Main Outcomes of a Peer-Led Healthy Lifestyle Intervention for People With Serious Mental Illness in Supportive Housing

Compared with the general population, people with serious mental illness (e.g., schizophrenia) continue to have reduced life expectancy because of higher rates of obesity, type 2 diabetes, and cardiovascular disease (CVD) (1). Belonging to a racial-ethnic minority group exacerbates these health inequities. Hispanics and Blacks with serious mental illness have elevated risk of obesity and type 2 diabetes, compared with non-Hispanic Whites with serious mental illness (2, 3). Reducing obesity among persons with these mental health conditions, particularly in racial-ethnic minority groups, is important for improving the health of these historically underserved populations.

Clinical trials based in the United States have shown that lifestyle interventions that increase healthy dietary habits and physical activity can improve general health of individuals with serious mental illness through weight reduction and improved cardiorespiratory fitness (CRF)—that is, the circulatory and respiratory systems’ ability to supply oxygen to working muscles during physical activity (4–7). Despite these results, several knowledge gaps remain. First, these trials used clinical staff in interventions, limiting the workforce that can deliver these interventions. The use of peer specialists, that is, people with lived experience of serious mental illness, to deliver healthy lifestyle interventions is a promising approach to increase the reach of these interventions, because peer specialists are a growing workforce in the United States (8). Rigorous studies are needed to test the effect of peer-led interventions on the health of people with serious mental illness (9). Second, these trials were conducted in clinical settings (e.g., outpatient clinics), restricting the accessibility of the interventions. Moving interventions closer to the community by delivering them in supportive housing agencies can help address access barriers by bringing interventions to “people’s doorsteps” (10). Supportive housing programs are an important setting for people with serious mental illness because they provide general medical, mental health, and social services (11). Finally, persons with serious mental illness from racial-ethnic minority groups are underrepresented in existing trials. However, two recent trials have actively recruited persons from racial-ethnic minority groups: 46% of participants in the InShape Replication trial (5) were largely Black and Hispanic, and 38% of ACHIEVE trial participants were Black (6).

To further close these gaps, we conducted a hybrid type 1 trial to pragmatically examine the implementation and test the effectiveness of a 12-month program, the Peer-led Group Lifestyle Balance (PGLB) intervention, in a sample comprising persons with serious mental illness predominantly from racial-ethnic minority groups who were overweight or obese (body mass index [BMI] ≥25) and living in supportive housing. This design tested the effectiveness of the intervention while collecting data about the intervention’s potential for community implementation (12). Here, we present results for our main effectiveness outcome, clinically significant weight loss, and our secondary outcomes, clinically significant improvements in CRF and reductions in CVD risk. We hypothesized that significantly larger proportions of PGLB participants, compared with participants receiving usual care, would achieve clinically significant weight loss, improved CRF, and reduced CVD risk at 12 and 18 months postrandomization, regardless of study site.

Methods

Study Overview

The trial protocol has been published elsewhere (13). This trial was conducted in three supportive housing agencies located in two northeastern U.S. cities. One site follows a Housing First model (14), and two follow a treatment-first model (11). All participants gave written informed consent, and the study was approved by the institutional review boards of Columbia University and the Philadelphia Department of Public Health.

Participants

Following a pragmatic trial design, the study inclusion criteria were minimal (15). Eligible participants were residents of their supportive housing agency, ages ≥18 years, and English or Spanish speakers; they had a chart diagnosis of serious mental illness and a BMI of ≥25 kg/m² assessed by a research assistant. Participants randomly assigned to PGLB obtained medical clearance from their primary care physician. We excluded participants who at the time of recruitment required detoxification services, posed a danger to self or others, failed a capacity-to-consent questionnaire (16), or self-reported medical conditions that contraindicated participation in a weight-loss program (e.g., cancer or stroke); for participants ages >64 years, those who received a score indicating cognitive impairment on the Mini-Cog clock test also were excluded (17).

Study Procedures

Participants were recruited between June 2015 and January 2018 via word of mouth and staff referrals. The participants were screened for study eligibility by the study team at each agency. Independent assessors employed by the study team and not blinded to participants’ group assignment conducted face-to-face interviews at the participant’s supportive housing agency at baseline and 6, 12, and 18 months postrandomization. After the baseline interview, randomization to PGLB or usual care was conducted at the participant level in blocks of four, stratified by site. Participants received $25 for completing each assessment but not for attending PGLB sessions. Measurement protocols are described elsewhere (13).

Intervention

PGLB is a 12-month, manualized healthy lifestyle intervention, delivered by peer specialists and adapted from the Group Lifestyle Balance (GLB), to meet the needs of people with serious mental illness living in supportive housing. Adaptations were reviewed by GLB developers, who deemed them consistent with GLB’s core components (18). PGLB consisted of weekly core sessions for 3 months, twice-monthly transition sessions for 3 months, and monthly maintenance sessions for 6 months, for a total of 22 sessions. Sessions lasted approximately 60 minutes and were delivered to groups of three to six participants in their housing agency, with the option of receiving individual sessions. PGLB focused on using behavioral techniques to improve dietary habits and physical activity (18).

Peer specialists delivering PGLB were employed at their respective housing agencies and were trained and supervised by the study team. Training included a 2-day GLB certification program delivered by a GLB master trainer and a 3-month session-by-session training that included using intervention elements (e.g., food logs) in their everyday lives and delivering mock sessions to supervisors before facilitating the intervention. Throughout the trial, study team members monitored intervention fidelity by reviewing session audio recordings and rating the degree to which key PGLB elements were present. To avoid intervention drift, weekly supervision meetings were held in person or by telephone (19).

Usual Care

All participants continued to receive usual care for general medical health throughout the trial. These services consisted of health promotion groups (e.g., cooking groups) and linkages to medical care and community resources (e.g., gyms). The health promotion groups were not manualized interventions and focused on health education. Agency staff (e.g., case managers) at study sites helped clients connect with primary care and specialized health services as needed. The use of usual care services was tracked at each assessment period.

Primary Outcome

The main outcome was the proportion of participants who achieved clinically significant weight loss, defined as weight loss of ≥5% total body weight from baseline at 12 and 18 months. Weight (in pounds) was measured by a research assistant with a calibrated digital scale (participants wore indoor clothing without shoes).

Secondary Outcomes

Secondary outcomes included the proportion of participants who achieved clinically significant improvements in CVD risk reduction and CRF at 12 and 18 months. CRF was measured with the 6-minute walk test (6MWT), an objective measure of functional exercise capacity that captures the distance (in meters) that participants walk at a normal pace along a flat and straight course for 6 minutes (20). The 6MWT is a reliable and valid measure among obese adults and has been used in previous trials that included persons with serious mental illness (4). Consistent with past studies, clinically significant improvement in CRF was defined as an increase of ≥50 meters in the 6MWT from baseline (4, 5). This level of improvement is associated with reduction in CVD risk (20, 21). Consistent with previous trials, clinically significant reduction in CVD risk was defined as either weight loss of ≥5% from baseline or an increase of ≥50 meters on the 6MWT from baseline (5).

Data Analysis

Bivariate analyses were used to examine baseline differences between PGLB and usual care groups in demographic factors and clinical variables. An intent-to-treat approach was used for all analyses. Logistic regression models tested our main hypothesis on the three dichotomous outcomes at 6, 12, and 18 months, comparing PGLB versus usual care and using listwise deletion for missing data. Generalized linear mixed-effects models with a logistic link explored within- and between-group changes in trends over time for each outcome. All models adjusted for study site. The models examining the main outcome and clinically significant reductions in CVD risk were adjusted for baseline weight. The models for all secondary outcomes were adjusted for baseline weight and 6MWT.

Because of baseline imbalances between PGLB and usual care groups in BMI, weight, and number of medical conditions (Table 1), sensitivity analyses that used the inverse probability of treatment weighting estimator were employed to correct for any selection biases generated by group assignment (22). This method incorporates the propensity score weights into the multivariate analyses to estimate the average treatment effect for the entire sample and average treatment effect for the treated samples (results are presented in a table in an online supplement to this article). Because we observed no differences between our primary and sensitivity analyses, we report results only of the primary analyses. Study site was a significant factor in our models, and we therefore conducted post hoc analyses to explore site differences. We conducted the aforementioned analyses for each outcome stratified by site. Two-tailed statistical tests were conducted, and p<0.05 was used to establish statistical significance.

Results

Sample Characteristics

Of 448 people screened, 340 were eligible, and 314 were enrolled and randomly assigned to PGLB or usual care (see flow diagram in the online supplement). The baseline characteristics of all study participants are given in Table 1. At baseline, the mean age of the participants was 48.7 years, and 43% were female. Most were from racial-ethnic minority groups (82%, N=255), with 58% being non-Hispanic Blacks. Most (62%) had at least a high school education. Most (90%) were unemployed, in the Supplementary Nutrition Assistance Program (91%), receiving Supplemental Security Income or Social Security Disability Insurance (78%), and covered by Medicaid (83%).

The most common lifetime psychiatric diagnoses at baseline were major depression (80%) and schizophrenia or schizoaffective disorder (57%), and 39% reported a lifetime alcohol or substance use disorder. At baseline, participants had a mean BMI of 33.7, and their mean weight was 218.8 pounds. On average, participants reported 3.7 general medical illnesses, most commonly hypertension (56%), high cholesterol (37%), and diabetes mellitus (33%), and 63% were current smokers. At baseline, 32% reported using usual care services in the past 6 months. PGLB and usual care groups did not differ significantly in any baseline characteristics, except for weight and BMI (which were higher for the PGLB group) and number of medical conditions (higher for the usual care group). Follow-up data collection was completed by 93% of participants (N=293) at 6 months, 84% (N=265) at 12 months, and 80% (N=252) at 18 months, with no differential attrition between groups at each time point. Missing data did not vary with group assignment.

Participation in PGLB and Usual Care

The median of total PGLB sessions attended was 18 of 22, with 59% of participants (N=93) attending ≥50% of sessions, and 36% (N=57) attending all 22 sessions. Most sessions (63%) were delivered in groups. PGLB and usual care participants did not statistically significantly differ in the use of usual care services for general medical health throughout the study (see online supplement).

Primary Outcome

Compared with the usual care group, a larger proportion of the PGLB group achieved clinically significant weight loss at 12 and 18 months, but none of these differences between the two groups was statistically significant (Table 2). The increases in the proportion of participants in both groups who achieved clinically significant weight loss from 6 to 18 months were statistically significant (PGLB, adjusted odds ratio [AOR]=2.26; usual care, AOR=1.90), indicating that participants in both groups achieved significant weight loss during the study period.

Secondary Outcomes

No statistically significant differences were detected for mean weight loss and mean increases in 6MWT between the usual care and PGLB groups (see online supplement). A larger proportion of PGLB participants, compared with usual care participants, achieved clinically significant improvements in CRF at 6, 12, and 18 months, yet none of these differences was statistically significant (Table 2). The increases in the proportion of participants from both groups achieving clinically significant improvements in CRF during the study periods were likewise not statistically significant. Compared with the usual care group, a larger proportion of the PGLB group achieved clinically significant reductions in CVD risk at 6 and 12 months, but these reductions were not statistically significant. The increases in the proportion of participants achieving clinically significant reductions in CVD risk over the course of the study were statistically significant for PGLB participants (AOR=1.54) and for those receiving usual care (AOR=1.73), indicating that CVD risk decreased for both groups during the study.

Site Differences

Study outcomes differed by study site (Table 3). At sites 1 and 2, individuals receiving usual care tended to experience better outcomes than did PGLB participants at all time points, with some minor exceptions, but none of the differences was statistically significant. Site 3 showed a different pattern. At this site, PGLB consistently outperformed usual care at all time points, particularly for the proportion achieving clinically significant weight loss at 18 months (PGLB, 42%, and usual care, 22%; AOR=2.57) and clinically significant reductions in CVD risk at 6 months (PGLB, 48%, and usual care, 27%; AOR=2.51) and 12 months (PGLB, 59%, and usual care, 33%; AOR=2.99).

Discussion

Our findings did not support our hypothesis. Although a larger proportion of participants in the PGLB group achieved clinically significant weight loss, increases in CRF, and reductions in CVD risk at 12 and 18 months, compared with usual care, differences between groups were not statistically significant. PGLB’s impact, relative to that of usual care, differed by study site.

These null findings do not appear to be a result of PGLB underperformance. Instead, our results indicate that PGLB achieved outcomes comparable to those from other U.S.-based healthy lifestyle trials involving people with serious mental illness (5–7) and those reported in a recent meta-analysis (23). The ACHIEVE trial tested the effectiveness of an 18-month behavioral weight-loss intervention for people with serious mental illness in psychiatric rehabilitation programs (6). ACHIEVE was associated with clinically significant weight loss for 32.5% of participants at 12 months and 37.8% at 18 months (6), comparable to the 29% and 32%, respectively, for these time points in the present study. The InShape trial tested the effectiveness of a 12-month health promotion program consisting of individualized sessions with a health coach and free gym memberships (5). InShape was associated with clinically significant reductions in CVD risk for 51% of participants at 12 months and 46% at 18 months (5), similar to the 48% and 49%, respectively, for the same outcome and time points among PGLB participants in our study. Our findings suggest that PGLB delivered in supportive housing can produce clinically significant health improvements in racially and ethnically diverse samples of persons with serious mental illness that are consistent with the outcomes of other non–peer-led healthy lifestyle interventions.

The lack of significant differences between the PGLB and usual care groups in the outcomes of the present study indicates that several methodological and contextual factors need to be considered. The imbalance at baseline between our two groups in weight—an average lower weight of 14 pounds for the usual care group—suggests that PGLB participants needed to achieve greater weight loss than those receiving usual care to counteract this difference. Although we corrected for this imbalance in our analyses, the PGLB group, on average, still did not achieve greater weight loss than the usual care group.

The use of usual care services for support of general medical health at our study sites might have increased during the study period, with more use among persons in the usual care group, thus influencing their weight loss, CRF, and CVD risk reduction outcomes. Our data do not support this pattern, because we observed no differential use of usual care services between groups, and the use of usual care decreased throughout the trial for both groups (see online supplement).

We noted improvements in study outcomes for the usual care group during the trial, which mirrored improvements in the PGLB group, suggesting that some usual care participants engaged in weight-loss strategies. Contamination between groups could account for these improvements in the usual care group, because we randomized at the participant level because of the small number of study sites. A potential source of contamination was the PGLB peer specialists, who were employees of the supportive housing agencies. However, we note that PGLB peer specialists’ sole responsibility was to deliver PGLB to participants randomly assigned to the intervention, and they had little contact with usual care participants.

Another source of contamination could have been interactions throughout the trial between PGLB and usual care participants, leading to sharing of PGLB materials and strategies for weight loss, thereby activating usual care participants to lose weight. To check for contamination, we used a six-item self-report measure asking all participants at 6, 12, and 18 months whether they engaged in PGLB strategies. On average, PGLB participants reported significantly greater engagement in PGLB strategies throughout the trial than did usual care participants (see table in online supplement). However, >40% of usual care participants reported engaging in PGLB strategies, particularly tracking their eating and exercise, and setting weight and exercise goals. Although this may have influenced our findings, it suggests that elements of PGLB resonated with participants to the point of their possibly sharing materials and strategies with usual care participants. A social proliferation process may have occurred whereby PGLB participants shared their new knowledge with their social circles within these agencies, helping diffuse PGLB strategies and activating usual care participants to engage in healthy lifestyle changes (24). More research is needed to better understand how this social proliferation process may have unfolded and to evaluate its impact on helping participants achieve clinically significant changes in weight and CRF.

Losing weight and improving CRF are major challenges for people with serious mental illness who are overweight or obese (25). Effective interventions use intensive manualized programs that combine coached and structured physical activity, support dietary changes with behavioral techniques, and last >9 months (26). Our null findings suggest that for some participants enrolling in a weight-loss study and having their weight, CRF, and other health indicators assessed every 6 months, and potentially learning about PGLB from other participants, may have activated them to lose weight and improve their CRF. A similar finding was reported in a small study of patients from a Hispanic background who attributed their weight loss at 40 weeks after baseline to regular weight checkups by staff, which increased their attention to weight-related issues (27). A one-size-fits-all healthy lifestyle approach may not be the most efficient way to help people with serious mental illness. Personalized strategies tailored to an individual’s characteristics, needs, and preferences may be needed for persons with serious mental illness; some may require intensive and structured behavioral approaches, whereas others may need fewer strategies, such as frequent weight monitoring and goal setting.

PGLB benefits were not uniform across the three study sites. At site 3, PGLB consistently outperformed usual care. This site included 40% of PGLB participants but accounted for more than half of the participants who achieved clinically significant weight loss (51%), improved CRF (68%), and reduced CVD risk (55%) at 18 months. We know of no published effectiveness trial of healthy lifestyle interventions for people with serious mental illness that has reported or examined site differences. More in-depth analyses of the qualitative and quantitative data collected as part of our hybrid trial are required to examine possible heterogeneity among sites, including whether potential differences in fidelity to the intervention, staff turnover, participants’ experiences with PGLB, or other contextual factors could explain these site differences. We are in the process of conducting these analyses and plan to publish their results in the future. This exploration may uncover contextual and implementation factors that are necessary to identify how and why PGLB works in certain settings and not others and that can inform the development of implementation strategies for PGLB (28, 29). More studies are needed, because little is known about the type of implementation strategies that can support adoption of healthy lifestyle interventions for people with serious mental illness in routine practice settings.

Our trial had several limitations. Because of logistical constraints, we did not use assessors blinded to participants’ group assignment (PGLB vs. usual care) to measure study outcomes, which could have biased study assessments. We conducted our trial in three supportive housing agencies that could be considered early adopters because these agencies typically do not include health interventions as part of their services. Studies with larger samples of supportive housing agencies are needed to examine the generalizability of our findings. The baseline imbalance in body weight that favored the usual care group suggests that the randomization was not successful because of unknown factors. Although we adjusted for these imbalances in our analyses, the differences in weight at baseline between study conditions potentially threatened the internal validity of our findings (30). Future studies could use different randomization strategies, such as stratifying randomization by BMI ranges (e.g., BMI, 25–29 and BMI ≥30) to reduce imbalance. Potential contamination between the study arms also may have contributed to our null findings. Future trials could avoid contamination by recruiting a larger sample of supportive housing sites and using a clustered randomized design in which sites rather than individuals are randomly assigned to study conditions. Most of the participants in our sample belonged to racial-ethnic minority groups, particularly non-Hispanic Blacks, and we did not have the statistical power to examine intervention differences among racial-ethnic groups. Additional studies are therefore needed to examine differential intervention effects among different minority groups.

Conclusions

Our null findings indicate that PGLB was not superior to usual care in helping participants achieve clinically significant weight loss, increases in CRF, and reductions in CVD risk. Although our findings suggest that some persons with serious mental illness from racial-ethnic minority groups who are overweight or obese and living in supportive housing could benefit from a peer-led healthy lifestyle intervention, multiple questions remain about how the PGLB intervention works and for whom and under which conditions it exerts the biggest impact. More research is needed to clarify why PGLB works in certain settings and not others. Increasing the life expectancy of people with serious mental illness requires bridging the gap between research and practice and developing evidence for how to best implement health interventions in routine practice settings to increase their reach and benefits.