Introduction
Peer support (PS) is an increasingly recognised psychosocial intervention in which people with lived experience of a mental health condition offer support to others facing similar challenges (Davidson et al., Reference Davidson, Chinman, Sells and Rowe2006). In global mental health, PS is increasingly gaining traction as a recovery-oriented, rights-based approach, particularly for individuals with severe mental health conditions (Stratford et al., Reference Stratford, Halpin, Phillips, Skerritt, Beales, Cheng, Hammond, O’Hagan, Loreto, Tiengtom, Kobe, Harrington, Fisher and Davidson2017; Fan et al., Reference Fan, Ma, Ma, Xu, Steven Lamberti and Caine2018; de Wet et al., Reference de Wet, Sunkel and Pretorius2022; Triece et al., Reference Triece, Massazza and Fuhr2022). While growing evidence supports its effects on psychosocial outcomes such as social inclusion, hope and empowerment (White et al., Reference White, Foster, Marks, Morshead, Goldsmith, Barlow, Sin and Gillard2020; Yim et al., Reference Yim, Chieng, Tang, Tan, Kwok and Tan2023; Cooper et al., Reference Cooper, Saunders, Greenburgh, Shah, Appleton, Machin, Jeynes, Barnett, Allan, Griffiths, Stuart, Mitchell, Chipp, Jeffreys, Lloyd-Evans, Simpson and Johnson2024), findings on broader impacts on personal recovery, clinical symptoms and service use remain inconsistent (White et al., Reference White, Foster, Marks, Morshead, Goldsmith, Barlow, Sin and Gillard2020; Yim et al., Reference Yim, Chieng, Tang, Tan, Kwok and Tan2023).
Recovery in this context refers to a multidimensional concept that goes beyond symptom reduction and includes connectedness, hope, identity, meaning and empowerment (Leamy et al., Reference Leamy, Bird, Le Boutillier, Williams and Slade2011). Constructs such as hope and social inclusion reflect important but partial aspects of this broader recovery concept. Qualitative research has offered valuable insights into the perceived impact and underlying processes of PS (King and Simmons, Reference King and Simmons2018; Cooper et al., Reference Cooper, Saunders, Greenburgh, Shah, Appleton, Machin, Jeynes, Barnett, Allan, Griffiths, Stuart, Mitchell, Chipp, Jeffreys, Lloyd-Evans, Simpson and Johnson2024). A key theoretical contribution is the Change Model of Peer Support, which conceptualises the effects of PS as indirect, operating through intermediate mechanisms including hope, empowerment and strengthened social networks (Gillard et al., Reference Gillard, Gibson, Holley and Lucock2015).
Despite growing conceptual clarity, quantitative evidence on these hypothesised mechanisms remains limited. Addressing this gap is critical for informing implementation strategies and the further development of evidence-based PS services. The UPSIDES project (Using Peer Support In Developing Empowering Mental Health Services) investigated the effects and implementation of PS across high-, middle- and low-income countries using a mixed-methods approach (Puschner et al., Reference Puschner, Repper, Mahlke, Nixdorf, Basangwa, Nakku, Ryan, Baillie, Shamba, Ramesh, Moran, Lachmann, Kalha, Pathare, Müller-Stierlin and Slade2019; Moran et al., Reference Moran, Kalha, Mueller-Stierlin, Kilian, Krumm, Slade, Charles, Mahlke, Nixdorf, Basangwa, Nakku, Mpango, Ryan, Shamba, Ramesh, Ngakongwa, Grayzman, Pathare, Mayer and Puschner2020). While prior studies showed positive effects on social inclusion, hope and empowerment, no direct effects were observed on recovery or health and social functioning (Puschner et al., Reference Puschner, Nakku, Hiltensperger, Wolf, Ben-Dor, Bugeiga and Slade2026). However, qualitative research and the UPSIDES Theory of Change suggest indirect pathways linking PS to recovery outcomes (Krumm et al., Reference Krumm, Haun, Hiller, Charles, Kalha, Niwemuhwezi, Nixdorf, Puschner, Ryan, Shamba, Epstein and Moran2022; Haun et al., Reference Haun, Adler Ben-Dor, Hall, Kalha, Korde, Moran, Müller-Stierlin, Niwemuhwezi, Nixdorf, Puschner, Ramesh, Charles and Krumm2024; Hiltensperger et al., Reference Hiltensperger, Ryan, Ben-Dor, Charles, Epple, Kalha, Korde, Kotera, Mpango, Moran, Mueller-Stierlin, Nixdorf, Ramesh, Shamba, Slade, Puschner and Nakku2024b). This study examines whether social inclusion, hope and empowerment mediate the relationship between PS and outcomes in personal recovery and health and social functioning.
Method
Study design
The UPSIDES project was a five-year international implementation initiative aiming to scale up PS for people with severe mental illness across diverse healthcare systems (Puschner et al., Reference Puschner, Repper, Mahlke, Nixdorf, Basangwa, Nakku, Ryan, Baillie, Shamba, Ramesh, Moran, Lachmann, Kalha, Pathare, Müller-Stierlin and Slade2019). It combined qualitative, quantitative and participatory methods to evaluate effectiveness, implementation processes and contextual determinants across high-, middle- and low-income countries. Within this broader programme, a pragmatic multicentre randomised controlled trial was conducted to examine the clinical and psychosocial effects of PS and to generate robust comparative evidence across sites (Moran et al., Reference Moran, Kalha, Mueller-Stierlin, Kilian, Krumm, Slade, Charles, Mahlke, Nixdorf, Basangwa, Nakku, Mpango, Ryan, Shamba, Ramesh, Ngakongwa, Grayzman, Pathare, Mayer and Puschner2020).
Data for the present analysis were collected between 2020 and 2022 at four time points: baseline (t0), four months (mid-point, t1), eight months (post-intervention, t2) and 12 months (follow up, t3) in six study sites (Ulm and Hamburg, Germany; Kampala, Uganda; Dar es Salaam, Tanzania; Be’er Sheva, Israel; and Ahmedabad, India). Following inclusion, participants were randomised into either the intervention group or the waitlist control group. Participants in the intervention group engaged in their first intervention session within one month after randomisation.
UPSIDES PS is a structured, formal intervention in which trained individuals with personal experience of mental health recovery provide support to people with serious mental health conditions (Nixdorf et al., Reference Nixdorf, Nugent, Aslam, Barber, Charles, Gai, Grayzman, Hiltensperger, Kalha, Korde, Mtei, Niwemuhwezi, Rameshh, Ryanb, Sladed, Wenzel and Mahlke2022). The intervention is grounded in core PS principles, including mutuality, reciprocity and a strength-based, recovery- and community-oriented approach. Interactions between PSWs and participants took place in individual or group formats, or a combination of both, adapted to the respective site context. Sessions were intended to take place weekly or fortnightly over a period of up to six months. For the present analyses, per-protocol participation was defined as attending a minimum of three sessions within this period, though frequency was otherwise flexible and could be adjusted to the needs of participants and PSWs.
PSWs completed the UPSIDES training programme prior to delivering the intervention, which comprises 12 core modules addressing topics such as the PSW role, recovery, communication and trialogue (Mahlke et al., Reference Mahlke, Nixdorf and Repper2020; Nixdorf et al., Reference Nixdorf, Kotera, Baillie, Garber Epstein, Hall, Hiltensperger, Korde, Moran, Mpango, Nakku, Puschner, Ramesh, Repper, Shamba, Slade, Kalha and Mahlke2024). The training was initially delivered in a train-the-trainer workshop and subsequently refined through site-specific focus groups. Throughout the intervention period, PSWs received regular supervision and additional support to ensure quality and well-being.
This mediator analysis was guided by the aforementioned change model (Gillard et al., Reference Gillard, Gibson, Holley and Lucock2015), which identifies, among other factors, social inclusion and hope as mediators and recovery and well-being (in this study, conceptualised as health and social functioning) as outcomes. The predictor PS is indicated by group allocation (intervention vs. waitlist control). Self-report and staff-rated measures were used to assess social inclusion, hope, empowerment, recovery and health and social functioning.
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2013. All procedures involving study participants were approved by: Ethics Commission of Ulm University, Germany (ref. 254/19), the Local Psychological Ethics Commission at the Centre for Psychosocial Medicine, Hamburg, Germany (ref. LPEK-0095), the Uganda National Council for Science and Technology (ref. SS 4990), the National Institute for Medical Research, Dar es Salaam, and Ministry of Health, Community Development, Gender, Elderly & Children, Dodoma, Tanzania (ref. NIMR/HQ/R.8a/Vol. IX/3328), the Human Subjects Research Committee of Ben-Gurion University, Israel (ref. 1787–1) and the Indian Law Society (ref. ILS/37/2018). Informed consent was obtained from all participants, and data were anonymised in compliance with the General Data Protection Regulation (GDPR) and relevant local data protection laws.
This paper adheres to the AGReMA reporting guidelines for mediation analyses (Lee et al., Reference Lee, Cashin, Lamb, Hopewell, Vansteelandt, VanderWeele, MacKinnon, Mansell, Collins, Golub, McAuley, Localio, van Amelsvoort, Guallar, Rijnhart, Goldsmith, Fairchild, Lewis, Kamper, Williams and Henschke2021) to enhance the transparency and reproducibility of the mediation analysis; see Supplementary File 1. The mediation analysis was conducted as outlined in the published study protocol (Moran et al., Reference Moran, Kalha, Mueller-Stierlin, Kilian, Krumm, Slade, Charles, Mahlke, Nixdorf, Basangwa, Nakku, Mpango, Ryan, Shamba, Ramesh, Ngakongwa, Grayzman, Pathare, Mayer and Puschner2020).
Participants
Participants were persons with severe mental health conditions. Inclusion criteria were adult age (18–60 years), a mental health condition with substantial functional impairment (Threshold Assessment Grid [TAG] [Rosen et al., Reference Rosen, Strathdee, Slade and Powell2000] > 5, illness duration > 2 years), sufficient proficiency in the host country’s language and the capability to give informed consent. Exclusion criteria were primary diagnoses of learning disability, dementia, substance use disorder or organic brain disorder, as well as cognitive impairments that prevented informed consent or completion of assessment. Potential participants were recruited via mental health services (inpatient, outpatient or community), patient and carer organisations, print and social media and word-of-mouth.
Measures
The measures used in this study included self-reported scales of hope, empowerment, social inclusion and recovery, as well as staff-rated health and social functioning. This combination of self-reported and staff-rated measures was used to capture both personal perceptions and external assessments. Social inclusion was measured using the Social Inclusion Scale (SIS) (Secker et al., Reference Secker, Hacking, Kent, Shenton and Spandler2009), which comprises 16 items on a four-point Likert scale, with total scores from 16 (low social inclusion) to 64 (high social inclusion). Hope was assessed with the HOPE scale (Snyder et al., Reference Snyder, Harris, Anderson, Holleran, Irving, Sigmon, Yoshinobu, Gibb, Langelle and Harney1991), an eight-item self-report questionnaire (plus four filler items), rated on an eight-point Likert scale. The total scores range from 8 (low hope) to 64 (high hope). Empowerment was measured using the Empowerment Scale ES (Rogers et al., Reference Rogers, Ralph and Salzer2010), which consists of 28 items on a four-point Likert scale, with total scores ranging from 28 (low empowerment) to 112 (high empowerment). Recovery was assessed using the Stages of Recovery Instrument STORI-30 (Andresen et al., Reference Andresen, Caputi and Oades2013), which consists of 30 items rated on a six-point Likert scale. This instrument consists of five subscales that capture five stages of recovery, ranging from stage 1 (Moratorium, lowest stage) to stage 5 (Growth, highest stage). The subscale with the highest score determines the recovery stage the person currently reports being in. The Health of the Nation Outcomes Scales (HoNOS) (Wing et al., Reference Wing, Beevor, Curtis, Park, Hadden and Burns1998) is a staff-rated tool designed to assess the severity of health and social problems. It covers 12 domains, such as physical health, mood, cognitive functioning or daily living activities. Each scale is rated from 0 (no problem) to 4 (severe problem), with higher scores indicating clinically significant concerns that may negatively affect a person’s health and social functioning and well-being. For the present analyses, the mean score across all 12 items was calculated (range 0–4), referred to as health and social functioning, and standardised values were used. In UPSIDES, trained research staff with professional backgrounds in psychology, psychiatry or social work conducted the HoNOS ratings.
For languages in which the measures were not readily available, translations were conducted into Hebrew, Gujarati, Luganda, Swahili or German, in accordance with the UPSIDES proportionate translation guidelines (Charles et al., Reference Charles, Korde, Newby, Grayzman, Hiltensperger, Mahlke, Moran, Nakku, Niwemuhwezi, Nixdorf, Paul, Puschner, Ramesh, Ryan, Shamba, Kalha and Slade2022). This methodology includes a structured sequence of forward translation, reconciliation and expert-panel review, back translation, harmonisation across sites and cognitive pretesting with service users and mental health workers. For the outcome measure for social inclusion, an additional psychometric evaluation was conducted, demonstrating adequate internal consistency, convergent validity and minimal floor/ceiling effects.
Analysis
Cross-lagged panel modelling (CLPM) within a structural equation modelling (SEM) framework was conducted to examine whether PS influenced participants’ scores on recovery (STORI) and health and social functioning (HoNOS) at t2 and t3 via social inclusion (SIS), hope (HOPE) and empowerment (ES) at t1 and t2. To estimate indirect effects via multiple parallel mediators, we applied a SEM-based path-analytic approach tailored to repeated measures control group designs (Montoya and Hayes, Reference Montoya and Hayes2017; Valente and MacKinnon, Reference Valente and MacKinnon2017). This approach allows simultaneous estimation of all relevant cross-lagged paths, accounts for autoregressive stability and correlations among variables and adjusts for measurement error, thereby isolating intervention effects. Autoregressive effects capture the temporal stability of a variable (e.g., SIS at t1 predicting SIS at t2), while cross-lagged effects assess whether one variable at an earlier time point predicts changes in another variable later in time (e.g., SIS at t1 predicting STORI at t2). Indirect effects were calculated as the product of the paths from PS to the mediators and from the mediators to the outcomes (Preacher and Hayes, Reference Preacher and Hayes2008). Several CLPMs were fitted separately for STORI and HoNOS, with mediators added stepwise. SIS was entered first as it was the primary outcome in the UPSIDES RCT (Puschner et al., Reference Puschner, Nakku, Hiltensperger, Wolf, Ben-Dor, Bugeiga and Slade2026). Model fit was evaluated using standard indices (χ2, CFI, TLI, RMSEA, SRMR, AIC, BIC), and the best-fitting models were selected for interpretation. Significance of indirect effects was tested using nonparametric bootstrap resampling (5000 iterations) with bias-corrected percentile confidence intervals. This robust method accounts for non-normality and provides more accurate inference in mediation analyses than traditional approaches (Preacher and Hayes, Reference Preacher and Hayes2008; Zhao et al., Reference Zhao, Lynch and Chen2010).
Before estimating the CLPM, several preconditions were checked. Missing data were analysed and imputed if necessary. Multiple imputation was conducted separately for intervention and control groups. Continuous variables (e.g., SIS scores) were imputed via the Classification and Regression Trees method, while for the ordinal STORI stages, Predictive Mean Matching was used. To ensure robustness, 20 imputed datasets were generated separately for intervention and control group data. Diagnostic plots confirmed that the distributions of imputed values aligned with expected values and did not differ systematically. Baseline group differences were tested using Welch’s t-tests and Wilcoxon rank sum tests, with results indicating no significant differences between groups at t0. Normality checks using Shapiro-Wilk and Mardia’s test identified deviations from normality in multiple variables. Consequently, Weighted Least Squares Mean and Variance (WLSMV) estimation was used for the ordinal outcome STORI and Robust Maximum Likelihood (MLR) estimation for HoNOS. Given these estimators’ robust corrections, scaled fit indices were reported to provide a more accurate assessment of model fit. The χ2 test was interpreted with caution, as it is highly sensitive in large samples and may indicate significance even when model misfit is trivial (Shi et al., Reference Shi, Lee and Maydeu-olivares2019). Multi-collinearity was assessed using the Variance Inflation Factor (VIF), with all predictors exhibiting VIF values below 5, indicating no relevant collinearity concerns. Significant autoregressive paths in the models confirmed the assumption of temporal stability. These results confirmed the suitability of conducting CLPM analysis.
IBM SPSS 29 was used for data management and sociodemographic descriptive statistics, while R (version 4.4.1) was employed for all other statistical analyses.
Results
The study included 565 participants with severe mental health conditions. Sociodemographic characteristics, including age, gender, study site, TAG score and type of mental health condition, are provided in Table 1. Descriptive statistics for mediators and outcomes at each time point for both intervention and control groups are provided in Supplement 2. A full flowchart is provided in Supplement 3.
Sociodemographic characteristics and distribution of study participants
Table 1 Long description
Participant characteristics are compared between an intervention group of 255 and a control group of 310 persons, including age, gender, study site, TAG score, and mental health condition type. Mean age was nearly identical at about 38.6 years in the intervention group and 38.5 years in the control group, with similar spread. Gender distribution was comparable between groups, with females 57.3 percent versus 51.6 percent, males 42.4 percent versus 48.1 percent, and diverse about 0.4 percent versus 0.3 percent. Study sites were similarly represented across both groups, with each site contributing roughly 12 to 24 percent; Butabika, Uganda was the largest site in both groups at about 24 percent and 22 percent. Mean TAG score was slightly higher in the intervention group at 9.16 compared with 8.83 in the control group, with similar variability. Mental health condition categories were broadly similar between groups, with psychotic, bipolar, depressive, and other conditions each contributing under about 11 percent per category as reported. Percentages may not sum to 100 within sections due to rounding or reporting conventions.
M = mean, SD = standard deviation, TAG = Threshold Assessment Grid.
The fit indices of CLPMs, where PS served as the predictor, and STORI and HoNOS were the outcomes, are shown in Table 2. Comparison of the model fits based on scaled CFI, TLI, RMSEA and SRMR revealed that Models 1B and 2B with social inclusion and hope as mediators overall have the best fit and therefore were used for further analyses. Models including empowerment (1C and 2C) did not show improved fit (AIC/BIC) and were therefore not considered further.
Scaled model fit indices
Table 2 Long description
Scaled fit indices are reported for six structural models linking SIS alone or with HOPE and ES (measured at times 1 and 2) to outcomes at later times (STORI or HoNOS). For STORI outcomes, Model 1A (SIS only) shows the strongest overall fit: chi-square 9.106 with 9 degrees of freedom and p-value 0.100, CFI 0.989, TLI 0.981, RMSEA 0.033, and SRMR 0.023. Adding mediators HOPE and ES for STORI (Models 1B and 1C) keeps CFI and TLI high and RMSEA and SRMR low, but chi-square p-values are reported as 0.000, indicating statistically significant misfit by that test. For HoNOS outcomes, all models have chi-square p-values reported as .000 and RMSEA values above typical “good fit” thresholds (about 0.076 to 0.092), while SRMR remains low to moderate (.024 to 0.056). Across HoNOS models, CFI stays high (.963 to 0.976) but TLI is notably lower (.817 to 0.864), suggesting weaker fit by that index. AIC and BIC are not available for the STORI models, while for HoNOS they increase from Model 2A to 2C, which does not support improved fit with added mediators under those criteria.
SIS = Social Integration Scale; HOPE = Hope Scale; ES = Empowerment Scale; STORI = Stages of Recovery Instrument; HoNOS = Health of the Nation Outcome Scales; df = degrees of freedom; CFI = Comparative Fit Index; TLI = Tucker–Lewis Index; RMSEA = Root Mean Square Error of Approximation; SRMR = Standardised Root Mean Square Residual. Model fit criteria: Good fit: CFI/TLI ≥ 0.95, RMSEA ≤ 0.05, SRMR ≤ 0.06. Acceptable fit: CFI/TLI ≥ 0.90, RMSEA ≤ 0.08, SRMR ≤ 0.08. N = 565. – indicates that AIC and BIC are not reported because the model was estimated using the WLSMV estimator, which does not compute a log-likelihood function.
Model 1B (see Fig. 1) demonstrated significant autoregressive effects for SIS (βt0-t1 = 0.564, βt1-t2 = 0.667, P < 0.001) and HOPE (βt0-t1 = 0.589, βt1-t2 = 0.691, P < 0.001). The influence of earlier recovery stages on subsequent STORI stages was also significant (βt0-t2 = 0.118, βt2-t3 = 0.314, P < 0.05). Regarding the predictor PS, allocation had significant effects on SIS and HOPE at T1 (SIS: β = 0.283, P = 0.001; HOPE: β = 0.288, P < 0.001). Path coefficients between allocation and the mediators at T2 were not significant (SIS: β = −0.037, P = 0.559; HOPE: β = 0.072, P = 0.287). Similarly, the direct effect of allocation on STORI was non-significant at all time points (STORI at T2: β = −0.025, P = 0.800; STORI at T3: β = −0.047, P = 0.649). Cross-lagged effects revealed that SIS and HOPE at T1 significantly predicted STORI at T2. SIS showed a stronger predictive value (β = 0.378, P < 0.001) compared to HOPE (β = 0.145, P = 0.043). The path from SIS at t2 to STORI at t3 was non-significant (β = 0.109, P = 0.094), while the path from HOPE at t2 to STORI at t3 reached significance (β = 0.159, P = 0.007). Mediation analysis showed significant indirect effects of PS on STORI at t2 via the mediators SIS and HOPE at t1. The indirect effect through SIS was β = 0.114, with a 95% confidence interval (CI) [0.049, 0.194]. The indirect effect through HOPE was β = 0.037, 95% CI [0.001, 0.086]. The total mediation effect was c = 0.151.
Cross-lagged panel model 1B showing estimated paths from peer support (PS) to recovery (STORI) at t2 and t3, mediated by social inclusion (SIS) and hope (HOPE) at t1 and t2.
Figure 1 Long description
Cross-lagged panel model with four time points labeled t0, t1, t2 and t3. At t0, boxes labeled Social Inclusion and Hope are connected by dashed arrows indicating autoregressive effects and covariances. Peer Support is positioned centrally, with arrows pointing towards Social Inclusion and Hope at t1, showing significant path coefficients of beta equals 0.283 and beta equals 0.288 respectively, both significant at p less than 0.001. At t1, Social Inclusion and Hope are connected to Recovery through significant paths, with coefficients beta equals 0.378 and beta equals 0.145, respectively. The path from Social Inclusion to Recovery at t2 is non-significant, beta equals 0.109, while Hope to Recovery at t3 is significant, beta equals 0.159. Peer Support has non-significant paths to Recovery at t2 and t3, with coefficients beta equals minus 0.025 and beta equals minus 0.047. Error terms, denoted by ′e′, are attached to Social Inclusion, Hope and Recovery at each time point. The diagram includes a legend explaining the symbols: solid arrows for significant unidirectional paths, dashed arrows for autoregressive effects and covariances and dotted arrows for non-significant paths. Path coefficients are marked with significance levels: three asterisks for p less than 0.001, two for p less than 0.01, one for p less than 0.05 and ′n.s.′ for non-significant paths.
Model 2B (see Fig. 2) yielded significant autoregressive paths for all variables (SIS: β t0–t1 = 0.458, β t1-t2 = 0.535; HOPE: βt0-t1 = 0.522, βt1-t2 = 0.540; HoNOS: βt0-t2 = 0.264, βt2-t3 = 0.551, P < 0.001). Allocation to PS had significant effects on SIS (β = 0.282, P < 0.001) and HOPE (β = 0.254, P = 0.001) at T1. No significant effects were observed for SIS (β = 0.003, P = 0.959) or HOPE (β = 0.110, P = 0.087) at T2. There were no significant direct effects of PS on HoNOS at T2 (β = −0.018, P = 0.756) or T3 (β = 0.003, P = 0.960). Cross-lagged paths indicated that HoNOS at T2 was negatively predicted by SIS (β = −0.124, P < 0.001) and HOPE (β = −0.102, P = 0.007) at T1. However, HoNOS at T3 was not significantly influenced by SIS (β = −0.045, P = 0.227) or HOPE (β = −0.068, P = 0.057) at T2.
Cross-lagged panel model 2B showing estimated paths from peer support to health and social functioning at t2 and t3, mediated by social inclusion and hope at t1 and t2.
Figure 2 Long description
Cross-lagged panel model illustrating the estimated paths from peer support to health and social functioning across four time points: t0, t1, t2 and t3. At t0, social inclusion, hope and health and social functioning are connected by autoregressive effects and covariances. At t1, peer support significantly affects social inclusion (β = 0.282, p <.001) and hope (β = 0.254, p <.001). Social inclusion negatively predicts health and social functioning (β = -0.124, p <.001) and hope negatively predicts it (β = -0.102, p <.01). At t2, social inclusion and hope have non-significant effects on health and social functioning. At t3, health and social functioning is not significantly influenced by social inclusion (β = -0.045, n.s.) or hope (β = -0.068, n.s.). The model includes error terms and path coefficients, with significant paths marked by solid arrows and non-significant paths by dashed arrows. Autoregressive effects and covariances are indicated by curved arrows. Path coefficients are annotated with significance levels: p <.001, p <.01, p <.05, n.s. p >.05.
The indirect effect of allocation on HoNOS via SIS was β = −0.035, 95% CI [−0.064, −0.013]; via HOPE, β = −0.026, 95% CI [–0.052, −0.006]; total mediation effect was c = −0.060.
Discussion
The results indicate that social inclusion and hope mediate the relationship between PS and both recovery and health and social functioning, highlighting their role as key mechanisms of action in PS interventions. Both models revealed strong autoregressive effects, which increased post-intervention, suggesting substantial temporal stability for both mediators and outcomes. This may imply that the effects of PS could be long-lasting, with mediators and outcomes maintaining stability over time. PS allocation significantly increased social inclusion and hope at t1 in both models, but the effects at t2 were non-significant, suggesting that the primary impact of PS on these mediators occurred early in the intervention. Mediation analysis revealed significant indirect effects of PS on recovery and health and social functioning via social inclusion and hope at t1. No direct effects of PS on recovery or health and social functioning were observed at any time point, further highlighting the importance of indirect pathways. The effect sizes identified in this study can commonly be interpreted as small to medium. However, such effect sizes are frequently observed in mediation models, particularly when these models involve complex relationships with multiple mediators (Preacher and Hayes, Reference Preacher and Hayes2008). In the context of this complex psychosocial intervention and longitudinal CLPM design, substantial autoregressive stability and several concurrently unfolding processes mean that relatively little variance remains for indirect pathways to explain.
Mechanisms of recovery
Our findings are consistent with other mediation studies of peer-delivered psychosocial interventions. For example, in a trial of the Thinking Healthy Programme for perinatal depression in India and Pakistan (Singla et al., Reference Singla, Kohrt, Murray, Anand, Chorpita and Patel2017), effects on depressive symptoms were mediated by increased patient activation and social support. Similarly, in a study of older adults receiving PS (Joo et al., Reference Joo, Xie, Choi, Gallo, Locascio, Ma, Mace, Solomon and Khemraj2025), reductions in loneliness and gains in self-efficacy and adaptive coping mediated improvements. It should be noted that the SIS used in this study captures both structural and perceived aspects of inclusion. Structural indicators of participation include social contact and engagement in community activities (items 1, 3, 5, 11, 12), whereas perceived inclusion reflects acceptance, safety, rights awareness and a sense of contribution (items 6, 7, 8, 13, 14, 15, 16). The scale also contains items assessing social exclusion and stigma (items 2, 9, 10). This blend of behavioural and subjective indicators is relevant for interpreting the mediation findings, as PS may influence opportunities for participation and experiences of belonging through different mechanisms.
While our findings pertain to individuals with severe mental health conditions, a recent meta-analysis of digital PS in non-clinical populations also identified hope, connectedness and emotional well-being as likely mechanisms (Yeo et al., Reference Yeo, Fortuna, Lansford and Rudolph2025). Although generalisability is limited, the convergence of findings across populations and delivery modes supports the broader relevance of psychosocial mechanisms such as hope and social inclusion. These mechanisms are central to the Change Model of Peer Support (Gillard et al., Reference Gillard, Gibson, Holley and Lucock2015) and were also observed in the UPSIDES trial: PS increased hope and social inclusion, without directly affecting recovery outcomes (Puschner et al., Reference Puschner, Nakku, Hiltensperger, Wolf, Ben-Dor, Bugeiga and Slade2026). Qualitative findings from UPSIDES (Krumm et al., Reference Krumm, Haun, Hiller, Charles, Kalha, Niwemuhwezi, Nixdorf, Puschner, Ryan, Shamba, Epstein and Moran2022; Haun et al., Reference Haun, Adler Ben-Dor, Hall, Kalha, Korde, Moran, Müller-Stierlin, Niwemuhwezi, Nixdorf, Puschner, Ramesh, Charles and Krumm2024) further support these results, highlighting hope and inclusion as perceived drivers of recovery. This body of evidence helps explain why PS improves psychosocial outcomes even when effects on symptoms or service use remain inconsistent (Cabassa et al., Reference Cabassa, Camacho, Vélez-grau and Stefancic2017; White et al., Reference White, Foster, Marks, Morshead, Goldsmith, Barlow, Sin and Gillard2020). Some reviews report improvements in self-reported recovery (Yim et al., Reference Yim, Chieng, Tang, Tan, Kwok and Tan2023; Cooper et al., Reference Cooper, Saunders, Greenburgh, Shah, Appleton, Machin, Jeynes, Barnett, Allan, Griffiths, Stuart, Mitchell, Chipp, Jeffreys, Lloyd-Evans, Simpson and Johnson2024), while others do not. Our results suggest that PS may primarily operate through indirect psychosocial pathways rather than direct symptom reduction. Interventions that explicitly target hope and social inclusion may therefore be more effective in fostering recovery and functional gains, especially in the absence of measurable symptom change. As CLPMs estimate within-person processes, the mechanisms identified here reflect population-average dynamics and cannot vary by time-invariant characteristics such as gender or diagnosis. This supports the general importance of strengthening hope and social inclusion across the overall population of service users. However, evidence indicates that social network resources and the capacity to benefit from them can differ across groups (McKenzie et al., Reference McKenzie, Collings, Jenkin and River2018; Brooks et al., Reference Brooks, Devereux-Fitzgerald, Richmond, Bee, Lovell, Caton, Cherry, Edwards, Downs, Bush, Vassilev, Young and Rogers2022). These differences highlight the value of future work using moderated mediation or multi-group approaches to clarify whether these pathways operate differently across subgroups.
Beyond the UPSIDES model, our findings are consistent with broader recovery theory. The CHIME framework (Leamy et al., Reference Leamy, Bird, Le Boutillier, Williams and Slade2011) identifies connectedness and hope as core elements of personal recovery, consistent with the mechanisms observed here. Social capital theory provides an additional perspective on this interplay (Bourdieu, Reference Bourdieu1986; Putnam et al., Reference Putnam, Nanetti and Leonardi1993; McKenzie et al., Reference McKenzie, Whitley and Weich2002). The distinction between bonding and bridging social capital, and their structural and cognitive components, remains central for understanding how social connections shape mental health (Almedom, Reference Almedom2005). PS fosters bonding social capital through shared experience, trust and reciprocity, and creates opportunities for bridging social capital by supporting participation, community engagement and access to resources. The SIS used in our study captures several of these dimensions. This alignment suggests that the indirect effects in our models may reflect broader processes of social capital formation that contribute to personal recovery and health and social functioning.
Implications
Our findings underscore the importance of fostering hope and social inclusion as central mechanisms through which PS interventions influence recovery outcomes. Understanding these indirect pathways is crucial because it clarifies how PS works in practice, helps refine intervention content and informs implementation strategies by identifying which elements require protection, strengthening or targeted delivery. These elements also distinguish PS from other professional interventions in mental health, as they are rooted in the lived experiences of individuals who have faced similar challenges. Unlike traditional mental health professionals, who focus on clinical expertise, PS interventions leverage personal narratives and shared understanding to foster hope and inclusion (Davidson et al., Reference Davidson, Chinman, Sells and Rowe2006). This distinction is also critical for informing clinical referrals, as it suggests that individuals experiencing lower hope and greater social exclusion should be prioritised for PS interventions. A key implication is that mental health service providers can use hope and social inclusion levels as indicators to guide referrals to PS services. This targeted approach may be particularly valuable in low-resource settings, where mental health workers, including PSWs, are limited. Prioritising service users with low levels of hope and social inclusion could improve the cost-effectiveness of PS delivery by focusing resources on those most likely to benefit.
Our findings indicate that hope and social inclusion are core mechanisms through which PS exerts its impact, underscoring that the effectiveness of PS depends on relational processes rather than symptom-focused techniques. This highlights the importance of maintaining role integrity to preserve PS’s unique relational dynamic (Stratford et al., Reference Stratford, Halpin, Phillips, Skerritt, Beales, Cheng, Hammond, O’Hagan, Loreto, Tiengtom, Kobe, Harrington, Fisher and Davidson2017). Unlike traditional mental health professionals, PS workers draw on personal experience, reciprocity and shared understanding to create connection and thereby strengthen social inclusion and foster hope. These relational elements are in turn essential to the recovery process, as PS reduces isolation – a common challenge for individuals with severe mental health conditions (Koenders et al., Reference Koenders, de Mooij, Dekker and Kikkert2017). Improvements in social communication and emotional support, which foster inclusion and reduce isolation, are also key outcomes of PS interventions (Fan et al., Reference Fan, Ma, Ma, Xu, Steven Lamberti and Caine2018; King and Simmons, Reference King and Simmons2018; de Wet et al., Reference de Wet, Sunkel and Pretorius2022). These aspects of PS align with the Change Model of Peer Support (Gillard et al., Reference Gillard, Gibson, Holley and Lucock2015), which emphasises the significance of social engagement and network strengthening as crucial pathways to improved recovery outcomes. In this sense, PS goes beyond traditional therapeutic practices, providing a space where individuals can truly relate to one another and offer a unique type of support rooted in shared lived experience.
Given the pivotal role of hope and social inclusion, training and intervention delivery should prioritise these elements. This means equipping PSWs with the skills to actively foster hope and promote social connections, while also being mindful of the diverse cultural and contextual factors of the setting they are working in. Tailoring training to the cultural nuances and mental health systems of different regions is essential to ensure PS remains effective in promoting social inclusion and hope across diverse populations (Nixdorf et al., Reference Nixdorf, Kotera, Baillie, Garber Epstein, Hall, Hiltensperger, Korde, Moran, Mpango, Nakku, Puschner, Ramesh, Repper, Shamba, Slade, Kalha and Mahlke2024), for example by incorporating culturally appropriate communication and interpersonal skills. Another foundational aspect of training should be the emphasis on the lived experience of PS workers as a key asset in their role (Charles et al., Reference Charles, Nixdorf, Ibrahim, Meir, Mpango, Ngakongwa, Nudds, Pathare, Ryan, Repper, Wharrad, Wolf, Slade and Mahlke2021). By leveraging their own recovery stories, PS workers can bring authenticity and empathy to their interactions with service users. The ability of PS workers to maintain their authentic, lived experience-based approach should be central to training and delivery (Kotera et al., Reference Kotera, Newby, Charles, Ng, Watson, Davidson, Nixdorf, Bradstreet, Brophy, Brasier, Simpson, Gillard, Puschner, Kidd, Mahlke, Sutton, Gray, Smith, Ashmore, Pomberth and Slade2025). Therefore, training should not only provide core recovery-focused knowledge and communication skills but should also offer practical exercises such as role-playing to help PS workers apply their learning comfortably and authentically in real-world settings (Nixdorf et al., Reference Nixdorf, Nugent, Aslam, Barber, Charles, Gai, Grayzman, Hiltensperger, Kalha, Korde, Mtei, Niwemuhwezi, Rameshh, Ryanb, Sladed, Wenzel and Mahlke2022). Further, unlike most mental health professionals, PSWs are often uniquely positioned to accompany service users in community and social activities. This flexibility enables PSWs to promote social inclusion not only through emotional support, but also through direct engagement in everyday life. Ensuring that PS services are adequately resourced to allow for such outreach – e.g., accompanying service users to social events or community spaces – could significantly enhance their impact.
Although UPSIDES training and delivery were locally adapted across all sites, the meaning and salience of social inclusion and hope may nevertheless vary between cultural and structural contexts. This underscores the importance of interpreting these mechanisms within the social environments in which PS is delivered.
Building on these findings, future research should examine the contextual factors that influence how well PS fosters hope and social inclusion. Future analyses within the UPSIDES process evaluation will assess the role of contextual variables and PS fidelity (Hiltensperger et al., Reference Hiltensperger, Kotera, Wolf, Nixdorf, Charles, Farkas, Grayzman, Kalha, Korde, Mahlke, Moran, Mpango, Mtei, Ryan, Shamba, Wenzel, Slade and Puschner2024a) to clarify how these factors shape the effects of PS on recovery and functional outcomes. These analyses will be particularly important for refining implementation strategies and optimising referral decisions for PS. The UPSIDES Theory of Change (Hiltensperger et al., Reference Hiltensperger, Ryan, Ben-Dor, Charles, Epple, Kalha, Korde, Kotera, Mpango, Moran, Mueller-Stierlin, Nixdorf, Ramesh, Shamba, Slade, Puschner and Nakku2024b) highlights that PS operates within interconnected pathways at individual, organisational and societal levels. Understanding how such contextual influences strengthen or weaken the mediating roles of hope and social inclusion will help refine implementation strategies and optimise PS referrals. This ensures that PS effectively fosters recovery and functional improvements across diverse settings.
Strengths and limitations
Strengths of our study include the large multi-national sample across countries in different income bands, the use of a multivariate repeated measures design and the quantitative evidence generated to supplement the existing qualitative evidence base. The inclusion of six culturally and structurally diverse settings is a particular strength, as it increases the external validity of our findings; however, it also introduces contextual heterogeneity that we could not model explicitly. Although randomisation was stratified by site and each site contributed comparable sample sizes, the overall sample was not sufficiently large to estimate hierarchical (multilevel) models that account for between-site differences or to test whether the identified mechanisms operate similarly across contexts. This limits our ability to draw conclusions about setting-specific effects. Further, the impact of the COVID-19 pandemic on PS and its potential effects was not considered as a confounding factor in this study. The pandemic disrupted the implementation of UPSIDES PS across settings, with particular challenges in low- and middle-income countries where technology barriers and service disruptions limited or even prevented the provision of PS altogether (Mpango et al., Reference Mpango, Kalha, Shamba, Ramesh, Ngakongwa, Kulkarni, Korde, Nakku and Ryan2020). In high-income countries (HICs), where digital tools and other adaptations made it possible for PS workers to at least stay in touch with their peers, the role of PS during the pandemic was more likely to be captured, although the full extent of its benefits may still have been constrained by the overall disruptions to services (Goldfarb et al., Reference Goldfarb, Grayzman, Meir, Grundman, Rabinian, Lachman, Epstein, Ben-Dor, Naaman, Puschner and Moran2024). Given these conditions, opportunities for in-person connection and community participation were likely reduced, potentially attenuating observable effects on social inclusion and hope during the trial period. These contextual factors may therefore have contributed to an underestimation of the true impact of PS.
Conclusion
This study provides empirical evidence identifying hope and social inclusion as primary mechanisms through which PS leads to recovery and functional improvement. The study emphasises the potential of the UPSIDES intervention in improving recovery and health and social functioning for individuals with severe mental health conditions. Importantly, our results offer practical guidance for mental health professionals and policymakers in determining which service users may benefit most from PS. Further, it provides valuable insights for refining intervention strategies, highlighting the importance of fostering hope and social inclusion to enhance the effectiveness of PS in diverse settings. Future research should explore how best to tailor PS referrals based on hope and social exclusion levels, ensuring that those most in need receive the greatest benefit.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S2045796026100900
Availability of data and materials
The data that support the findings of this study, including analytic code, will be available in the repository OPARU at https://oparu.uni-ulm.de/xmlui/ following an embargo until 31 December 2026, to allow for prioritised generation of research findings by members of the UPSIDES consortium.
Acknowledgements
The study Using Peer Support In Developing Empowering Mental Health Services (UPSIDES) is a multicentre collaboration between the Department for Psychiatry and Psychotherapy II at Ulm University, Germany (Bernd Puschner, coordinator); the Institute of Mental Health at University of Nottingham, UK (Mike Slade); the Department of Psychiatry at University Hospital Hamburg-Eppendorf, Germany (Candelaria Mahlke); Butabika National Referral Hospital, Uganda (Juliet Nakku); the Centre for Global Mental Health at London School of Hygiene and Tropical Medicine, UK (Grace Ryan); Ifakara Health Institute, Dar es Salaam, Tanzania (Donat Shamba); the Department of Social Work at Ben Gurion University of the Negev, Be’er Sheva, Israel (Galia Moran); and the Centre for Mental Health Law and Policy, Pune, India (Jasmine Kalha).
Author contributions
B.P., M.S., C.M., J.N., G.R., D.S., G.S.M. and J.K. conceived the trial and wrote the trial protocol. R.H., A.C., M.H., J.K., C.M., J.Na., J.Ni., R.N., M.R., G.R., M.S., D.S., P.W. and B.P. contributed to implementation in various roles (recruitment, training and delivery of the intervention, data collection, data check). R.H., with support from B.P., formulated the research questions for the mediation analysis and wrote the manuscript. R.H. analysed the data. All authors critically reviewed and approved the final manuscript.
Financial support
UPSIDES has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 779263. Mike Slade acknowledges the support of the NIHR Nottingham Biomedical Research Centre. This publication reflects only the authors’ view. The Commission is not responsible for any use that may be made of the information it contains.
Competing interests
None.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2000.
Open access
