Structured lifestyle education to support weight loss for people with schizophrenia, schizoaffective disorder and first episode psychosis: the STEPWISE RCT

Rationale

Pragmatic interventions that offer long-term weight control or reduction in people with schizophrenia or first episode psychosis are needed. The STEPWISE trial aimed to evaluate if a structured lifestyle education programme, delivered in a community mental health setting, supported weight loss at 1 year in adults with schizophrenia, schizoaffective disorder or first episode psychosis.

Objectives

• To develop a group-based, structured self-management lifestyle education programme for people with schizophrenia, schizoaffective disorder and first episode psychosis.

• To conduct a multicentre RCT to investigate if the intervention leads to a clinically important difference in weight change, as well as physical activity and diet, compared with usual care.

• To conduct a mixed-methods process evaluation to explore intervention delivery, participant and facilitator experiences and explain discrepancies between expected and observed outcomes.

• To conduct an economic evaluation of the intervention.

• To assess the fidelity of intervention delivery when undertaken at 10 different sites.

Recruitment and reminders for attendance

Recruitment and retention to the first two cohorts were difficult, and novel recruitment strategies, such as the offer of transport, and text and telephone-call reminders, were needed to increase attendance. This led to improved participation in the third and fourth iterations.

Logistical considerations

Participant feedback showed that they valued the intervention and discussion mostly focused on practical ways of facilitating attendance and active engagement. Organisational issues were seen as key factors (e.g. arranging taxis helped participants to arrive in a timely manner, while reducing the anxiety of using public transport). The local mental health venue was described as a convenient and familiar place to reach. The participants reported that the number of sessions was reasonable.

Many people with schizophrenia have altered sleep patterns, which make early-morning appointments difficult, and so the time of the sessions (12.30 for a 13.00 start) was considered to be ideal. During the first cohort, participants struggled to attend on time; this was partly solved by providing taxis, but we also introduced a pre-session healthy lunch. This meant that the participants did not disrupt sessions if they arrived late, and also provided a practical demonstration of how to eat healthily on a limited budget.

Many people with schizophrenia may experience cognitive deficits and limited ability to concentrate over time. To overcome this, we incorporated sufficient and flexible breaks into the pilot intervention.

Sessions

Participants enjoyed the small-group setting, which enabled them to assimilate information more efficiently, share experiences with fellow participants and reinforce existing messages. Participants and facilitators found the resources, such as flipcharts, laminates and booklets, valuable and engaging. Participants commented on the benefit of using the same facilitators throughout the intervention. Contrary to our expectations, participants wished to attend alone and thought that they would not benefit from bringing an accompanying person to the sessions. Participants felt that undertaking the intervention was something they could accomplish on their own and that it would be easier to share information with fellow participants without the presence of people who did not have mental illness.

The use of supporting tools (e.g. samples of low-calorie drinks and snacks, kitchen scales, cookery books and pedometers) reinforced the messages provided to participants about the benefits of participation, improved internal motivation and supported engagement and attendance.

Trial design

We undertook a multicentre, two-arm, parallel-group RCT of the STEPWISE structured lifestyle education programme compared with usual care plus written lifestyle advice in 10 NHS Mental Health Trusts in England. Participants were individually randomised to the STEPWISE programme or usual care. This report is concordant with the Consolidated Standards of Reporting Trials (CONSORT) statement (2010). 31

Important changes to methods after trial commencement

Trial recruitment commenced on 12 March 2015 (first patient, first visit) and, following this, in response to early observations and feedback from the intervention development study, a number of changes were made to the protocol. The protocol was published32 and the current version is available via the National Institute for Health Research (NIHR) Journals Library website. 33

Between the initial Research Ethics Committee approval and study commencement, there were two substantial amendments. These revised the sample size, clarified the eligibility criteria, described the control arm and updated the screening and consent process to ensure that recruitment could take place closer to scheduled intervention sessions.

In March 2015 (substantial amendment 3, protocol version 4.0), following feedback from the intervention development study, the option for participants in the intervention arm to bring along a friend, relative or carer to the intervention sessions was removed. This amendment also added details about referring any concerns or participant risk to the clinical care team. Amendments were made before delivery of the first foundation session (23 April 2015).

In June 2015 (substantial amendment 4, protocol version 5.0), the data collection windows for the Operational Criteria Checklist for Psychotic Illness and Affective Illness (OPCRIT+) and the outcome assessments were increased, to provide more opportunity for sites to follow up participants, allowing for missed appointments.

A further amendment in November 2015 (substantial amendment 5, protocol version 6.0) clarified the use of ‘Community Mental Health Teams’ in the eligibility criteria. This allowed potential participants within a variety of services, including those stepping down from inpatient to community services, to participate, provided that they could fully implement the learning from the intervention. This amendment also allowed 1 additional week to obtain the fasting blood sample at the 12-month follow-up.

In February 2016 (substantial amendment 6, protocol version 8.0), changes were made to the protocol to allow over-recruitment beyond the initial recruitment target, in order to allow centres to recruit in waves and run intervention groups with sufficient participants.

Participant selection and eligibility

The trial was co-ordinated from the Clinical Trials Research Unit (CTRU) in the Sheffield School of Health and Related Research (ScHARR). Delegated study staff located at participating centres identified and gained consent from potential participants. Eight of the 10 centres [and their principal investigators (PIs)] were involved at the proposal development stage, with two further centres identified through expressions of interest via the NIHR Clinical Research Network.

The trial was promoted within clinical teams and in community areas in which mental health services are delivered. Clinicians used a standardised script to ensure that potential participants received consistent information about the trial. The research team at participating sites worked with clinical teams to identify potentially eligible patients from clinic lists and other caseloads. Self-referrals were also accepted, following the use of information displayed on posters and leaflets.

The study took place within CMHTs, including early intervention services, in 10 mental health NHS trusts in a range of locations, urban and rural, in Sheffield, Leeds, York, Bradford, Greater Manchester, South London, Sussex, Hampshire, Devon, Somerset and Cornwall.

Adults were eligible for inclusion in the study if they:

• were aged ≥ 18 years

• had a diagnosis of schizophrenia or schizoaffective disorder [defined by the International Classification of Diseases, Tenth Revision (ICD-10) codes F20 and F25] or first episode psychosis (defined as < 3 years since presentation to the mental health team) using case note review

• were being treated with an antipsychotic

• were able to give written informed consent

• were able and willing to attend and participate in a group education programme

• were able to speak and read English

• had a body mass index (BMI) of ≥ 25 kg/m² or were concerned about their weight (in the case of participants from South Asian and Chinese backgrounds, the BMI threshold was reduced to ≥ 23 kg/m²).

People were excluded from the study if they:

• had a physical illness that could seriously reduce their life expectancy or ability to participate in the trial

• had a coexisting physical health problem that would, in the opinion of the PI, independently affect metabolic measures

• had a mental illness that could seriously reduce their ability to participate in the trial

• had a current pregnancy or were < 6 months post partum

• had a condition associated with significant weight gain (e.g. Cushing syndrome)

• engaged in significant alcohol or substance misuse which, in the opinion of the PI, would limit the patient’s ability to participate in the trial

• had a diagnosis or a tentative diagnosis of psychotic depression or mania

• had a primary diagnosis of a learning disability

• were currently (or within the past 3 months) engaged in a systematic weight management programme, to ensure that other programmes did not have an impact on baseline measures.

During the trial, participants were not prevented from joining weight loss or physical activity programmes (intervention arm participants were encouraged to make positive behaviour change); however, any uptake of systematic programmes (other than STEPWISE) was captured at the 3- and 12-month follow-up by asking all participants if they had taken up any other weight management or physical activity programme outside the trial.

All potential participants had a minimum of 24 hours in which to decide whether or not they wished to participate, before attending a consent visit.

The CTRU co-ordinated the follow-up and data collection in collaboration with centres. Participant study data were collected and recorded on study-specific case report forms (CRFs) by site research staff and were entered onto a remote web-based data capture system at each site.

Interventions

All participants received written lifestyle advice on diet, physical activity and alcohol and smoking use before randomisation. Participants were randomised to receive either the STEPWISE education programme or usual care.

Measurement of outcomes

Following consent but before randomisation, research staff completed CRFs for each participant, which covered medical and psychiatric history, demographics and current medication information (Table 2). All assessments and questionnaires were undertaken either at the participant’s home or in the NHS trust. A fasting blood sample was taken, as well as measurements of vital signs and anthropometry. Baseline self-report instruments were completed, with research staff encouraged to read questions aloud to participants, with all available answer options, to ensure understanding. Participants were provided with a wrist-worn accelerometer to wear 24 hours per day for 7 days following the visit.

Additional information, such as medication details and details of hospital admissions, was obtained and/or verified using the patient’s medical notes. At both the 3- and the 12-month follow-up visits, research staff who were masked to treatment allocation confirmed medication information with participants, particularly doses and side effects of antipsychotic medication. They took measurements of blood pressure, weight and waist circumference (and a fasting blood sample at 12 months only) and provided the participant with a wrist accelerometer to wear for the next 7 days. Participants were asked to complete the same questionnaires as at baseline, with the addition of a form to capture the uptake of any weight loss programmes outside the STEPWISE trial. The follow-up windows at 3 and 12 months were defined as minus 2 weeks and plus 4 weeks to allow time for missed and rearranged appointments.

Derivation of outcome measures

The EuroQol-5 Dimensions (EQ-5D) tariff was scored using the EQ-5D-5L for UK population norms;51 no score was calculated if any of the five items were missing. The eight subscales of the SF-36 were calculated as per McHorney et al. ;40 subscores were calculated when at least half of the questions within the domain had been answered. When all eight domains were completed, the aggregate physical and mental component scores were calculated. The adapted B-IPQ was scored by summing the responses to items into a single score41 if at least six of the eight questions were answered. The BPRS was scored by summing up the responses into a single score42 and was calculated if at least 14 of the 18 items had been answered.

Sample size

The sample size calculation was based on data from two sources, both of which assessed behavioural interventions for weight loss in people prescribed antipsychotics for schizophrenia. First, a systematic review undertaken by Das et al. 52 examined both randomised and non-randomised controlled trials, and reported between-group differences of 1.5–6 kg (SD ≈5 kg). Data on overweight and obese UK patients with severe mental illness from a second study also contributed to the sample size calculation. A total of 51 people with schizophrenia were followed up for at least 1 year, and the weight change reported was 7.7 kg (SD 6.5 kg). 53

The sample size used in STEPWISE aimed to detect a difference of 4.5 kg, which is clinically meaningful (average ≈5% reduction in body weight)54 and also appears to be compatible based on previous work. A conservative estimate of a SD of 10 kg, 95% study power and a two-sided significance level of 5% was assumed, meaning that 130 participants per intervention arm (260 participants in total) were required to detect a minimum clinically important difference of 4.5 kg.

Owing to the group nature of the intervention, it is possible that the outcomes of participants within the same group may be correlated. Therefore, an average group size of seven participants was assumed, with an intraclass correlation of 5% in the intervention arm. For this reason, the sample size was inflated by a design effect of 1.3 in the intervention arm, which gives revised sample sizes of 169 participants in the intervention arm and 130 participants in the control arm (299 in total).

To ensure a 1 : 1 allocation, 158 participants were required per arm to reproduce this power. Assumptions were made for a conservative dropout rate of 20%, which is higher than that observed in similar studies,55 giving a final sample size of 198 participants per trial arm. This equates to 40–50 participants at each centre, with 20–25 of these receiving the intervention in up to four groups per site.

Randomisation

Blinding

All research team members performing outcome assessments were blind to treatment allocation. Blind (or suspected) breaks were recorded. Owing to the nature of the intervention, participants were not blinded.

Ethics aspects

The study received a favourable opinion from the National Research Ethics Committee, Yorkshire & the Humber – South Yorkshire on 4 February 2014 (reference 14/YH/0019).

Patient and public involvement

Angela Etherington (a person with severe mental illness and experience of taking antipsychotic medication) and David Shiers (the carer of a family member with schizophrenia) were involved in the design of the study and the intervention, management meetings, the qualitative research analysis and the drafting of the report. They reviewed, made changes to and approved the final lay summary.

Outcome measures

Analysis of weight change

The primary objective (weight change at 1 year post randomisation) was assessed by fitting a marginal generalised estimating equation (GEE) model using robust standard errors and an exchangeable correlation structure. The difference between intervention and control arms was adjusted for baseline weight, site and years since antipsychotic medication initiation. The intraclass correlation coefficient (i.e. the ‘cluster’ term) was derived from the correlation matrix of the GEE model. The following preplanned sensitivity analyses were undertaken:

• alternative covariates (in which any imbalanced baseline characteristics were added into the GEE model)

• alternative model structure (multilevel model in place of the GEE to estimate the cluster effect)

• alternative assumptions for missing data.

The last of these analyses used approaches proposed by von Hippel,56 Carpenter et al. 57 and White et al. 58 The complete-case analyses were augmented with a reanalysis assuming a missing-at-random (MAR) mechanism, achieved by incorporating all baseline covariates that were associated with the probability of missing weight at 12 months and/or with weight change among those who were followed up; these included baseline demographics, disease characteristics, recruiting site, treatment group, weight, BPRS score, B-IPQ score and physical domains of the SF-36 at baseline and 3 months.

The first model incorporated baseline measures as covariates, following which the treatment effect was re-estimated (model MAR 1). Following this, predictive mean matching multiple imputation via chained estimation incorporated all available baseline and 3-month values to impute missing 12-month weight using the original model covariates (model MAR 2). Thereafter, the sensitivity to missing not at random (MNAR) was assessed by adding a range of fixed quantities (delta) to the MAR prediction; for example, delta = 1 corresponds to the assumption that an individual with no 12-month data has a weight change of 1 kg more than predicted. The values of delta used ranged from –5 to + 5 kg and included different values of delta for the two treatment groups.

The treatment comparison was calculated in the following preplanned subgroups:

• uptake of therapy (non-attender; attended one or two foundation sessions; attended three or four foundation sessions but no booster sessions; attended at least three foundation sessions and one booster session)

• recruiting centre

• clinical diagnosis (first episode vs. other diagnoses)

• time since starting antipsychotic medication (≤ 12 months, > 12 to ≤ 24 months and > 24 months)

• BMI (≤ 25 kg/m², > 25 to ≤ 30 kg/m², > 30 to ≤ 35 kg/m², > 35 to ≤ 40 kg/m² and > 40 kg/m²)

• principal reason for dietary concern, as assessed by the B-IPQ.

Analysis of other outcomes

Other outcomes were analysed using a GEE model, with the covariates being treatment group, site, years since antipsychotic medication initiation and the baseline measurement of the respective outcome.

General considerations

A comprehensive statistical analysis plan was developed before the database freeze and while the statistician was blinded to treatment allocation. Data were reported and presented in accordance with the revised CONSORT statement. 31,59 Analyses were performed on an intention-to-treat basis, unless otherwise stated. Analyses of outcome in relation to protocol compliance were undertaken by looking at the level of course attendance (a subgroup analysis in Analysis of other outcomes) and by complier-average causal effect (CACE), using two-stage least squares regression; the latter defined compliance as attendance of at least one foundation course.

All statistical tests were two-tailed at a 5% significance level and CIs were two-sided, with 95% intervals. No adjustment was made for multiplicity, but the number of outcome measures used necessitates cautious interpretation of statistical significance. All analyses were performed in the Stata^® version 14.2 (StataCorp LP, College Station, TX, USA) statistical software, using the user-written Stata package rctmiss for the first MNAR model. 60

Overview

The process evaluation was undertaken ‘to explain discrepancies between expected and observed outcomes, to understand how context influences outcomes, and to provide insights to aid implementation’. 30 Specifically, we investigated whether or not (1) treatment is consistent with the underpinning behaviour change theories (treatment theory or theory of change) and (2) contextual factors affected implementation. The process evaluation used a pipeline logic model, showing causal links between resources, activities and outcomes, integrating the National Institutes for Health Behaviour Change Consortium (NIHBCC)’s approach to treatment fidelity24 and a modified version of Linnan and Steckler’s framework for process evaluation. 61 We described context qualitatively and took a mixed-methods approach to characterising recruitment, reach, dose delivered/received and fidelity both qualitatively and quantitatively, with triangulation between data sources. 62 Interviews were held with intervention designers, health professionals and RCT participants, and the analyses were combined with RCT data and quantitative fidelity data.

Researchers

The qualitative researchers, Rebecca Gossage-Worrall, a female graduate sociologist [Master of Arts (MA); Research Associate], and Daniel Hind, a male graduate anthropologist [Doctor of Philosophy (PhD), Reader], had 8 and 10 years’ experience of interviewing, respectively. No relationship was established with any participant outside or before the interview. The interview purpose was explained to participants twice, at consent to trial and at interview. Two facilitators had a prior relationship with Rebecca Gossage-Worrall (through their study research role) before participating in the interviews. Daniel Hind knew two intervention developers prior to interview via project meetings.

Theoretical and thematic framework

Theory

We used the International Classification of Functioning (ICF) as a conceptual framework for describing the schizophrenia-specific context for implementation. 75,76 We used an a priori framework, based on similar studies,77,78 to inform the topic guide for interviews with participants (Table 3). Topic guides for interviews with health professionals were designed around the normalisation process theory (NPT). 80–83 We used the theoretical domains framework (TDF)84 to characterise stakeholder understandings of the intervention. Codes from the TDF were later mapped to constructs from the principal theories underpinning the STEPWISE intervention: Bandura’s self-efficacy theory,85 the self-regulation theory of Leventhal et al. 86 and Marlatt and George’s relapse prevention model. 87 We used the NIHBCC’s framework for understanding intervention fidelity79 and the framework of Sekhon et al. for understanding the acceptability of health-care interventions. 88

TABLE 3 - Example questions derived from the a priori framework

A priori theme	Example interview question
Acceptability77,78	How did you find the STEPWISE programme? Was there anything in particular that you liked about it? Was there anything in particular that you did not like about it? How easy or difficult would it be to pay for the travel to the sessions? How long were the sessions? Was that OK? How many sessions did you attend? (Probe: if sessions were missed, why?) How do you feel now that it is the end of the weekly course?
Being in a group77,78	How many people were in your group? (Probe: was this number OK?) How easy or difficult did you find talking in the group? (Prompt: talking about weight, experience of attending other weight management programmes and how they differ)
Presentation of content78	Did you feel it got the balance right in terms of talking about both diet and physical activity? What did you think of the people who led your group?
Changes in behaviour77,78	Did you receive anything to help you to put into practice what you learned? (Prompts: water bottle; scales; printed materials) How useful or not were these items?
Processes of change78	Did the programme meet your needs?
Types of interventions78	Did you feel it got the balance right in terms of talking about both diet and physical activity?
Fidelity79	Did you have any of the one-to-one contact with your facilitator?

We developed a programme theory to identify essential elements for the successful replication and causes of failure in the contracts, actions, interactions and emergent relationships between people and organisations that surround the STEPWISE intervention. 89–91 The programme theory development was deductive, through literature review and by articulating mental models in discussion with the LDC team, and inductive, through interviews with participants and professionals. 92 To illustrate how sequences of events were to bring about desirable outcomes, as per the programme theory, we developed a logic model (Figure 5). 93,94

FIGURE 5.

Logic model for the implementation of the STEPWISE intervention.

Participant selection and setting

Consent for participant interviews was requested, face to face, at the time of consent to the RCT (when participants specified a preferred method of approach) and reconsented immediately before interview. The majority of participants were approached by telephone. Professionals were approached directly by telephone, e-mailed the information sheet and consented by telephone; intervention designers were approached during project meetings and by e-mail. The RCT intervention arm participants were purposively sampled (n = 24) from those consenting and allocated to the intervention (n = 188) to reflect different study centre, gender and age (Table 4). A total of 63 participants (34%) were sampled, of whom 22 (35%) were non-responsive, despite a minimum of three attempts to contact. Ten participants declined when invited to participate, one declined at (re-)consent and six consented but did not attend the appointment. Forty professionals were purposively sampled to reflect differences in site, occupation, sex and prior group facilitation characteristics (derived from a short survey after completing facilitator training) for invitation to study, of whom 20 were interviewed (Table 5); none formally declined, but many left their employment or were non-responsive, or it became unnecessary to interview them because of accrual of the target sample. All those involved with the intervention design were interviewed. All interviews were conducted by telephone.

Data collection

Semistructured interview guides for participants (see Appendix 1) and professionals (see Appendix 2) were pilot-tested with the relevant project team members. No topic guide was used for the unstructured interviews with intervention designers, although sections from the behaviour change wheel,95 the logic model and the NIHBCC framework24 were used as prompts. Interviews were recorded on an encrypted digital recorder and field notes were made during and after the interview. The median (range) length of the interviews was 18:57 minutes (13:06–30:33 minutes) for participant interviews, 46:13 minutes (29:29–76:32 minutes) for facilitator interviews and 39:20 minutes (43:39–64:00 minutes) for intervention designer interviews. Data saturation was achieved in the participant, professional and intervention designer data sets. No repeat interviews were carried out and transcripts were not returned to participants for comment/correction.

Data analysis

Two coders (RG-W and DH) coded participant interviews systematically in NVivo version 11 (QSR International, Warrington, UK) to behaviour change theory, intervention functions, theoretical domains95 and dimensions of acceptability,88 as well as opportunistically to the NIHBCC framework79 (see Table 6) and logic model constructs (see Figure 5). Staff interviews were coded systematically to NPT constructs and opportunistically to the NIHBCC framework and logic model constructs. Developer interviews were coded systematically to intervention functions, logic model and NIHBCC framework constructs (see Table 6).

Fidelity assessment

Leicester Diabetes Centre staff monitored the fidelity of intervention delivery through direct observation of sessions, using two instruments. First, the STEPWISE Core Facilitator Behavioural Observation Sheet assesses the relative presence or absence of 35 behaviours in six domains: non-judgemental engagement of participants (five items); eliciting and responding to emotions/feelings (two items); facilitating reflective learning (eight items); behavioural change, planning and goal-setting (nine items); overall group management (nine items); and other behaviours (two items). Second, LDC staff objectively assessed participant–educator interaction during observation visits by means of the DOT (DESMOND Observation Tool). The coder sat at the back of the room, with a compact disc playing in a headphone, from which a beep sounded every 10 seconds, whereupon the coder recorded whether an educator or a participant was currently talking at that point. Silence, laughter or multiple conversations were classed as ‘miscellaneous’. In self-management programme research, a link has been proposed between less facilitator talk and a more effective participant receipt of the education process, defined as a less didactic/more facilitative approach. 96 The national and local infrastructure available for quality control and mentoring in education programmes such as DESMOND was not available for STEPWISE, so feedback of observations in pursuit of accreditation was conducted. Other steps to ensure intervention fidelity are described in Table 6.

Triangulation protocol

Different methods and informants were used and a formal framework was employed for comparison of the findings to comprehensively address different questions, increase confidence in findings and provide a basis for feedback from participants and professionals on the project team. 97 No priority was granted to either quantitative or qualitative methods, which were used concurrently to assess the intervention. A modified triangulation protocol62 was employed for the methodological triangulation of data sets, in five stages:

1. Data sets were reviewed to compare for presence and examples (‘sorting’) of logic model constructs.

2. The level of convergence between data types was coded for each of the 22 logic model components as ‘agreement’ (full interpretive agreement between data sets); partial agreement (some disagreement within/between data sets); silence (a logic mode component covered by only one data set); and dissonance [disagreement between data sets (‘convergence coding’)].

3. The global level of convergence was characterised (‘convergence assessment’).

4. Differences in the data set contribution to the case study were summarised (‘completeness comparison’).

5. The triangulated results were shared and points of disagreement were discussed with stakeholders at a face-to-face meeting on 14 July 2017 (‘feedback’), with changes in interpretation being incorporated when supported by the data. A formal comparison of researcher coding was not conducted, owing to time constraints and a rapidly evolving analysis strategy.

Introduction

The economic evaluation of the STEPWISE trial was conducted using both health and social care and societal perspectives. Individual participant data were used in the analysis to generate results for the cost-effectiveness of the intervention regarding its impact on cost-relative quality-adjusted life-years (QALYs), derived from the EQ-5D-5L and the SF-36 questionnaires, and weight change. The time horizon of the trial (3- and 12-month follow-up) was the same in the economic evaluation. The economic analyses included all trial participants. The comparison was between the group receiving the STEPWISE intervention and the group receiving a leaflet with advice. This analysis determined cost-effectiveness using the £20,000-per-QALY-gained threshold set by NICE. A sensitivity analysis was used to establish the effect of changes in costs and QoL measurement (comparing results using utility values from the EQ-5D and the SF-36)51,98 on the results. Value sets for health-related QoL measures related to value sets containing preferences from a population in England for the EQ-5D-5L51 and in the UK for the SF-36. 98

Health and social care costs

Health and social care resources were calculated using the unit costs from the Personal Social Services Research Unit (PSSRU)99 and NHS reference costs,100 with service use data. Responses to the Client Service Receipt Inventory (CSRI)101 questionnaire, which captured health and social care as well as other resource use, were collected for the 3 months before the baseline assessment and the 3-month follow-up and the 9 months before the 12-month follow-up. The CSRI includes information on hospital care (inpatient and outpatient), social care, primary care and criminal justice service contacts. Less frequently used health-care professional contacts were reported in the ‘other health-care professional’ category, including drug advisors, outreach workers, early intervention services, occupational therapists, drug service facilities and communities, using the CSRI. Medication prices from the British National Formulary102 in March 2017 were applied to the data on medication use by the participants in the study. All drugs for physical and mental health conditions, including those that could be bought over the counter, were included. Ongoing medications recorded on the CRF were used as a source of medication data.

Intervention costs

The intervention comprised four foundation sessions (2.5 hours each), three booster sessions (at 4, 7 and 10 months, of 2.5 hours each) and support contacts (telephone calls, texts and postcards) every 2 weeks. The base case was conducted to represent the scenario in which only four patients attended each session. This was to reflect lower attendance than the seven full sessions for the intervention. The cost of the accelerometer was omitted.

The foundation and booster sessions of the intervention were costed as group therapy sessions, using the mental health nurse wage level and related oncosts, combined with capital costs and other costs of the service at 2015–16 prices. The intervention unit costs, using the wages of a mental health nurse and a dietitian, were costed for a 2.5-hour session of the intervention. For most analyses, the results are reported for two mental health nurses facilitating a group session with six participants. The unit costs per person, calculated using the costs of a mental health nurse, were £88 for a group of four, £59 for a group of six and £44 for a group of eight (the costs for groups of six and eight were used in the sensitivity analyses). Support contacts by telephone were costed at the price of one nurse or dietitian per minute, multiplied by the length of the call. Text messages and postcards were costed at a nominal value of £5. The costs of the intervention were split between time periods, as the intervention booster sessions and support contacts continued after the 3-month period.

Societal costs

Average wage rates were applied to productivity losses and informal care time, which were obtained using data from the Office for National Statistics. 103 In the sensitivity analyses, we used minimum wage rates,104 which were also applied to productivity losses and informal care, and the unit costs for a home-care worker for informal care using the figure from the PSSRU. 99

Police officer contact time was calculated from police salary scales105 and workforce statistics. 106 A cost of police contacts was included, but no other criminal justice system costs were included. The cost of a lost education day was calculated as the cost of a working day; as it was assumed that as the participants were all adults, this was more appropriate. The CSRI also recorded time lost from work, education and informal care from friends and family.

No discount rate was applied to the costs and QoL measures, as the follow-up period for participants was 1 year.

Health-related quality of life

Health-related QoL was measured using the EQ-5D-5L107 and the SF-36/Short Form questionnaire-6 Dimensions (SF-6D). 98 The EQ-5D is the measure more commonly used in health technology assessments in England and is recommended by NICE. However, the EQ-5D is known to have a ceiling effect in schizophrenia, which is why the SF-6D was also used, as this measure is more likely to be normally distributed. The SF-36 questionnaire responses were converted to the SF-6D QoL measure for the analysis, using domain weightings from Brazier and Roberts. 108 QALYs were calculated from the tariff scores using the area under the curve method for both the EQ-5D-5L and the SF-6D QoL measures. It was assumed that there was a linear change between any two time points.

Statistical analyses

The net benefit approach was used in a cost–utility analysis of costs and both types of QALY data, and in a cost-effectiveness analysis with health/social care and societal costs. The net benefit is derived from the multiplication of a threshold value for a unit of outcome minus the cost.

Service use and cost differences between the two groups were described. The analysis of costs and outcomes was carried out in four combinations: (1) EQ-5D-5L QALY and health and social care costs; (2) EQ-5D-5L QALY and societal costs; (3) SF-6D QALY and health and social care costs; and (4) SF-6D QALY and societal costs.

Incremental cost-effectiveness ratios (ICERs) were computed by dividing incremental costs by incremental outcomes. Cost-effectiveness planes were constructed by obtaining, through bootstrapped regression models, 1000 pairs of cost and outcome differences and showing these on scatterplots. The proportion of replications in each quadrant could then be obtained.

Cost-effectiveness acceptability curves

Cost-effectiveness acceptability curves (CEACs) show the probability that one group is more cost-effective than the other for a chosen value of willingness to pay per unit of outcome. The STEPWISE economic evaluation uses the net benefit approach109 with the 12-month follow-up results for these CEACs. The willingness-to-pay value that is used is the £20,000 NICE threshold for total health and social care costs. CEACs are also produced for societal costs at the 12-month follow-up; however, the relevant analysis for NICE methodology is from the health and social care perspective. The value of the willingness-to-pay threshold was also varied in values of £10,000 between £0 and £200,000, to show how the probability of cost-effectiveness changes as the threshold rises.

Missing data

The multiple imputation (chained equations) method110 was applied to impute missing values for cost and QoL data, assuming that data were MAR before bootstrapping to carry out the cost-effectiveness analysis.

Sensitivity analysis

Three types of sensitivity analysis will be reported in the final analysis, with varying costs. First, costs of the intervention were used [mental health nurse (£1.18 per minute, £59 per session – two facilitators with a group of four patients)]. Second, informal care costs were calculated using two alternative unit costs: home care worker (£20 per weekday hour of a home care worker based on independent and social services home care) and minimum wage rate (£7.20 per hour in 2016). Third, productivity losses (lost employment time), valued using the average wage rate in the base case, are also calculated using the minimum wage rate104 (£7.20 per hour in 2016). Each of these sensitivity analyses were carried out using values from both the EQ-5D and the SF-6D. No adjustments were made to take into account deaths during the study, as the numbers were so low.

Weight change (primary end point)

Among participants completing follow-up, the weight change was almost identical for the intervention and control arms. At 3 months, there was a reduction of 0.2 kg in the intervention arm and an increase of 0.4 kg in the control arm (difference = –0.55 kg, 95% CI –1.44 to 0.35 kg; p = 0.230). At 12 months (the timing of the primary comparison), the mean reduction in weight was 0.47 kg in the intervention group and 0.51 kg in the control group (difference = 0.04 kg, 95% CI –1.59 to 1.67 kg; p = 0.964). The distributions are shown in Figure 8 and further summaries are presented in Table 13. There was considerable variation in the weight change in both groups, ranging from –24 kg to +30 kg in the intervention arm and from –25 kg to +30 kg in the control arm.

FIGURE 8.

Weight change. (a) 3 months and (b) 12 months. Reproduced from Holt et al. 23 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original.

Reproduced from Holt et al. 23 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original.

Figure 9 shows the mean and 95% CI for the adjusted mean weight change by time point and group. The means are adjusted for the primary covariates (recruiting site, years since starting antipsychotics, baseline weight and course identifier). The mean weight loss was similar between arms and was some distance from the 4.5 kg defined as the minimally clinically important difference in the sample size calculation.

FIGURE 9.

Mean and 95% CI for weight change.

The weight change by recruiting centre is displayed in Figure 10. The difference between the intervention and control arms ranged from –2.6 kg to +7.0 kg. Adding an interaction term between site and treatment to the primary model gave a test statistic of χ²₍₉₎ = 14.9 (p = 0.094), indicating some evidence of a variation in the size of intervention effect between sites.

FIGURE 10.

Mean weight change by recruiting site.

Other weight-related end points

The weight, BMI and waist circumference outcomes are also presented in Table 13. Although the proportion of participants reporting any weight reduction was slightly higher in the intervention arm, the magnitude of the difference was almost identical between the two trial arms. None of the comparisons reached statistical or clinical significance.

Sensitivity analyses of weight change

A number of sensitivity analyses were undertaken for weight change with regard to alternative model types, covariates and missing data assumptions (Figure 11). The weight change at 12 months was refitted with sex included as a covariate, as the groups were imbalanced in this; doing so made little difference, as sex was not associated with weight loss. The model assumptions were also reassessed using a mixed-effects model, by CACE (using the same covariates) to assess the impact of uptake, and with missing data assumed to be MAR using predictive mean matching imputation. In all cases, the difference between groups remained small.

FIGURE 11.

Sensitivity analyses for weight change. PMM, predictive mean matching.

It is possible that some participants did not remain in the study because of their weight (i.e. informative dropout or MNAR), and the above models do not account for this. As a final sensitivity analysis, missing weight was imputed in which those who withdrew were assigned weights up to 5 kg more than their predicted value; this was done in the intervention arm only, the control arm only and both. Although this affected the estimated treatment differences, the differences remained small and were not statistically significant.

The impact of the intervention by subgroup was estimated in four subgroups, as described in Chapter 3, Analysis of other outcomes. In all cases, the difference between groups was small and not of clinical or statistical significance (Figure 12).

FIGURE 12.

Weight change by subgroup.

Predictors of weight change

A series of linear regressions were undertaken to assess whether or not any participant characteristics were associated with weight change. The univariate analyses are presented in Table 14. Few of the characteristics appeared to systematically explain weight gain or loss. Weight loss was modestly associated with age, with weight reduction increasing by 0.8 kg per 10 additional years (95% CI 0.0 to 1.5 kg; p = 0.042), participants with a diagnosis of schizoaffective disorder had a greater mean weight loss (2.7 kg) than those with first episode psychosis (0.3 kg) or schizophrenia (0.01 kg gain; global test, p = 0.075). Weight loss was greater among female participants and was, on average, greater among those with higher baseline weight and longer duration on antipsychotics, but none of these findings was statistically significant. Additional analyses based on non-linear associations, using fractional polynomial regression and multivariable regression models, did not reveal any important predictors of weight change. There was no association between total (group, telephone and face-to-face) contact time and weight change (see Figure 13).

Participants who were prescribed lithium had a statistically significantly greater weight reduction than those who were not (mean difference = 4.4 kg, 95% CI 0.4 to 8.5 kg; p = 0.033), although this accounted for only 16 participants completing the trial. Among participants taking antipsychotics, weight loss was greatest among participants on paliperidone (n = 12, average weight loss 3.4 kg) and lowest among those on aripiprazole (n = 63, 1.2-kg gain) and other antipsychotics (n = 20, 1.6 kg gain). Participants who changed medication within 3 months of randomisation had a higher average weight gain than those who did not (antipsychotic change: n = 29, mean difference = 3.4 kg, 95% CI 0.3 to 6.5 kg; p = 0.031; antidepressant change: n = 18, mean difference = 6.1 kg, 95% CI 2.3 to 9.9 kg; p = 0.002), but changes over the full 12 months were not associated with weight change.

Finally, there was no association between total contact time (summed across foundation sessions, booster sessions and support contact) and weight loss (Figure 13).

FIGURE 13.

Weight loss at 12 months vs. total contact time (group session and support contact). The line is a locally weighted scatterplot smoother with bandwidth 0.4.

Other vital signs

The vital signs (blood pressure and pulse) at baseline, 3 months and 12 months are presented in Table 15. Systolic (but not diastolic) blood pressure was significantly higher in the intervention group at 3 months (difference = 2.4 mmHg, 95% CI 0.1 to 4.6 mmHg), but otherwise no change was apparent either between groups or across time points.

Laboratory measurements

The laboratory measurements at baseline and 12 months are presented in Table 16. No change was apparent either between groups or across time points.

Accelerometry

Table 17 shows the results of the accelerometry. A total of 352 participants had baseline recordings for the primary accelerometry end point [(ENMO) i.e. average acceleration], of whom 302 had worn the GENEActiv for at least 4 days and therefore had a valid baseline measure. Of these, valid ENMO data were available in 222 participants at 3 months and in 209 participants at 12 months, meaning that the change in accelerometry could be estimated in 54% and 51% of the trial population at 3 and 12 months, respectively.

Overall baseline physical activity was low in both groups. Intervention arm participants were, on average, more active than control arm participants both before randomisation and after randomisation, but, after adjusting for baseline, few of the differences were statistically significant. The exception was MVPA, measured at weekends at the 3-month follow-up, which was modestly greater in the intervention group. The mean length of MVPA in the intervention and control groups was 70 versus 58 minutes per day (difference of 9.2 minutes per day, 95% CI 1.7 to 16.7 minutes per day; p = 0.016) for 5-minute bouts and 11 versus 7 minutes per day for 10-minute bouts (difference of 5.6 minutes per day, 95% CI 2.0 to 9.3 minutes per day; p = 0.003). The corresponding outcome for acceleration was not statistically significant (20 vs. 19 mg, difference = 1.0 mg, 95% CI –0.3 to 2.4 mg; p = 0.140). None of the remaining outcomes was different between the trial arms.

Dietary intake as assessed by the Dietary Instrument for Nutrition Education questionnaire

Figures 14–18 depict food intake as assessed by the DINE questionnaire (blue lines denote mean values). The measures (including the categorisations as low, medium and high) are reported in Table 18. The baseline self-reported diet of the participants in both groups indicated a high consumption of refined sugar from sugary drinks and low fibre. There was some evidence that alcohol intake had reduced in the intervention arm, although the intake was highly skewed and not robust to alternative models, including non-parametric analyses. No other dietary changes were observed as a result of the intervention or trial participation.

FIGURE 14.

Dietary Instrument for Nutrition Education fibre intake. (a) Baseline; (b) 3 months; and (c) 12 months.

FIGURE 15.

Dietary Instrument for Nutrition Education fat intake. (a) Baseline; (b) 3 months; and (c) 12 months.

FIGURE 16.

Dietary Instrument for Nutrition Education unsaturated fat intake. (a) Baseline; (b) 3 months; and (c) 12 months.

FIGURE 17.

Dietary Instrument for Nutrition Education daily sugar intake from drinks (g). (a) Baseline; (b) 3 months; and (c) 12 months.

FIGURE 18.

Dietary Instrument for Nutrition Education weekly alcohol intake (units). (a) Baseline; (b) 3 months; and (c) 12 months.

Patient-reported outcome measures and disease severity

Health utility (EQ-5D) and QoL (SF-36) are reported in Tables 19 and 20. Self-reported QoL post randomisation was generally slightly higher in intervention participants (physical functioning, role limitations), but emotional well-being was slightly higher in the control group participants.

Smoking status

At baseline, approximately half of the participants were current smokers (Table 24). Smoking rates did not change during the trial in either group.

Weight loss programmes

During the trial, 25 participants [(7.4%) eight control, 17 intervention] reported attending one or more weight loss programmes outside the trial (Table 25).

Ten-year cardiovascular and diabetes risk

The Framingham cardiovascular risk scores are presented in Table 26. Twenty-six participants in each trial arm had a pre-existing cardiovascular diagnosis and 84 (40 intervention, 44 control) were aged < 30 years; the Framingham risk score is not defined for these subgroups. The mean ‘desk’ Framingham 10-year risk (which does not incorporate laboratory measures) was 10.9% for the intervention group and 10.6% for the control group at baseline, and remained similar at both 3 months and 12 months. The mean laboratory-based Framingham scores were substantially lower in the control group at baseline (7.7% vs. 8.6%), and both demonstrated minor change (< 1%) at 12 months. Only one participant in the intervention arm developed a cardiac event during the trial; this individual died.

Table 27 shows that 60 participants (35 in the intervention group and 25 in the control) group had diabetes on entry to the trial. Among the remainder, the 10-year risk of developing diabetes was 17.2% in both trial arms at baseline, with over half being in the ‘moderate’ risk category. By 12 months, the risk had decreased by 0.5% in the intervention arm and by 0.4% in the control arm. No participant was recorded as having developed diabetes during the trial, although one participant (intervention arm) was hospitalised twice because of diabetes-related complications.

Change in medications

During the trial, the vast majority of participants remained on antipsychotics (Table 28), although around one-fifth of participants in both groups experienced a change of antipsychotic treatment. Approximately half of participants in both groups were treated with antidepressants. By the end of the trial, 17% of intervention arm participants and 15% of control arm participants had experienced a change of antidepressant medication.

Session feedback

Intervention group participants could provide anonymous feedback following each group session. It was not possible to link feedback to their clinical outcomes, but the feedback could be linked to the study centre results. Seven hundred and eight session feedback forms were returned after the seven group sessions across the 10 centres, although we did not know who responded or on how many occasions. Therefore, the numbers and percentages in the tables below are based on ‘forms returned’ rather than on ‘participants’. Five of the returned forms had no study centre identifiers and were excluded. When two responses were indicated (n = 3), the higher score was taken, and missing responses (n = 10) were replaced with the mean of the participants’ other responses.

Table 29 shows that the vast majority of responses were positive for all six statements, with almost 90% ‘agreeing’ with each statement either ‘strongly’ or scoring ‘2’ on a Likert scale of 5. A total of 87.2% of participants strongly agreed or agreed that the sessions had met their needs.

Participants had the opportunity to add comments to their forms, although only four did so. Three of these comments were positive:

I felt this has been a fantastic group and really looking forward to meeting up in 3 months’ time.

One thing that went well? Talking together about different weight control.

What went well. Everything except the cold.

The only negative comment made was:

Mental health as bad as ever which affects my answer.

Session feedback by study centre

Session feedback scores could range from 6 to 30, with 6 indicating the most positive response on all five items and 30 indicating the most negative response on all items. Table 29 shows the average feedback scores across the 10 centres and Table 30 shows the average scores by centre. Feedback was similar across the centres: in most centres > 50% of forms recorded the most positive score of 6 and in all centres at least 75% of forms recorded scores below 12 (equivalent to scoring 2 for each statement). There were no significant correlations between mean weight change and mean feedback scores for centres, at 3 months or 12 months (Spearman’s rank-order correlation = –0.20; p = 0.476, and Spearman’s rank-order correlation = 0.042; p = 0.454, respectively).

Participant deaths

Four deaths were reported, all in the intervention group. Three deaths occurred during the trial and one occurred several months after the trial ended. Detailed information is given below.

Patient 1 was a 53-year-old white man with a BMI of 34 kg/m² and concurrent diagnoses of asthma and depression, who had been on antipsychotic medication for 32 years. His death occurred 264 days after randomisation, after he had attended four foundation sessions and two booster sessions. The cause of death was a pulmonary embolism, which is likely to have developed secondarily to a ruptured Achilles tendon.

Prior to the study, the participant reported walking a lot, achieving more than 10,000 steps per day, as assessed by pedometer. Objective accelerometry data indicated that he had not changed his physical activity after the intervention. Nevertheless, as a ruptured Achilles tendon could have occurred as a result of following advice from the intervention, the relationship was reported to the ethics committee as possibly intervention related. This death was therefore reported within 15 days to the research ethics committee, which determined that no further action was required.

Patient 2 was a 48-year-old white woman with a BMI of 40 kg/m² and concurrent diagnoses of hypertension and dyslipidaemia. She had been taking antipsychotic medication for 22 years. Her death occurred 260 days after randomisation, after she had attended three foundation sessions. The cause of death was determined by the coroner as 1a) left ventricular hypertrophy, 1b) hypertension and obesity. The death was unrelated to the intervention.

Patient 3 was a 35-year-old black woman with a BMI of 38 kg/m² and a concurrent diagnosis of type 2 diabetes mellitus who had been on antipsychotic medication for 7 years. Her death occurred 57 days after randomisation after four foundation sessions. Her diabetes mellitus had been treated with oral antidiabetes agents, but not sodium–glucose co-transporter-2 inhibitors. She was found unresponsive at her home and the death was determined as resulting from diabetic ketoacidosis leading to cardiac arrest. The death was unrelated to the intervention.

Patient 4 was a 36-year-old Asian woman with a BMI of 36 kg/m² and a concurrent diagnosis of type 2 diabetes mellitus who had been taking antipsychotic medication for 10 years. She died from a myocardial infarction 1 month after she completed the study, while she was detained under section 3 of the Mental Health Act 1983. 111 She had been under the care of the local cardiology team, which advised that she stop taking clozapine, which precipitated the deterioration in her mental health. The death was unrelated to the intervention.

Body functions

Half of the facilitators interviewed commented that, although most participants were receptive to the intervention, an individual’s psychosocial functioning could affect how well they could engage with STEPWISE. Facilitators discussed this most frequently in terms of the transient severity of hallucinatory or delusional symptoms associated with paranoid schizophrenia; some also spoke of comorbid learning disabilities, implying problems with attention, memory, goal-directed/logical thought, insight, executive function or language, which are commonly associated with disorganised-type schizophrenia:

. . . when we’ve gone out together we’ve all been discussing . . . university degrees they’ve got . . . they are really quite intelligent people but they still do have a psychotic illness . . . they could be distracted if there’s something on their mind, or if they’re paranoid at the time . . . that has to be taken into consideration a little bit more when you’re planning.

Facilitator S03/F02

We’ve got quite a small group and one of the persons in the group . . . he has a mild learning difficulty and I’m not sure that he’s quite sort of clear about what, why he’s there particularly sometimes, which is difficult.

Facilitator S06/F05

I think some of them have mild learning disabilities and umm I’m not sure that you know they take in everything we tell them in the session.

Facilitator S09/F01

There could be like mild learning disabilities running alongside their mental illness as well and that, I have definitely noticed that that’s hindered . . . someone’s ability to engage.

Facilitator S09/F03

In addition, developers characterised this group as sometimes less interactive than other populations and occasionally somnolent or impulsive:

. . . there’s been a couple of occasions where people do seem a bit overmedicated . . . Asleep . . . or saying very random things that really have no you know no connection to what is being talked about.

Developer D05

We all said that the medication that we’re on, different people different reactions, but quite a few of them said you know, I’m sluggish from the sedation. And it’s hard to wake you up in the morning.

Participant S08/Q06

. . . so when the sandwiches came out there was definitely some people [sic.] it was literally . . . automatic that reaching for sandwiches and eating and reaching for another sandwich . . . and thinking do you know, ‘have you noticed how many of those sandwiches that you’ve eaten?’.

Developer D03

Activities and participation

Cognitive problems meant that some participants had problems learning and applying knowledge, specifically, acquiring skills, reading and problem-solving. Undertaking tasks could take longer than expected, and that could be difficult:

We did have to sort of adjust to the people that were in the group . . . the ones that were struggling with the reading . . . we would just sort of put more time in with them.

Facilitator S09/F01

Costs notwithstanding, facilitators felt that the use of taxis was essential for attendance; the barriers to using transportation were great for those with severe symptoms or poor cognition, and many lived in areas that were poorly served by public transport. Aside from managing diet and fitness, the principal objective of STEPWISE, a self-care theme that often arose was the importance of adherence to antipsychotic medication in the face of weight gain:

Quite a few service users get put off continuing to take their medication because of the side effects and this gives them a way to control it so it isn’t only kind of benefiting their physical health, but it will benefit their mental health as well if it’s helping them to stay on the medication.

Facilitator S06/F01

In some cases, family or residential care staff were responsible for shopping and food preparation. Meeting strangers and forming and conducting informal social relationships could cause anxiety. Some had never met others with their condition; others talked about the difficulty of discussing antipsychotic-induced weight gain at Weight Watchers owing to the stigma of mental illness:

. . . they are finally coming in to a group of people they have never met. It’s quite anxiety provoking for them, isn’t it?

Facilitator S06/F04

So, the STEPWISE programme was really the first time I had had exposure to other service users.

Participant S09/Q01

I would never have said anything [at Weight Watchers] about my medication sort of being a contributor to me putting on weight because they’d say ‘What is it?’ and . . . I would’ve felt pressure to tell them. And then it was like a ‘oh no I’ve told them I’m mentally ill now’ sort of thing.

Participant S10/Q03

Many STEPWISE attendees were not in remunerative employment and some were living in economically difficult circumstances:

You end up having more of a sedentary lifestyle because you’re told you’re not fit to work.

Participant S08/Q09

I was struck by some of the levels of deprivation . . . there was somebody who was living in a caravan. Their caravan had been washed away in the floods so they actually didn’t have anywhere to live . . . they were there at the group, which was pretty amazing, but if you think about . . . where this fits on their . . . continuum of needs . . . some people are just getting by.

Developer D03

Environmental factors

The attitudes of family members were not always supportive vis-à-vis lifestyle changes, and stigmatising treatment by other members of the public was reported:

They found these sessions were the opportunity for them do it for themselves instead of having someone else there . . . [one of] the participants brought someone with them . . . it was a relative and you could see, there was a lot of . . . judgemental attitude, like, ‘oh, you should be doing this,’ or, ‘you shouldn’t be doing that,’ and . . . ‘why are you doing this?’ and it was a lot of pressure on that participant.

Developer 01

The individual attitudes of facilitators were characterised by developers as being generally sensitive to the needs of an autonomy-enabled approach to behaviour change, with only a small minority being overly didactic in delivery. The following quotation from a facilitator is illustrative of how many facilitators characterised their roles, interactions and relationships with clients:

Going to an acute [psychiatric] ward and you’re saying to the person take these tablets, they’ll turn around and say to you, ‘why should I?’ Whereas, if you’ve got somebody who’s got a general illness they just sit there and put them in their mouth [laughs] . . . you get used to having to discuss things with people and in that respect I suppose it was natural for me to do things that way [i.e. in accordance with the STEPWISE philosophy] . . . you’ve got to be more open-minded, I think, with people with mental health . . . a lot of the people you’re talking to are extremely intelligent people and so therefore you do respect them . . . they’re not an illness, they’re a person [with] need, wants feelings, ways of doing things . . .

Facilitator S03/F02

However, some facilitators thought that health professionals who were meant to be referring patients to STEPWISE (‘gatekeepers’) sometimes erected barriers to patient engagement:

A lot of clinicians have said their clients don’t like group work or they don’t like to leave their home and they’ve put a lot of barriers in the way of people engaging.

Facilitator S06/F01

Interviewees described the availability of, and referral patterns associated with, existing lifestyle interventions as ad hoc, fragmented and locally idiosyncratic, often involving referral to other organisations. This was confirmed by our survey of usual care. 36 Monitoring of weight was not systematic for this population in all trusts, and, although all trusts talked about physical health as a priority, scarcity of resources was a problem, although quality-improvement initiatives were changing this in some cases (see also Sense-making, Operational work and Appraisal work).

How context might affect the success of the intervention

Facilitator views of STEPWISE and Implementation process concentrate on how organisational context affects outcome. To more vividly illustrate and explore potential context effects (see also Table 4), we present six vignettes of interviewees from the intervention arm who experienced clinically important weight gain (Box 1) and weight loss (Box 2). These vignettes suggest that, although weight control is possible in people with first episode psychosis and in people living in the most deprived neighbourhoods (S04/Q02), it is particularly difficult for those with first episode psychosis (S04/Q08, S04/Q12, S07/Q06). Those with clinically important weight loss may have less severe psychiatric symptoms and comorbidities (S04/Q02; BPRS score of 34) than those with clinically important weight gain (S04/Q08; BPRS score of 44). Those who experienced clinically important weight loss generally talk about needing more ongoing support or, especially, closer monitoring of behaviour change enactment and outcomes, as if they are more alert to the risk of relapse.