Identifying effective characteristics of behavioral weight management interventions for people with serious mental illness: A systematic review with a qualitative comparative analysis

Summary People with serious mental illness (SMI) have identified barriers to engaging in behavioral weight management interventions (BWMIs). We assessed whether BWMIs that addressed these barriers were more effective. First, we systematically reviewed qualitative literature and used a thematic analysis to identify the characteristics of BWMIs that promote engagement for adults with SMI. Second, we systematically reviewed randomized controlled trials (RCTs) of BWMIs in adults with SMI. Data on the characteristics that promoted engagement and weight outcomes were extracted. We then used a crisp‐set qualitative comparative analysis (CsQCA) to identify which characteristics were associated with weight loss. For the qualitative review, 20 studies in 515 people with SMI were analyzed and nine characteristics were reported to promote engagement in BWMIs. For the systematic review, 34 RCTs testing 36 interventions in 4305 participants were included. The active interventions resulted in more weight loss (mean = −4.37 to +1 kg at 6 weeks to 18 months follow‐up) compared with controls (−1.64 to +3.08 kg). The CsQCA showed BWMIs that offered regular contact, tools to support enactment, and tailored materials were associated with effectiveness. As these are all supplementary strategies, it may be possible to augment BWMIs available for the general population to engage people with SMI.


| INTRODUCTION
The global prevalence of overweight (body mass index [BMI] 25-29.9 kg/m 2 ) and obesity (BMI > 30 kg/m 2 ) is increasing and its adverse effects on health are well-documented. 1,2 Overweight and obesity are 2 to 3 times more common in people with serious mental illness (SMI) defined as psychotic disorders like schizophrenia and bipolar disorder. 3 These disorders are often long-term mental health diagnoses marked by hearing, seeing, or believing things that are not real. 4 Antipsychotic medications are sometimes used to manage the symptoms of SMI but contribute to excess weight through increased appetite and metabolic changes. 5 The risk of excess weight and metabolic disturbance appears higher with second-generation drugs, particularly olanzapine and clozapine. 6 Poor diet and physical inactivity also cause excess weight and these are more common in people with SMI compared with the general population. 7 The higher prevalence of overweight and obesity contributes to a higher incidence of cardiovascular disease (CVD) in people with SMI, which is the main factor that reduces their life expectancy by 15 to 20 years. 8 Hence, addressing overweight and obesity in people with SMI is of utmost importance.
In the general population, randomized controlled trials (RCTs) of behavioral weight management interventions (BWMIs) have supported people to follow an energy-restricted diet and increase physical activity. These trials have produced greater weight loss than without support, 9,10 and have shown to reverse type 2 diabetes, lower hypertension, and improve lipid profiles. 11 Accordingly, national guidelines in the United States and United Kingdom suggest offering BWMIs to achieve weight loss for anyone with overweight or obesity. 12,13 These BWMIs are the mainstay treatment for overweight and obesity in many high-income countries and are provided as part of healthcare services. 14 However, people with SMI have reported barriers to engaging with standard BWMIs. 15 These include anxiety in social situations arising from fear of harm from others (i.e., persecutory beliefs) or hearing threatening or critical voices (i.e., auditory hallucinations). 16 Distressing beliefs about oneself related to low self-esteem can undermine persistence with weight loss attempts. 16 People with SMI can also experience difficulties in concentration and motivation. 17 Such barriers have led researchers to develop and test BWMIs that are bespoke for people with SMI.
Previous systematic reviews of these bespoke BWMIs show evidence that, overall, they can be effective but with heterogeneity. For example, Speyer et al. reported BWMIs were effective in reducing weight compared with treatment as usual (TAU) but with moderate heterogeneity: pooled effect = À2.20 kg, 95% CI À3.01 to À1.42 kg, p < 0.001, I 2 = 35.1%. 18 Differences across the intervention characteristics may explain these results. Furthermore, while bespoke BWMIs for people with SMI can be effective, they are rarely provided as part of routine healthcare provision. Therefore, we assess how BWMIs have tailored support to overcome the barriers to engagement people with SMI experience, and assess how differences in these intervention characteristics explain difference in weight loss.
Our aim is to inform researchers on how standard BWMIs may be adapted to better serve people with SMI. Specifically, we a. systematically review qualitative studies to identify which characteristics of BWMIs promote engagement for people with SMI using a thematic analysis; b. systematically review RCTs to identify the characteristics of behavioral weight management interventions associated with weight loss using a crisp-set qualitative comparative analysis (CsQCA).

| METHODS
A protocol was registered in advance and is available in PROSPERO (CRD42020189897). Reporting followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. 19

| Patient and public involvement
We consulted 12 members of the UK public with lived-experience of SMI. We aimed to ensure the research question was relevant and to use their feedback to inform data interpretation. Ethical approval was obtained from the University of Oxford Medical Sciences Interdivisional Research Ethics Committee (R68892/RE001).
The patient and public involvement (PPI) contributors were recruited via local networks within the University of Oxford and The McPin Foundation. We obtained informed consent over the telephone. We then conducted individual telephone interviews or online focus groups between August 14 and October 9, 2020. All discussions were guided by a semistructured topic guide (Appendix A).
In total, we conducted five telephone interviews and two focus groups-one of four contributors, one of three contributors. Each consultation lasted 2 h with scheduled breaks every 30 min. All consultations were facilitated, audio-recorded, and transcribed by the first author. Next, we used a thematic synthesis of the data guided by the Enhancing Transparency in Reporting the Synthesis of Qualitative Research (ENTREQ) guidelines. 20 Thematic synthesis aims to accumulate and summarize descriptive patterns in data rather than transform it for new theories. [21][22][23] Using this method, the first author coded line-by-line each transcript to produce an initial coding frame of intervention characteristics that promote engagement in BWMIs. This coding frame was developed by the lead author and reviewed by members of the research team. The coding frame was then augmented with our systematic review of qualitative studies (Section 2.2 below).

| Eligibility criteria and search strategy
We aimed to review qualitative studies to identify which characteristics of BWMIs promote engagement for people with SMI.
We included peer-reviewed qualitative studies. This included studies reporting any qualitative element of an intervention and RCTs that reported the results of nested qualitative studies. We searched MEDLINE (OvidSP) (1946 to present) from database inception to September 23, 2020, using text word terms (Appendix B).
We also searched for studies that reported qualitative enquiries that aimed to assess the response of people with SMI to eating healthy outside of an intervention. In addition, we searched reference lists of all included studies. We excluded studies that solely focused on children and people without a nonpsychotic mental illness (i.e., eating or neurodevelopmental disorders or stakeholders only).
We also excluded entirely quantitative studies. No restrictions were set on the date of publication, language, or care setting.

| Data synthesis and analysis
We used a thematic synthesis of the data guided by the ENTREQ guidelines. 20 Using this method, data analysis proceeded as follows.
First, we used the coding frame developed from the PPI consultations to inform our subsequent data interpretation. Next, the lead author coded line-by-line the result and discussion sections of the included studies to augment the coding frame with new themes. Codes were then grouped into broader categories of shared meaning. Categories were then summarized to produce top-level analytical themes of intervention characteristics that promote engagement in BWMIs for people with SMI. A second reviewer, who was closely involved with both the PPI consultations and the systematic review of qualitative studies, verified the finalized groupings of analytical themes. Finally, all data were presented to our PPI contributors for validation. Data were coded and managed using NVivo 11 software. 24 Selected quotations are presented in the results section and names have been anonymized.

| Systematic review of randomized trials
We conducted a systematic review of RCTs of BWMIs to identify which characteristics are associated with clinical effectiveness. The systematic search started on June 11, 2020, after the protocol was approved and registered in PROSPERO, though data extraction began once the above intervention characteristics were finalized on October 28, 2020. Methods for the searching, screening, data extraction, and quality assessment of studies followed the Cochrane handbook guidelines. 25

| Eligibility criteria
Articles included met the following criteria: • Population: Adults (aged ≥18 years, no upper limit); with SMI defined by a primary diagnosis of psychosis (i.e., schizophrenia, schizophreniform disorder, schizoaffective disorder, delusional disorder, brief reactive psychosis, psychosis not otherwise specified) or bipolar disorder; and who had overweight (BMI 25-29.9 kg/m 2 ) or obesity (>30 kg/m 2 , no upper limit). Studies on people with a diagnosis of a nonpsychotic mental illness were excluded. There was no restriction on medication use.
• Intervention: Individual or cluster RCTs of any behavioral (i.e., nonpharmacological or bariatric) intervention that aimed to support weight management (i.e., defined as weight maintenance or weight loss) through diet alone or diet and physical activity. To refine the scope of this review, we excluded studies that focused solely on physical activity. No restrictions were set based on intervention characteristics or duration.
• Comparison: Any comparison conditions including other BWMIs or TAU. For studies including another BWMI as a comparison, we isolated the intervention characteristics not included in the control group (i.e., only included in the active intervention group[s]) and recorded these in the data extraction form.
• Outcomes: Mean weight change (kg), BMI (kg/m 2 ), or percentage weight change (kg). When measured on multiple occasions, only data at the first follow-up postintervention was extracted.

| Search strategy
The search strategy was co-developed by the research team with a specialist health science librarian at the University of Oxford. The following databases were searched from database inception until June 11, 2020, using medical subject headings, or similar when possible, or text word terms: Medline, EMBASE (OvidSP) (1974 to present), Psy-chINFO (OvidSP) (1806 to present), and CINAHL (EBSCOHost) (1982 to present). We also searched reference lists of included studies and previous systematic reviews. 18,[26][27][28][29] No year or language limits were set. The Medline search strategy is provided in Appendix C.

| Study selection and data extraction
All studies identified were imported into Covidence for screening. 30 After duplicates were removed, titles and abstracts were doublescreened for eligibility. Discrepancies regarding study inclusion were resolved through discussion. Data were double extracted by five researchers using a piloted form. The data extracted included: participant characteristics (i.e., age, sex, and SMI diagnosis); characteristics of the intervention identified from the qualitative review, as well as characteristics of the control group; length of follow-up; and weight outcomes. Authors were contacted for further information where necessary.

| Risk of bias assessment
Risk of bias (RoB) assessments were conducted in duplicate using the Cochrane risk of bias tool. 25 The following bias domains were assessed as low, high, or unclear risk: allocation sequence generation, allocation concealment, blinding of outcome assessors, incomplete outcome data, selective outcome reporting, and other bias. It is not possible to blind participants or study personnel to allocation in behavioral intervention trials so we omitted this domain.

| Data synthesis and analysis
We did not perform a meta-analysis due to anticipated heterogeneity across intervention design and implementation. Instead, we conducted a narrative synthesis of the data guided by the Synthesis Without Meta-analysis (SWiM) reporting guidelines. 31 Using this approach, we grouped studies by end-of-intervention duration (i.e., ≤6 or 7-12 months). The results were augmented with an exploratory crispset qualitative comparative analysis (CsQCA). 32,33 This method aims to establish causal relationships through systematic comparisons.
Using this method, data analysis preceded in the following stages. The first stage relied on our systematic review of qualitative studies which identified characteristics (i.e., conceptual categories) from the literature. These characteristics formed the conditions that were examined in the CsQCA. In the next stage, each intervention arm (i.e., case) identified from systematic review of randomized trials was coded for either the presence (=1) or absence (=0) of the characteristic. Interventions were also coded as effective (=1) or not (=0) depending on whether there was a statistically significant (p ≤ 0.05) difference in weight at end-of-intervention follow-up. Next, a raw data matrix and truth table were created to code these characteristics and outcomes, which was used in the CsQCA. In interpreting the results of the CsQCA, two concepts were key: consistency and coverage. Consistency refers to the percentage of characteristics that were present in interventions that resulted in a statistically significant between-group difference in weight at follow-up. Consistency is the proportion of times an intervention is effective when a particular characteristic is present. Characteristics that contribute to effectiveness would lead to high consistency (possible range from 0 to 1, with high consistency indicated by ≥0.75). Coverage refers to the proportion of effective interventions in which a particular characteristic is present. Given there are several plausibly effective characteristics, low coverage does not indicate lack of a valid association between cause and effect, only that it is less commonly present in effective interventions.

| Patient and public involvement
Overall, people with a lived experience of SMI recognized the need to manage their weight and were positive about the opportunity for more support. The results of the interviews and focus groups are presented in the coding frame in Appendix D. The coding frame was further developed using the results of the systematic review of qualitative studies and the final (combined) themes are presented below.

| Peer support
Similarly, participants valued opportunities to connect with other participants in the BWMI (e.g., attending an exercise or cookery class together). It was noted when this was absent.
One of the most important things was being part of the group; I enjoyed being with people and not having to do things on my own. 40 [Being] in a group, we have the support, safety and strength from your friends rather than being frightened or anxious with strangers. 50

| Interim booster support
People with SMI reported difficulties initiating weight loss tasks owing to fluctuating symptoms, medication side effects, and varying motivation. Participants valued proactive support between sessions (e.g., telephone calls) to help translate intentions into action. It also provided an added opportunity to foster therapeutic rapport with the person who was facilitating the intervention, and reduced feelings of isolation. Text or phone OK. 48 … having somebody to report to … it makes me feel good to say "Shirley, I went to the gym three times this week," and she's proud of me because I did it. That's important to me, having somebody to say I did it …. 43

| Supporting tools
Participants valued tools (e.g., intervention handbooks, pedometers, cookery books) that could help initiate a weight loss activity.
The introduction of supporting tools … supported the messages provided to participants about the benefits of participation, improved internal motivation, and supported engagement and attendance. 48

| Tailored materials
Tailored content (e.g., materials written in plain and simple language) and structure (e.g., shorter or repeated sessions) could make it easier for participants to engage in the intervention while experiencing symptoms of SMI (e.g., psychotic experience or anxiety).   3.3 | Systematic review of randomized trials

| Study selection
As shown in Figure 2, the title and abstracts of 2121 unique studies were screened. Full-text studies were assessed for 184 records. In total, 34 studies met the inclusion criteria and were included for the CsQCA.  Two studies were included twice in the CsQCA because they each contributed to two intervention arms. 71 64,65,70,83 The behavioral therapy comprised goal-setting and problem-solving strategies to promote control over calorie intake and cues to eat.
The comparison group were offered TAU (i.e., no weight loss support) in all but three studies. 59,72,73 In one study, the control group received a monthly newsletter about healthy eating. 59 In the other two studies, the control group were offered a free membership to the same local fitness club plus educational materials without access to a health mentor. 72,73 On average, BWMIs included a mean of three of the nine intervention characteristics identified in the qualitative thematic analysis.
The BWMIs that were specific to people with SMI included, on average, six of the characteristics. Across all studies, the most common characteristic was an intervention that was facilitated by a mental health professional, which was included in 25 interventions representing 23 studies. 54 In five interventions, this was a weekly telephone call, 59,60,71,75,83 and fortnightly in one intervention. 78 The nature of the call was not specified. The mean weight change in the intervention groups lay between À4.37 to +1 kg at 6 weeks to 18 months follow up, compared with À1.64 to +3.08 kg in the control group.
A summary of participant-and study-level characteristics is provided in Table 2 (see also Appendices F and G).

| Risk of bias
Sixteen studies were judged to be at high risk of bias. 54

| Qualitative comparative analysis
The results from the exploratory CsQCA are presented in Table 3 and Appendix I. The characteristic with most support for effectiveness was supporting tools, which meant prompts like pedometers and cookery books. The consistency was 0.60 implying that in 60% of interventions that supporting tools were used, the intervention was shown to be effective. The coverage was 0.42, meaning that 42% of effective interventions included this characteristic. Interim booster support was linked with a significant difference in weight loss in favor of the intervention compared with the control 60% of the time and was included in 21% of the effective interventions. Tailored materials achieved a consistency rating of 58% and coverage of 50%.
The variety of configurations suggested that no single characteristics or combination of characteristics accounted for all weight loss outcomes. Therefore, we examined patterns among those con-

figurations. Each configuration represents an intervention scenario
that is linked to weight loss. An initial examination of these configurations revealed that some configurations appear more consistently than others. The following configurations had the highest consistency and highest coverage: (1) interim booster support plus tailored materials, (2) interim booster support plus a knowledgeable facilitator, (3) interim booster support plus supporting tools. The consistency of these configurations was 0.75 and coverage was 0.21 (see Table 3).

| Overview of findings
In the systematic review of qualitative studies, nine characteristics were identified as promoting engagement for people with SMI in weight management interventions. These included the following: (1) education on the specific contributors to weight gain for people with SMI, (2) emphasis on success and achievements, (3) a knowledgeable facilitator, (4) peer support, (5) interim booster support, supporting tools, (7) tailored materials, (8) practical support, and (9) incentives. In the systematic review of RCTs, three of these characteristics were most commonly associated with weight loss. First, interventions that offered supporting tools like pedometers and cookery books. Second, interventions that offered interim booster support between sessions such as low-intensity telephone calls. Third, interventions that tailored the materials and session structure to account for the impact of a mental health diagnosis-such as low motivationoften faced by people with SMI. There was little evidence that including other intervention characteristics improved effectiveness.

| Strengths and limitations
The protocol was published a priori and we used gold standard Note: In crisp-set qualitative comparative analysis (CsQCA), each intervention characteristic scores 1 or 0 to describe whether the intervention did or did not have the characteristic of interest. The outcome in our analysis was whether or not the intervention was associated with statistically significant changes in weight in the desired direction (i.e., weight loss or weight gain prevention). Together these scores form an intervention's configuration, which is the set of conditions associated (=1) or not associated (=0) with statistically significant changes in the outcome. a Consistency represents the proportion of times interventions were effective when that characteristic was present. Coverage indicates the proportion of interventions that were effective that included this characteristic. The interventions included in this review undoubtedly differed in characteristics that are common to BWMIs for the general population, and variation in the effectiveness between them could be explained by these other generic behavioral characteristics. That said, a previous review that examined these characteristics found little evidence that variation in their inclusion explained variation in effectiveness. 90 We found some of the most effective characteristics of interven-

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author (CL) upon reasonable request. $ no difference in outcome (i.e., no change in weight); + outcome change in desired direction (i.e., weight loss); À outcome change in undesired direction (i.e., weight gain); NS, not significant. Outcome results: Time point closest to the intervention completion.