total number of days in hospital, time to first readmission, use of accident and emergency (A&E) services for a psychiatric emergency (measured as number of episodes of liaison psychiatry contact), and number of contacts with crisis resolution and home treatment teams in the year post discharge, plus standardised measures of psychiatric symptom levels, subjective quality of life, social inclusion, hope for the future and strength of social network, measured at end of intervention (4 months post discharge). Adherence to the intervention was assessed using a structured online survey completed by PWs following each contact. Serious adverse events (SAEs), as defined in the trial protocol,41 were recorded and followed up until the end of the 12-month follow-up.

We required a sample of 530 participants, allocated on a 1 : 1 ratio, to detect a reduction of 12% in readmission (from 34% to 22%) in the intervention group compared to the CAU group, with 80% power at the 5% significance level. This calculation allowed for clustering by PW in the intervention group only,44 assuming an intracluster correlation of 0.05 with an average cluster size of 10 participants. We inflated the sample size by 10% to allow for missing primary outcome data at follow-up,45 resulting in a final sample size of 590. All analyses were conducted according to the intention-to-treat (ITT) principle, meaning that all randomised participants with a recorded outcome were included in the analysis and analysed according to the group to which they were randomised. We also estimated the CACE for the intervention on the primary outcome46 (where compliers were participants who had at least two PW meetings, at least one of which was in the community following discharge). The CACE was estimated with a two-stage estimation procedure. In the first stage, a logistic regression of treatment receipt regressed on randomisation was conducted. In the second stage, a Poisson regression of the outcome on treatment receipt was conducted. The analysis was adjusted for the same covariates as the ITT analysis. A bootstrap (1000 samples) was used to obtain bias corrected and accelerated CIs. Subgroup analyses for the primary outcome were pre-specified: ethnicity (any black ethnicity, all other ethnicities); primary diagnosis at index admission (psychotic disorders, personality disorders, other eligible disorders); first language (English, other). Full details of the analyses undertaken are given in the trial protocol. 41

Gillard et al.47

Results for the trial are given here pending peer review and publication of the main trial paper. 47 We note that recruitment into the full trial at additional sites was delayed while we waited for approval for the checkpoint report at end of the pilot trial, and then as some sites secured approvals to employ PWs locally. A seventh site (in addition to the original six) was opened to support recruitment, with recruitment taking 6 months longer than originally planned. The target of 590 participants was successfully recruited, with flow of participants into the study given in Figure 11. Participant characteristics were well-balanced between groups (see Tables 3 and 4).

FIGURE 11.

Flow of participants in the ENRICH trial.

In the PW group, 136 (47.4%) participants were readmitted to psychiatric inpatient care within 12 months post-index admission, and 146 (50.2%) in the CAU group (see Table 5). The unadjusted risk difference was −3% (95% CI −0.11 to 0.05; p = 0.5070) in favour of the peer support group. The adjusted relative risk of readmission in the ITT analysis was 0.97 (95% CI 0.82 to 1.14; p = 0.6777), and the adjusted odds ratio (OR) was 0.93 (95% CI 0.66 to 1.30). In the CACE analysis, the relative risk of readmission according to the natural indirect effect (RR 0.88, 95% CI 0.76 to 0.99) was lower than from the ITT analysis and was significant. In subgroup analyses, for participants of any black ethnicity the adjusted OR of readmission was 0.40 (95% CI 0.17 to 0.94), while for any other ethnicity the OR was 1.12 (95% CI 0.77 to 1.63; interaction p = 0.0305). No other subgroup effects were found (see Table 6).

Observed differences in the secondary outcomes collected at 4 months were small and none were statistically significant (see Table 7). There were no statistically significant differences between the groups in any of the secondary outcomes assessed at 12 months (see Table 8). Adherence to the intervention was assessable in 268 (91.2%) participants with a mean of 1.8 [standard deviation (SD) = 2.9) face-to-face contacts with a PW in hospital, 4.4 (SD = 4.6) post discharge. Assessors were unmasked by 52/306 (17.0%) patients revealing their allocation during collection of secondary outcome data at 4 months (38 in the peer support group, 14 in the CAU group).

There was a total of 67 SAEs reported in the trial (34 in the peer support group, 33 in the CAU group) from 51 participants (26 in the peer support group, 25 in the CAU group). One SAE in the peer support group, an incident of self-harm, was reported as related to the intervention. Number and type of SAEs included 12 deaths, none of which were reported as related to the study.

The trial indicated that one-to-one peer support for discharge from psychiatric inpatient care, offered in addition to CAU, did not have a significant effect on readmission when compared to CAU only. There were neither significant effects on any secondary service use outcomes in the year following discharge nor any reduction in symptom severity or benefit to psychosocial outcomes after 4 months. The CACE analysis indicated that a smaller proportion of participants in the peer support group who received a pre-defined minimal amount of the intervention were readmitted than participants in the control group who – according to the analysis – would also have received the minimal amount of peer support if such support had been offered to them. Additionally, compared to CAU, participants of any black ethnicity in the peer support group were significantly less likely to be readmitted than participants of any other ethnicity (see Table 6).

The study reflects findings from our systematic review of one-to-one peer support in a range of mental health service settings which suggests that, on the basis of pooled data from trials to date, peer support is unlikely to have an effect on psychiatric hospital admission, length of stay in hospital or clinical severity. 38 Engagement in the peer support intervention was low. However, the findings of the CACE analysis, suggesting that participants who engaged better may have benefited from the intervention, raised the question as to why engagement with the intervention was not more complete. We were encouraged that black participants were significantly less likely to be readmitted in the year post-discharge compared to control than those of any other ethnicity (see Table 6). While we need to better understand how and why peer support seemed to work better for this group of participants, this finding offers hope that peer support might help address persistent inequalities. 48,49

Predictors of readmission

A secondary aim of WP4 was to explore predictors of readmission across our sample. Of the 590 participants recruited to the study, EPR data, which provided all service use data, was available for 578 (98%). Of these 578 participants, 282 (49%) were readmitted within 1 year post discharge. Fifty-nine (10%) were readmitted in 30 days or less, 123 (21%) in 90 days or less.

The following variables were tested as predictors of readmission (binary outcome) and time to readmission (time to event outcome): age, gender, ethnicity, diagnosis, SIX, BPRS, Manchester Short Assessment of Quality of Life (MANSA), number of admissions in the previous year, number of compulsory admissions in the previous year, number of A&E attendances in the previous year, compulsory or voluntary index admission, length of index admission. Univariate analysis was carried out using logistic regression to estimate the relationship between each potential predictor and the readmission outcome. Those predictors found to be statistically significant (significance level = 5%) univariately were entered into a multiple logistic regression model. This strategy was followed for the time to readmission outcome using Cox regression.

Diagnosis, BPRS, MANSA, number of admissions in the previous year and number of A&E attendances in the previous year were found to be univariately associated with both outcomes. Number of compulsory admissions in the previous year and type of index admission (compulsory or voluntary) were also significantly associated with time to readmission. However, both variables had more than 10% missing data and so were not entered into the final regression models. All univariate analyses are reported in Report Supplementary Material 2.

In the multiple logistic regression model (n = 529) readmission was associated with higher BPRS (OR 1.03, 95% CI 1.01 to 1.05; p = 0.009) and number of admissions in the previous year (OR 1.32, 95% CI 1.10 to 1.57; p = 0.002). In the multiple Cox regression model (n = 529) shorter time to readmission was associated with higher BPRS [hazard ratio (HR) = 1.01, 95% CI 1.00 to 1.03; p = 0.031] and number of admissions in the previous year (HR=1.24, 95% CI 1.14 to 1.38; p < 0.001). These findings are in line with previous similar analyses (see Table 9). 50,51

Stage 1: Generating intervention components

The systematic review and expert workshops were used to identify potential components, following an intervention mapping approach. 68 Components were given a short label and a descriptor, and mapped on to five domains: (1) Recruitment and Role Description; (2) Training; (3) Delivery; (4) Supervision and Support; (5) Organisation and Team.

Systematic review. Procedures for the review are as described in Systematic review of one-to-one peer support in mental health services above. 38 For the purposes of intervention development, papers reporting studies of any design were included, from database inception until end of April 2015, where they reported description of intervention components. In addition, grey literature – unpublished evaluations and experiential testimonies – were identified using a snowball approach through emails to contacts known to be working in peer support. A member of the LEAP screened each article with decisions checked by the first author. Data detailing peer support components were extracted from included studies and coded to the five intervention domains.

Expert workshops. Workshops were held with the LEAP and the research team to suggest potential components for the intervention. A third workshop was held with five members of the research team to consider how a taxonomy of behaviour change techniques (BCT)69 might be relevant to peer support in mental health services. Again, components were mapped on to the five domains, and where similar components were identified from different sources these were coded together.

Stage 2: Producing the intervention handbook

Prioritising components. LAGs were convened in each of six study sites (mental health trusts). LAGs prioritised components using a card sort exercise structured into the five domains. 70 A large visual grid was produced with five columns labelled with the five domains and cards were printed for each component comprising a name and brief definition. Components identified by three or more sources in stage 1 (e.g. LEAP, team and review) were considered core to the intervention and those cards were placed within relevant domains on the grid. LAGs were given the remaining set of cards and invited to discuss the relevance and importance of each in the context of peer support and mental health service delivery locally. LAGs were asked to prioritise and identify up to five more components to add to each domain of the grid. Notes were made of LAG discussions, including the rationale for prioritising components. The research team produced a single grid based on the output from all LAGs. Components were discounted from further discussion (not included on the grid) if not prioritised by any LAGs, added as core components (added to the grid) if prioritised by a majority of LAGs, or otherwise retained for further discussion.

Producing and refining the intervention ‘blueprint’. Using output from the LAGs we produced a blueprint of the intervention in the form of a flow diagram, specifying the processes of recruiting PWs, training, delivering the intervention, and support received by PWs. The blueprint included all components retained for discussion, as well as the full set of core components, so that local implementation issues could be considered. In a second round of meetings, LAGs were presented with the flow diagram and invited to discuss the blueprint using well-established talk-aloud approaches. 71 Discussion began at the top left of the flow diagram (recruiting PWs) and proceeded sequentially, including consideration of alternative pathways as appropriate, with LAGs to consider the appropriateness of each component, reflecting on practicalities of implementing and supporting the intervention locally. Again, notes were made of all discussions.

Drafting the intervention handbook. The output of LAG meetings was used to draft the ENRICH intervention handbook and PW training programme. Development of the intervention was also informed by our ‘peer support principles’ (see below). Further workshops with the LEAP and research team were held to inform writing the handbook and training content.

Stage 3: Piloting the intervention

A pilot randomised controlled trial of the intervention was conducted in two study sites to test feasibility of delivering trial procedures and implementation of the intervention (see WP3 below). Following the pilot, feedback workshops were held with the PWCs who trained and supervised PW teams at both sites, and the PWs who had delivered the peer support at one site, exploring their experiences and views on what worked well and what might be improved about the training and other aspects of the peer support. Workshops were semi-structured discussions rather than formal focus groups, inviting participants to feedback on any issues arising in their experiences of delivering the peer support. Changes were made to the handbook and training programme, based on the feedback, following a further research team workshop.

Stage 1: Generating intervention components

A total of 3800 studies were identified in the literature search, of which 97 were included in the review, 85 peer-reviewed and 12 from grey literature (see Appendix 1, Figure 13). Components generated by the literature review and expert workshops were mapped on to the five intervention domains (see Appendix 1, Table 11). Forty-four components were identified in the review (including 6 from grey literature), 29 by the LEAP, 37 by the research team and 6 from the BCT workshop; a total of 66 distinct components, once similar components were combined. Twelve core components, identified by three or more sources, are identified with an asterisk in Appendix 1, Table 11.

FIGURE 13.

Flow diagram of studies in the WP1 literature review.

Stage 2: Producing the intervention handbook

Prioritising components. Following the first round of LAGs, 6 components were discounted, 19 added as core components (see Appendix 1, Table 12), and 29 retained for further discussion.

Producing and refining the intervention blueprint. The flow diagram used in the walkthrough exercises in the second round of LAGs is shown in Appendix 1, Figure 14.

FIGURE 14.

Flow diagram depicting intervention blueprint.

Drafting the intervention handbook. A detailed handbook was produced specifying a full set of procedures defining PW and the PWC roles, recruitment process, training, support and supervision for PWs, and how the peer support is delivered in hospital and in the community (see Appendix 1, Table 13).

An 8-day, manualised training programme was developed, underpinned by 10 knowledge and skill sets derived from components prioritised to the training domain in stage 2, and the peer support principles. 57 The resulting training matrix (see Appendix 1, Figure 15) guided writing and development of training materials. Each training day comprised session plans, slides, handouts, exercises and other materials.

FIGURE 15.

ENRICH training matrix.

Stage 3: Piloting the intervention

Five PWCs involved in delivering training at both pilot sites attended a feedback workshop, with one PWC providing individual feedback. Four of five PWs at one site attended a workshop. Feedback was used to make amendments to the training and aspects of the handbook on recruitment of PWs and post-training support (see Appendix 1, Figure 16).

FIGURE 16.

Feedback from the pilot trial.

Service contacts

The primary health economic analysis and secondary cost-effectiveness analysis at 4 months were carried out using EPRs data supplied by participating mental health trusts. This included all recorded NHS hospital- and community-based mental health service contacts and psychiatric bed days utilised by trial participants up to 12 months post discharge from the index hospital admission and over the 12 months prior to the index admission. The PW contact log was used to record face-to-face contacts between PWs and trial participants.

For the secondary analysis of costs relating to wider service use, a version of the Adult Service Use Schedule (AD-SUS)72 was administered to collate trial participant self-reported contacts with services not delivered by NHS mental health providers, including: primary care, hospital for non-psychiatric reasons, social worker, voluntary sector services, private therapists and contacts with the police for any reason, including number of times placed in custody. The AD-SUS was administered at baseline with reference to the 3-month period prior to index admission; and at 4 months post discharge reference for that period.

Unit costs

Unit costs required for costing service contacts were obtained from the Unit Costs of Health and Social Care published annually by the Personal Social Services Research Unit73 and NHS Reference costs. 74 The unit cost of police contacts and police custody were based on costs reported by Heslin and colleagues. 75

An estimate of the cost per minute of face-to-face contact between peer support workers and trial participants was developed specifically for this study. Activity logs relating to a sample period of 10 working days were completed by PWs and the PWCs. These were used to identify the ratio between face-to-face and non-face-to-face time allocated to a specific patient, including time spent by PWCs training PWs, and subsequently to estimate a cost per hour of face-to-face contact inclusive of costs of training and non-face-to-face contact time. PW and PWC time was costed using the appropriate corresponding NHS salary grading (reported with relevant salary ‘on-costs’ in the Unit Costs of Health and Social Care). 73 Additional overhead costs and capital costs related to employment were applied proportionately in accordance with unit costs reported for similar professionals in the Unit Costs of Health and Social Care. 74

Cost and outcome measurement

The total cost per trial participant was calculated as the sum of costs attributed to all community- and hospital-based contacts made over 12 months from the date of discharge from the index admission (for the primary cost analysis) or over 4 months post discharge date (for the secondary cost-effectiveness analysis). Costs were also split into those relating to: CMHT contacts; psychiatric bed days; crisis team and A&E contacts (summed together); and all other service contacts.

For the secondary analysis of wider service costs over 4 months post hospital discharge, the number of self-reported contacts for each trial participant was multiplied by the relevant unit cost. For this analysis, costs were divided into total community-based service contacts and the total cost of contacts made with the police, including the cost of police custody. Costs of peer support were excluded from the secondary analysis of wider service costs. Health-related quality of life outcomes converted to a 4 month ‘QALY’ equivalent were measured using the EuroQol-5 Dimensions, five-level version (EQ-5D-5L) instrument. 52 Following recent guidance issued by National Institute of Health and Care Excellence (NICE),76 health states reported by trial participants were mapped back on to the health state valuation system applicable to the 3-level version of the EQ-5D. 77 The cross-mapping utilised the algorithm reported by Van Hout and colleagues. 78

Analysis

All analyses were conducted on a complete case basis and restricted to randomised participants included in the analysis samples at 4 months and 12 months used for the clinical evaluation (see Appendix 2, Table 17). Missing data from EPRs on service contacts and other patient covariates was not a serious problem and, in line with the main clinical analysis, we assumed that it was ignorable and likely to be missing at random. Missing data on self-reported service contacts and quality of life at 4 months were more problematic due to patients not being interviewed at follow-up – as self-report data is likely to be not missing at random, we did not seek to impute values where data were missing. For the primary analysis (12-month cost of mental health service contacts) complete cost data were available for n = 537 participants (91% of those randomised); complete health state utility data required for a complete case cost-effectiveness analysis were available for n = 289 participants (49% of those randomised); and for secondary analysis of the cost of wider service contacts, complete data on the cost of community service contacts at 4 months were available for n = 291 participants (49% of those randomised) and n = 293 (50% randomised participants) for cost of police contacts.

Multivariate modelling: costs and health state utilities

The estimation of differences in the mean total cost of mental health service use over 12 months (for the primary analysis) and over 4 months post discharge (for the secondary cost-effectiveness analysis) for intervention and control participants was undertaken using GLM with a logarithmic link function. 53 Total cost for each period was regressed on a dummy variable identifying treatment allocation (1 = Peer support; base category = usual care) and additional ‘baseline’ covariates: ethnicity; trial site; diagnostic group; total cost of mental health service contacts over 12 months prior to the index admission. The sub-categories of cost defined above were also modelled (over 12 months only) using the same approach. Wider service costs over 4 months were modelled similarly with total community-based costs and total costs of police contacts modelled separately. Utility scores applicable to EuroQol-5 Dimensions, three-level version (EQ-5D-3L) health states were also analysed using GLM framework using a log-link. Covariates included ethnicity, diagnostic group, number of previous psychiatric admissions, trial site and baseline utility scores.

Cost-effectiveness analysis at 4 months: combining costs and QALY outcomes

A ‘net benefit’ framework was used54 to evaluate whether the peer support intervention offered a cost-effective alternative to usual care after 4 months post discharge from the index hospital admission. For the ‘base case’ analysis, we adopted a cost-effectiveness threshold of £20,000 per QALY gained. This is the lower end of the range currently used by NICE (£20,000 to £30,000) when assessing health programme cost-effectiveness.

Uncertainty in cost-effectiveness analysis

For the primary analysis of costs over the 12 month follow-up, non-parametric bootstrap sampling was performed to facilitate an evaluation of uncertainty in relation to the estimated incremental effect of the PW intervention on total costs over 12 months arising from trial sampling error. The probability of drawing an incorrect conclusion about cost impact of peer support based on this expected value was determined with reference to the proportion of the bootstrap incremental values that are either negative or positive. We also present 95% confidence limits using values at the 97.5th and 2.5th percentiles of the distribution.

Primary analysis

Appendix 2, Table 14 presents descriptive statistics for the cost of mental health service contacts measured over 12 months post discharge from hospital separately for intervention and control participants, including the cost of participant contacts with peer support workers. A positive mean cost of PW support is shown for the control arm of the trial on account of two trial participants receiving the intervention in error.

Appendix 2, Table 15 contains the main results from the primary analysis carried out on total costs of mental health service contacts over 12 months, inclusive of the costs of contacts with peer support workers in the intervention group. Adjusting for baseline covariates, exposure to peer support was associated with a reduction in mean total costs of £2631 (95% CI −£21,546 to £3845) – this amounted to a 10% reduction in mean total costs over 12 months post hospital discharge compared to usual care (95% CI −31% to 15%). Given sampling uncertainty there was an estimated 82% probability that peer support was associated with lower total costs over 12 months (or an 18% chance that usual care was the lower cost alternative).

Multivariate analysis undertaken on sub-categories of cost suggests that most of the cost reduction associated with peer support arose from reduced cost of bed day utilisation (estimated reduction in bed day costs of −£3765, 95% CI −£9696 to −£2167), although peer support was also associated with an 8% reduction in the costs of CMHT contacts (95% CI −21% to 7%).

Sensitivity analysis

Estimated treatment effects using multi-level (GLM) random intercepts model resulted in only marginal differences in the standard errors estimated around treatment effects and the predicted difference in mean total costs between intervention and control participants.

Two additional post hoc sensitivity analyses were also conducted on the primary analysis of 12-month total costs. First, we evaluated whether the exclusion of the estimated reduction in costs of community mental team contacts associated with peer support made any difference to the estimated differences in total cost. While there was a small reduction in effect of peer support on total cost over 12 months when CMHT costs were excluded, much of the intervention effect on mean total costs over 12 months was retained, again suggesting that most of the overall effect on total costs can explained by a reduction in bed day utilisation.

Second, we tested the sensitivity of the estimated reduction in total costs over 12 months to the exclusion of bed day costs above the 95th percentile of the cost distribution. As total costs have a distribution with a long right tail (a comparatively small percentage of participants generating a disproportionate volume of cost over follow-up), it is plausible that the observed effect of peer support at the sample mean will be driven by prevention of lengthier and more costly psychiatric admissions. We found that exclusion of the participants in the top 5% of bed day costs substantially diminished estimated reduction in mean total costs associated with peer support to £549 (compared to the ‘full’ effect of −£2631).

Secondary analysis

Appendix 2, Table 16 presents the baseline covariate adjusted EQ-5D-3L health state utility values for intervention and control participants at 4 months post hospital discharge. These differences amount to a gain of just 0.002 QALYs over 4 months for the intervention over the control group (95% CI −0.01 to 0.01) – an improvement equivalent to less than one extra day spent in full health. As at 12 months, exposure to peer support was associated with lower total costs of mental health service contacts over 4 months (again inclusive of the incremental costs associated with peer support) – a reduction in mean costs of −£2066 (95% CI −£24,158 to £1155).

Appendix 2, Table 17 presents the combined analysis of QALY and cost outcomes over 4 months at a cost-effectiveness threshold of £20,000. The expected incremental net benefit (INB) of peer support (expressed in QALY units) is positive (0.11 QALYs, 95% CI −0.08 QALYs to 1.29 QALYs). Given sampling error, there is a 75% chance that this conclusion is correct (or conversely a 25% probability that usual care was the more cost-effective alternative).

The expected incremental net benefit of peer support remains positive over all values of the cost-effectiveness threshold (see Appendix 2, Table 17) with little variability in estimated uncertainty.

Appendix 2, Table 18 contains descriptive statistics for the costs of wider service contacts. Appendix 2, Table 19 presents estimated differences in the costs of wider service contacts between intervention and controls, adjusting for baseline covariates. The mean cost of self-reported community-based service and police-related contacts was estimated to be £51 (95% CI −£164 to £265) and £53 (−£43 to £149) higher for intervention participants respectively.

Participants

All trial participants who had been randomised to the PW group and who had a known allocated PW were included in this analysis.

Data

The dependent variable, ‘engaged with PW’, was defined in the CACE analysis as participants who had at least two PW meetings, at least one of which was in the community post-discharge.

Pre-randomisation and pre-discharge predictor variables were obtained from a number of sources: baseline research interview which was conducted prior to randomisation, EPR extracted by site information management personnel for the period 12 months pre-index admission and a PW contact log completed using an online survey after each attempted contact.

Pre-randomisation predictor variables were: index admission type (compulsory/voluntary), sexual orientation (heterosexual/not heterosexual), gender, diagnosis (using stratification variable), ethnicity (black/other), age, social inclusion (SIX), severity of symptoms (BPRS), hope, number of admissions in year prior to index admissions, treatment stigma barriers to accessing care evaluation, severity of anticipated stigma and count of life areas of stigma questionnaire on anticipated discrimination (QUAD).

Pre-discharge predictor variables were: matching characteristics between participant and PW, relationship-building in first contact, length of first contact with PW (minutes), days between recruitment and discharge and whether the participant had two or more PW contacts pre discharge.

To calculate the ‘matching characteristics’, variable intermediate variables were calculated to indicate whether allocated PWs matched participants on age (matching within age bands: 18–35, 36+), gender, ethnicity (matched within ethnicity groups; Asian, black, mixed, other and white) and diagnosis (matched with diagnosis categories: psychotic disorder, personality disorder, other diagnosis). The variable ‘matching characteristics’ was calculated as the sum of the four characteristics PWs and participants may have matched on, giving a possible range of 0–4.

For each contact PWs indicated which of the following activities were covered in the contact: using your own lived experience, learning from the lived experience of the person you are supporting, negotiating (or renegotiating) the boundaries in your relationship, accompanying or ‘being alongside’ the person you are supporting, using active listening skills. Using these data, ‘relationship-building’ was calculated as the count of these activities covered in the first contact between PW and participant and had a possible range of 0–5.

Statistical analysis

Logistic regression was used to model the relationship between the two groups of predictor variables and ‘engaged with peer support’ dependent variable. Results are reported with ORs and 95% CIs.

Descriptive data

The sample had a mean age of 39.7 years (SD = 14.0) with a range from 18 to 74 years. There were similar numbers of male (48.3%) and female (51.4%) participants with one identifying as transgender. Just over half of the sample was white, 144/257 (56%) and 43/257 (16.7%) were black. Two hundred and seventeen participants were heterosexual. The largest diagnostic group in this sample was the psychotic disorders group (44.5%). Approximately half of participants (50.8%) had been admitted under a compulsory admission. In the year prior to the index admission, participants had a median of 1 psychiatric admission.

Engagement with peer support

Sixty-two per cent (163/265) of participants received peer support as defined as having had at least two PW meetings, at least one of which was in the community following discharge. For the 163 who engaged with peer support their mean number of face-to-face contacts was 9.1 (median = 9; SD = 4.7) ranging from 2 to 22. Post discharge they had a mean number of face-to-face contacts equal to 7.1 (median = 7; SD = 4.0) ranging from 1 to 15.

Pre-randomisation predictors of engagement

Descriptive statistics of the pre-randomisation variables being tested as predictors are reported in Appendix 3, Table 20, split by engaged with PW group. These are accompanied by the parameter estimates of the logistic regression model. It can be seen that the only demographic characteristic predictive of engaging with peer working is sexual orientation, heterosexual participants being 70% less likely to engage with PW than non-heterosexual participants, OR 0.3 (95% CI 0.08 to 0.87; p = 0.029).

This model was repeated twice, (1) removing SIX and (2) removing count of life areas of stigma (QUAD). SIX was removed as this variable had the most missing data, its removal increasing n to 210. The QUAD variable was removed as the two versions of the scale, mean and count of life areas, were highly correlated, r = 0.73. These two models produced results consistent with the reported model in Appendix 3, Table 20. Little’s test was non-significant, indicating there was no evidence that data were not missing completely at random indicating that a complete case analysis is appropriate.

Pre-discharge predictors of engagement

Looking at the individual characteristics that may match or not between the participant and their PW it can be seen in Appendix 2, Table 19 that there is no evidence that matching on age, gender, ethnicity or diagnosis impacts on whether the participant engages with the PW. These are incidental matches – no attempt was made to match on characteristics apart from gender if a participant expressed a preference or if this was indicated as necessary by a member of the participant’s clinical team. The individual matching variables were not analysed further.

Appendix 3, Table 22 reports the results of the logistic regression model exploring the association between predictors relevant to the instigation of the peer support relationship. It can be seen that a shorter time between recruitment and discharge and a longer duration of first contact with PW is associated with being more likely to ‘engage with peer support’. Relationship-building is nearly significant at the 5% level with those who engaged with peer support having on average 1 more relationship-building activity than those who did not engage with peer support.

Measures of well-being, job satisfaction, multidisciplinary team working and burn out

Data were available for 20 PWs at T2 and 21 at T3. Descriptive statistics are reported in Appendix 4, Table 25. The well-being of the PWs remained fairly constant over the year, marginally lower than the general population norm. The job satisfaction subscales and overall score mean values were higher than norm values apart from satisfaction with pay and prospects subscales. Interdisciplinary Team Scale subscales were higher than norm values across the three timepoints apart from team effectiveness and workplace resources. Mean scores of Maslach Burnout Inventory subscales indicate lower levels of burn out and depersonalisation, and slightly higher levels of personal effectiveness than norms.

Change in service use

Appendix 4, Table 24 indicates very little change in use of services for both mental and physical health from T1 to T3, with the exception of an increase in use of counselling and psychotherapy, and new use of a range of community-based well-being services, including attending community and faith-based groups.

Change in outcomes

Change in outcomes from T1 to T2, and T1 to T3 is reported in Appendix 4, Table 26. Between baseline and 4 months follow-up there was a statistically significant decrease of nearly 4 points in well-being, a medium effect size, 0.56. However, over the course of the year there was no significant change in well-being. There was a statistically significant decrease in some job satisfaction subscales at the 4-month follow-up; personal satisfaction (medium effect size, 0.50), satisfaction with workload (small effect size, 0.31) and satisfaction with prospects (small effect size, 0.33). This decrease in satisfaction with prospects continued over the year with T1 to T3 scores reducing by 0.45 points, a statistically significant change (small – medium effect size, 0.43). Over the course of the year there was also a statistically significant decrease in satisfaction with training (small – medium effect size, 0.46). There was no statistically significant change in any of the Multidisciplinary Team subscales at T2 or T3. Of note however is the relatively large effect size for a decrease in the Communication subscale of 0.86 to T3. Examining the Maslach Burnout Inventory, there was a significant increase in depersonalisation at both T2 and T3, the increase occurring in the first 4 months and seemingly maintained to T3 (small-medium effect size to T2, −0.48).

Overall levels of well-being and employment outcomes

Participants articulated a positive sense of well-being, attributing both feeling and functioning well aspects of well-being to the PW role:

For me it was about having the opportunity to be able to help others and to be able to realise that I was a human being who was valued, and it was an empowering experience to be able to now be in a position to contribute and have a meaningful fulfilling role in life again by supporting others.

(BMS-PW01-T1)

In general, satisfaction with the PW role was reported as high, participants were pleased that the role was, in most cases, properly salaried and they were, at least initially, optimistic about future job prospects:

This has been my first paid job in probably eight years I think … It’s been the most fantastic experience, I’ve absolutely loved it.

(SWL-PW05-T3)

… maybe gain some more experience, maybe go on to a more senior role or managerial role. I’d like to do some training development for my own personal growth. And maybe go on to peer training.

(CNW-PW04-T1)

A strong and supportive sense of team within the peer support team was evident in a number of interviews:

Supervision has been a really nice way to end the week because we all come together and we discuss ideas and issues and explore ways to make things better to improve our practice. I do enjoy working with my other peer workers as well, they’re really nice. We get on really well and I think having the people that you surround yourself with at work and a good work environment is really essential to your mental health.

(CNW-PW02-T2)

However, a sense of disconnect with the work of clinical teams might have contributed to the slightly lower team effectiveness and resources scores:

… it’s been a bit blurred about what our role is as peer support workers [when] there hasn’t been a care co-ordinator or other support workers involved with a particular peer that you are working with. I think sometimes you are a little bit forced into helping with things like accommodation … and that’s not your role.

(CNW-PW01-T2)

PWs, by and large, did not report feeling burnt out in their work, and also experienced a high degree of personal effectiveness in the role:

… so previously when I was at [name of service] they were very long days and it’s a crisis service so it’s extremely busy and you just don’t stop all day so I was tired. But this role, because it’s completely different, it’s good for me, it’s pushing me and I think it’s having a better impact on my well-being.

(BDF-PW03-T1)

Low levels of de-personalisation were indicated by enthusiasm about interpersonal relations and working alongside others, especially where the work was emotionally demanding:

… as soon as I walked on the ward … I was like, ‘I want to be back here on the ward with all these people’ … every time I come into work, the more I’m around people … it’s just like I’m on a complete high … It’s just so positive.

(ELN-PW01-T1)

Change in outcomes