The Effectiveness of Peer Support in Personal and Clinical Recovery: Systematic Review and Meta-Analysis

We conducted a comprehensive literature search in PubMed, PsycInfo, CINAHL, Cochrane Library, and Web of Science. The search strategy included terms related to or describing the intervention, the relevant patient groups, and methodological design. (The search terms are available in the online supplement to this article.) Articles written in English, Danish, Norwegian, or Swedish published since 2000 were included. The initial search was conducted in April 2018, and an updated search was performed in June 2019. In addition, reference lists of the included studies, as well as of reviews, were searched and authors of published randomized controlled trial (RCT) protocols were contacted by e-mail for any unpublished results. All citations retrieved through the search were screened independently by three authors (C.H.E., S.S.M., M.N.N.) in Covidence (24). The full texts were retrieved and independently assessed for eligibility (described below) by the same three authors. Disagreements were resolved through discussion with a fourth author (L.F.E.).

The included studies were approved by research ethics boards or committees of the study authors’ institutions and were performed in accordance with the ethical standards of the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.

All published and available unpublished RCTs and cluster RCTs were eligible for inclusion.

We excluded studies on individuals with intellectual disabilities, psychosomatic disorders, sexual offenders, or forensic psychiatry. We also excluded studies with a primary focus on general medical illness, substance use disorder, or prevention or on relatives of individuals with mental illness.

Risk for bias was assessed independently by four authors (C.H.E., S.S.M., L.H., M.N.N.) according to a modified Cochrane Collaboration tool (26) that included the following aspects: random sequence generation, allocation concealment, blinding of outcome assessment, incomplete outcome data, selective outcome reporting, power calculations, and intent to treat. Assessment of power calculation and intent to treat replaced the category “other” bias in the risk-of-bias tool. The item “blinding of participants and personnel” was not included in the assessment because of the nature of the peer support services, where the common understanding of lived experience is a vital component of peer models. The risk-for-bias assessment was converted to Agency for Healthcare Research and Quality (AHRQ) standards (27) to get an assessment of good quality (all criteria met low risk for bias for each domain), fair quality (one criterion not met, such as high risk for bias for one domain, or two criteria unclear and an assessment that this was unlikely to have biased the outcome, as well as absence of a known important limitation that could invalidate the results), and poor quality (two or more criteria categorized as high risk for bias or unclear risk for bias and an assessment that this was likely to have biased the outcome, as well as presence of important limitations that could invalidate the results).

Data were extracted by two authors (C.H.E., S.S.M.) for the meta-analysis and summary of findings. Extracted information included study methodology, setting, results, and population; participant demographic characteristics; details of the intervention and control conditions; outcomes and times of measurement; descriptions of peers; and categorization of the peer intervention. Means±SDs for relevant outcomes at the end of the intervention were extracted for the meta-analysis, and missing data were requested from study authors.

Comprehensive meta-analysis software (Biostat) was used for the statistical analysis. Pooled estimates at the end of an intervention were calculated as the standard mean difference (SMD) and 95% CIs in a random-effects meta-analysis design with inverse variance weights. The heterogeneity of study results was assessed through visual inspection of forest plots, the p value of the chi-square test, and the I² statistic (a descriptor of the variability in effect estimates due to heterogeneity rather than chance). A p<0.1 and I²>50% suggest substantial heterogeneity. Funnel plots were used to visually inspect the risk for publication bias. The statistical significance level was set at p<0.05. Effect sizes were interpreted according to research guidelines (28), and Cohen’s d effect sizes defined as small (Cohen’s d=0.2), medium (Cohen’s d=0.5), and large (Cohen’s d=0.8) were used as guidance (28, 29). We included all studies with applicable data in the meta-analysis, and studies with treatment-as-usual and active control groups were analyzed together. One unpublished study included more than one eligible peer support intervention with applicable data (Miller R., personal e-mail communication, 2018). The data from the two interventions were pooled for the meta-analysis. In post hoc exploratory meta-analyses, we analyzed studies with peer-led peer support compared with clinician-led peer support, as well as co-led, one-to-one, and group-based peer support interventions compared with treatment-as-usual or active control groups.

The search resulted in 12,161 records; one unpublished record was identified by contacting an author to request a study protocol, and 119 records were found through hand searching of reference lists. After removal of duplicates, the titles and abstracts of 11,789 records were screened. After the screening, the full texts of 155 studies were assessed for eligibility, and 59 studies were included, reporting on findings in 49 RCTs (a flow diagram of the selection process is available in the online supplement). Of these, 34 reported data that could be used in the meta-analysis. The authors of the remaining 14 trials were contacted and asked to provide data. However, none of the authors provided applicable data for the meta-analysis; hence, these trials were included only in the narrative part of the systematic review results (below).

Peer support interventions were compared with treatment as usual in 25 trials (30, 32, 36, 37, 39, 40, 43–45, 47–50, 53–57, 60–62, 66, 68–70, 76, 78) (Miller R., personal e-mail communication, 2018) and waitlist control conditions in nine trials (31, 33, 42, 51, 52, 63, 64, 67, 73, 74, 80–82); five trials (34, 41, 59, 75, 79) had active control conditions such as nonpeer group counseling, clinician-led psychiatric advanced directives, and cognitive-behavioral therapy. The remaining eight trials (38, 46, 58, 65, 77, 83–87) were three- or four-armed trials with treatment-as-usual and active control groups (see table in the online supplement).

The included RCTs were characterized by high or unclear risk for bias in many domains of the modified Cochrane risk-of-bias tool checklist (Figure 1). According to AHRQ standards (27), nine of the trials in the “peers added to hospital services” category (31, 34–37, 39, 42–45, 51) were of fair quality, and 18 trials (30, 32, 33, 38, 40, 41, 46–50, 52–57) (Miller R., personal e-mail communication, 2018) were of poor quality. In the category “peers independent of hospital services,” two trials (63, 64, 72–74) were of fair quality, and 11 trials were of poor quality (65–70, 75–79). In the “peers in established service roles” category, all three trials (59–62) were of poor quality. The “peers online” category consisted of one trial (83) that had no risk for bias in all domains and two trials of fair quality (80, 81, 84–86) and two trials of poor quality (82, 87).

Pooled effect sizes for personal and clinical recovery outcomes at the end of interventions are presented in the online supplement. If only one trial reported on an outcome, a pooled effect size was not applicable.

Detailed information on the heterogeneity of the studies included in the meta-analysis is presented in the online supplement. We included 34 trials in this meta-analysis investigating outcomes of personal and clinical recovery. Of the 47 analyses conducted, 20 had a high heterogeneity as indicated by I²>50% and p<0.1.

Funnel plots were created for visual inspection of publication bias in the included studies (see the online supplement). Few of the analyses included ≥ 10 studies. However, several of the funnel plots indicated a risk for publication bias.

Post hoc subgroup analysis was carried out to explore any potential differences in pooled effect sizes in one-to-one and group-based peer support, respectively. Additionally, post hoc analyses of co-led peer support versus treatment as usual and peer-led versus clinician-led peer support, as well as analyses excluding interventions without an explicit aim of personal recovery, were conducted (see table in the online supplement). No clear differences between one-to-one and group-based peer support in personal or clinical recovery outcomes were observed. Co-led peer support added to hospital services had statistically significant moderate effects on both personal and clinical recovery outcomes, compared with treatment as usual. Additionally, a significant moderate effect on social network and a small increase in depression symptoms was observed in peer-led versus clinician-led interventions. Last, post hoc analyses of interventions that explicitly aimed to improve personal recovery indicated that peers added to hospital services slightly decreased personal recovery compared with treatment-as-usual, waitlist, or active control groups.

The findings of this systematic review and meta-analysis, comprising 12,477 participants in 49 trials, provide evidence that peer support targeting individuals with any mental illness has small effects on outcomes related to personal recovery and little or no effect on clinical recovery—although we did find a small reducing effect on anxiety symptoms. We found that peers employed and trained in supporting roles in the mental health hospital treatment service had a small positive effect on personal recovery, had a small-to-moderate positive effect on self-efficacy, and were associated with a small reduction in overall psychiatric symptoms. Additionally, we found that peer support delivered in peer-designed or cocreated interventions developed independently of hospital settings and provided in community settings had a modest positive effect on self-advocacy. We had no analyzable data on peers employed in established service positions. We also found that peer support delivered online had a small but nonsignificant positive effect on personal recovery. These results indicate that peer support might have an impact on different aspects of personal recovery that reflect the core values of peer support provided in different organizational settings. However, more high-quality RCT studies coproduced by clinicians, researchers, and peer supporters, especially of peers working independently of hospital services or in online settings, are needed to further contribute to the literature on the efficacy of peer support in mental health care.

The findings that peer support has small positive effects on personal recovery and little or no effect on clinical recovery in poor-to-fair-quality trials with high heterogeneity are consistent with previous findings of reviews evaluating peer support for individuals with severe mental illness (15–17). The present review adds evidence that the positive effects of peer services on personal recovery and decreased psychiatric symptoms occur across different diagnostic categories of mental illness. Additionally, and contrary to our initial hypothesis, the effect on personal recovery was most evident in peer support delivered as and as an add-on to standard mental health treatment in hospital services. Still, our findings support observations in previous reviews that the evidence of peer support is weak given a lack of high-quality studies (17). Previous meta-analyses have shown small positive effects of peer support on outcomes such as personal recovery, hope, and empowerment (15–17). In the present meta-analysis, we excluded data on empowerment from one trial (30) because of very large inconsistencies in the data; this exclusion may partly explain why we did not confirm previous findings of a positive effect of peer support on empowerment. Moreover, differences in time of data extraction at the end of an intervention or at follow-up, along with differences in categorizing trials in different organizational settings across the studies in this review and previous reviews (15, 16), may also have had an impact on the comparison of findings across studies.

We found that the effect on improved personal recovery was most pronounced for peers added to hospital services compared with other peer categories. One explanation for this finding is that peers added to established mental health services is the most effective form of peer support. However, hospital services have a longer tradition of conducting RCTs, possibly explaining why this category included most of the effective trials of reasonable quality compared with the other categories. For instance, the authors of a recent review (20) concluded that online peer support interventions have potential but are in the early stages of development and require well-powered clinical trials, in line with our finding of a small and nonsignificant effect of online peer support on personal recovery. Moreover, it has been posited that recent trials indicating effectiveness of peer support in the established mental health services likely evaluated some form of peer-supported individual self-management rather than more mutual peer support. That is, individuals are supported in their recovery by peers who provide skills and confidence that enable service clients to manage their own health problems, rather than by an approach in which the recipient and the peer support provider mutually support and help each other (88). We also note that the focus in the field has shifted to primary outcomes related to individual empowerment and advocacy (18, 89), such as peer support facilitating a belief about the ability to exert control over one’s own life by learning to communicate and advocate for one’s own needs.

Peer-designed or cocreated interventions such as WRAP or BRIDGES were developed to be provided independently of mental health hospital care, and we found that these interventions have a modest positive effect on self-advocacy. Nevertheless, it has been argued that these peer support approaches belong to a traditional medical model of mental health, in which peer workers operate in complementary functions by enabling clients to cope better with mental health problems (22). We found no applicable data for peers in established service roles, in line with a recent review noting that most of the studies comparing peer workers with other mental health workers performing a similar role were published >10 years ago; none of them included data on personal and clinical recoveries (18), suggesting that employing peers in established clinical functions is outdated.

Because peer support is a complex intervention, we recommend that qualitative studies better capture the needs, active ingredients, and mechanisms of change experienced by peer support providers and recipients (8, 19, 21). Some qualitative studies have reported that identification with a role model and the building of trusting relationships based on shared experiences are key mechanisms to increase empowerment, self-efficacy, and social function (90). Additionally, future interventions should be transparent about the type of peer support approach used. These approaches may include trauma-informed interventions with a focus on what has happened in a client’s “whole life,” rather than illness-focused approaches centered on symptoms and disabilities (91); intentional peer support with a focus on equal power relationships and reciprocal roles of helping and learning with a focus on community rather than individual change alone (92); and a peer-supported individual self-management approach (34).

We note that it is important to state whether peers are employed as paid staff or volunteers, because a peer’s role and position may have an impact on the mechanisms of the intervention. Accordingly, to secure the active ingredients of peer support, we recommend that future interventions should be developed and evaluated via a cocreation approach that involves peers on an equal footing with researchers, taking advantage of peers’ lived experiences (9, 19, 22). This tactic should include the development of a program theory based on a clear description of the theory of change and a hypothesis about proposed mechanisms of any effects (18, 19, 93); this knowledge could inform high-quality RCTs and the development of fidelity measures, along with both quasi-experimental and process evaluations. Moreover, it has been noted that the journey of personal recovery is gradual (94), emphasizing why longitudinal intervention measurement is recommended. Furthermore, we recommend that identification of factors that hinder or promote implementation should be prioritized (8), and we also note that cost-effectiveness studies of peer support interventions are needed (18, 19).

Results of post hoc explorative analyses revealed that interventions that were co-led by a mental health professional and a peer provider had significant positive effects on measures of both personal recovery and clinical recovery (see tables in the online supplement). These findings were in agreement with results from a meta-analysis by Thomas et al. (95), showing that co-led interventions yielded the greatest effects on recovery compared with control groups. These findings suggest that partnership and collaboration between mental health professionals and peers enhance recovery. A qualitative review has indicated that clinicians working with peer support workers gained increased belief that recovery from mental illness is possible (96). In additional post hoc analyses, we confirmed recent findings that one-to-one peer support added to hospital services has a small but significant impact on personal recovery (18). Moreover, we confirmed findings that group-based peer support has a modest impact on personal recovery (19), suggesting that this impact may be independent of whether peer support is delivered in an individual or group format. Last, we showed that exclusion of interventions without an explicit aim of personal recovery, such as the HARP program (57) and Living Well (32), slightly decreases effects on personal recovery, indicating that although the programs did not explicitly aim to aid personal recovery, they still improved recovery. Nevertheless, further high-quality studies with a focus on peer support delivery and type of peer support in different organizational settings are needed to confirm these explorative findings.

The strengths of this study included the distinction between personal and clinical recovery outcomes and division into different types of peer support. We included a wide spectrum of peer support interventions, allowing for subgroup analyses of clinician-led versus peer-led, co-led, one-to-one, and group-based peer support. The first author of this review has lived experience of mental illness and as a peer-group facilitator, which was a strength in the operationalization of peer support, as well as in the interpretation of the implications of findings. The findings are generalizable to adult mental health hospital practice, independent community settings, and online platforms, primarily across samples in the United States but also in Europe. Nevertheless, the generalizability of the findings is primarily restricted to populations from Western countries.

Limitations included the poor quality of several of the reviewed studies, a generally high heterogeneity across the studies, and inadequate descriptions of intervention settings. Additionally, the definition of peer support is very challenging because of differences in health care system delivery and organization across countries, as well as a lack of transparency about intervention content, peer recruitment, training, and supervision. Especially the category “peers independent of hospital services” contained very disparate types of interventions where the only common feature was that they are peer-facilitated interventions or services in nonhospital organizations. The results of the meta-analyses of this category should therefore be interpreted with great caution, and the nonsignificant effect on personal recovery could be explained by the differences between the interventions compared. However, all interventions had an overall aim of improving personal recovery and were therefore pooled in the meta-analyses, which underlines the complexity of the field exploring the effectiveness of the peer element across a broad variety of interventions. None of the studies included measures of attrition among the peer support providers or potential worsening adverse effects among clients.

Several of the funnel plots indicated a risk for publication bias in the studies examined. Risk for bias included lack of transparency in randomization, insufficient blinding of outcome assessors, incomplete outcome data, and selective reporting, increasing the risk for selection and detection bias. Additionally, all trials had high risk for bias for the item “blinding of participants and personnel” because of the nature of the peer support intervention, increasing the risk for expectancy and performance bias. Nevertheless, we do not expect peer service research to overcome this limitation, because the building of safe relationships based on exchange of lived experiences is a vital component of peer models. Other limitations included the fact that studies with treatment-as-usual and active control groups were analyzed together, heterogeneity in study procedures such as group and individual format, and varying duration of peer support services. We addressed some of these limitations in post hoc subgroup analyses. Further post hoc analyses of the content and the length of intervention, manualized versus nonmanualized delivery, and volunteer versus paid peers were considered but were not possible because of inadequate descriptions of the interventions. Also, this meta-analysis evaluated the addition of peer support to highly varied standard services. Finally, we conducted several separate meta-analyses in this review, increasing the risk for type 1 error.