Where Does Value Lie in Peer Support? An Exploratory Discussion of the Theories and Methods Underpinning Effective Research in Peer Support

Search Terms

A search of health-related databases Medline, CINAHL, PsycINFO, PubMed, and Cochrane Library using relevant terms and key words was undertaken. ¹

In addition, a hand search was conducted to identify qualitative research into peer support reports and discussion papers. This was done through ResearchGate, Google Scholar, and sites such as NICE UK, U.S.-based Peers for Progress, and the Canadian Agency for Drugs and Technologies in Health (CADTH).

Inclusion Criteria

The strategy was to include articles on randomized controlled trials (RCTs) and systematic reviews (SRs) from January 2011 to July 2018 in English. Specifically, articles covering research projects undertaken with people who attended peer support groups for chronic and complex conditions were sought. Peer support delivered through social media and other technologies (mobile phone apps, interactive online programs) were included. The initial search delivered 284 peer-reviewed articles.

Selection Strategy

After removal of duplicate results, all abstracts were read independently by C.W. and a research assistant. Articles on single or small projects, those only marginally concerned with peer support issues or no evidence on effectiveness, such as protocol papers were excluded. The result was 34 articles to review in full. Eighteen of these were RCTs; additionally, there was one rapid review; a qualitative study, and two reports. The remaining were 11 SRs.

Another facet of the 2019 project was to explore other methods of evaluating peer support effectiveness in long-term behavior change that do not rely solely on clinical outcomes such as HbA1c or weight loss. The above search strategy did not find these alternative methods of evaluation so following reflection the reviewers targeted Patient Activation Measures (PAM) to test their potential to evaluate the effectiveness of peer support in changing health behaviors. Methods involved in longitudinal studies were also considered since effectiveness of peer support may be better gauged over longer timeframes. Finally, Grading of Recommendations, Assessment, Development and Evaluations (GRADE) was used to assist in estimating the effectiveness of RCTs.

Analysis

Complete published copies of those articles meeting the inclusion criteria were obtained and read with key attributes summarized. Reviewers (C.W. and a research assistant) applied GRADE to the RCTs estimating the levels of evidence including bias, limitation in research design and reporting.

We compared the results of this literature review with those of the 2011 literature review to estimate if more recent research showed methodological improvement.

Implications for Researching Effective Behavior Change

Research of this nature is generally undertaken to demonstrate effectiveness of a support program to replicate it in the broader community. Accordingly, sample sizes are important to detect differences between control and intervention groups, but this aspect was rarely discussed and numbers appeared to relate to availability. They varied from the lowest of 30 participants divided between intervention and control groups and the highest of 400 divided between intervention and control.

Studies reported general differences between the intervention and control groups, though these differences could have been associated with influences other than peer support. Only five studies reported blinding of researchers involved in data collection and analysis. Few RCTs had numbers calculated, possibly relying on self-selection. Small sample sizes can restrict the power of the study: it can also affect the variability and therefore reliability of results.

Similarly, timeframes for the delivery of a program, or dosage required to show an effect varied from 4 weeks to 12 months. The time required for effectiveness was rarely discussed. It is likely that time spent in delivering an intervention was dependent on funding. In a few cases there was follow-up at 6 or 12 months. Outcome measures were often clinical such as HbA1c, but also treatment adherence, well-being, and readmissions or ED use. The value of such measures is undermined, when there is inadequate baseline data collected on each person in the trial regarding stage of disease, social supports, and activity levels (Krauss, 2018). Of the17 RCTs, six reported significant changes though changes were not always evenly distributed across all measures. Seven other RCTs reported no or marginally significant improvements with several more reporting that both intervention and control arms improved. Only one (Houlihan et al., 2017) had a design demonstrating outcomes that could be attributed to peer support.

Peer support was delivered either in face-to-face groups or via internet platforms. Peer supporters were generally volunteers. Training of peer supporters to deliver a program is integral to a peer support program and an important variable to be detailed in research. The curriculum and delivery methods were not reported in five RCTs and in three other RCTs were mentioned but not described.

Problems with RCT design are compounded when we consider the shortcomings in design of RCTs in peer support including ineffective blinding, small numbers, and short timeframes. The extent of methodological problems is demonstrated when only one RCT (Houlihan et al., 2017) had a design that effectively demonstrated outcomes that could be attributed to peer support.

Articles lacked sufficient detail to assess if the length of time, the training, or other factors across the RCTs influenced the outcomes. There were no reports of evaluations of the contributions of training to outcomes of RCTs, suggesting that its role as an influencing variable was undervalued. As well, the impact of the training on support group participants, including their views on its value was not evaluated.

More cost-effectiveness studies have been undertaken since 2011 but much of the evidence relies on inference rather than modeling and measures such as Disability – Adjusted Life Years (DALYs). Burton et al. (2018) found six studies mentioning cost-effectiveness, only one of which provided evidence that peer-led programs were as effective as professionally led programs. CADTH (2013) found only one cost-effectiveness study but it did not include all risk factors related to Type 2 diabetes.

The systematic reviews made greater attempts at defining their terms. Haines et al. (2018) cited Pfeiffer et al. (2011), defining peer support as “bringing together non-professionals with similar stressors or health problems for mutual support or unidirectional support from an experienced to novice peer.” Mead et al. (2001) defined peer support as a “System of giving and receiving help founded on key principles of respect, shared responsibility and mutual agreement of what is helpful . . .” (p. 135). The importance of definitions is demonstrated as the first definition limits peer support to nonprofessionals with the same or similar experiences of health conditions, while the second suggests that health professionals may be included.

The 11 systematic reviews pointed to methodological problems or poor evidence. There are recommendations for better designed RCTs such as better blinding or more adequate powering (Stubbs et al., 2016), and fuller reporting of results. A number report the need for better outcome measures (Shilling et al., 2013), or a need to focus on what peer support does best (Barker & Maguire, 2017). Some authors speculated on the disappointing results and suggested that there were other aspects of peer support that might be more rewarding to investigate, for example, who most benefits from peer support (Fuhr et al., 2014), why it is popular with people with chronic illnesses, and which aspects are popular (DeMello et al., 2018; Small et al., 2013), and concentrating on overall improvement in well-being, possibly including improved health literacy, reduced isolation, increased personal control rather than clinical physical outcomes. Shilling et al. (2013, p. 608) in their SR articulate the need for theory to relate to outcome measures, timeframes, and comparators to strengthen evidence. They consider measures of health outcomes should be consistent with theories of peer support as an intervention; that the timeframes chosen to deliver per support should be appropriate to show health outcomes and greater attention paid to choosing appropriate control groups. With regard to control groups, they suggest it is not ethical to prevent people from seeking support.

Improving Research Design

Some of the inadequacies of the research may relate to poor design at the outset. The lack of definitions and theoretical underpinning of methodologies contribute to this inadequacy.

Definitions are important because they name the essence of the thing to be researched. Without this, it leaves others to make their own inferences. In peer support, there is no one accepted definition; its meaning may vary from one researcher to another. Some of this problem is exemplified in the RCTs. Thom et al. (2013) distinguish between peer support, peer health coaches, and peer educators, arguing that peer health coaches are more flexible than peer educators, though these differences were not explained. Barker and Maguire (2017) defined peer support as “experts by experience” though experience is not specified. While some definitions assume that peer support is integral to personal experience of illness and recovery, others relate it to observable outcomes or applications of training and education. Commencing with a defined account of what peer support is, will assist in moving toward some internal consistency and establish better what is being researched. In addition, it assists developing hypotheses and theories.

A theory of peer support makes clear how and by what mechanisms peer support may benefit the participants. This may then be applied as a working hypothesis in the different situations where peer support is employed. While some articles referred to theories most often, these were only nominal and were not integrated in the research design. There was little evidence of theory driving the research design and providing the framework (Noar & Zimmerman, 2005). Often, theories of peer support relate to individual behavior change, though the focus on clinical indicators as measures of changes in behaviors in both physical and mental health conditions may not fully represent this form of change. No article provided a detailed, theoretically derived description of peer support, hypothesized methods of effect, or the relationship between delivery, outcomes, and the theoretical basis. Similarly, Davis et al. (2015) consider the evidence regarding the abilities of programs to change health behaviors is mixed and may be due to poor application of theory or employment of an inappropriate theory. They suggest theories including social and environmental factors should be considered when research involves health behaviors.

The design of training is best related to accompanying definitions and theories. The importance of training those delivering peer support was demonstrated in the development of resources such as a training manual, didactic sessions to ensure information was consistently presented, communication skills and role plays, lack of reporting means. However it is not possible to evaluate if training was consistent with any definitions or theories. However, Cane et al. (2012) argue that there “is some evidence that behavior change interventions informed by theory are more effective than those that are not” (p. 7); they maintain that in reality, designs of such training are more likely to be based on “common-sense models of behavior.” Self-management programs offer an insight into the value of evaluations. Kennedy et al. (2014) undertook a process evaluation of a self-management program in U.K. primary care. This evaluation used a multimethod approach in line with the U.K. Medical Research Council recommendations to explore what worked and did not work for organizations, the staff, and patients. Normalization Process Theory (NPT) was used as the basis of design and the results of the evaluation were included in the results of the main trial.

Critiques of the RCT

RCTs, as they are now employed in peer support research, are essentially reductionist, in this case reducing peer support, to predetermined component parts (Hawe et al., 2004). Greene and Loscalzo (2017) explored the limits of reductionism regarding the complex interrelationships between genetics and environments, arguing for a more nuanced “network” approach. Deaton and Cartwright (2017) argue that RCTs in economic research have advantages in convincing skeptical audiences that an intervention works but will only advance scientific progress as a part of a whole cumulative process that includes understanding why it works. What works with RCTs needs to be couched in the circumstances it works under and describe what these may be. Boon et al. (2007) made a similar argument with regard to understanding complementary medicines in health care.

Even earlier, Pawson and Tilley (1997) were skeptical of the use of RCTs to research complex interventions involving human behavior. They argued for “realist” evaluations, which avoid assumptions of direct causal links between interventions and outcomes and argue for examining the contexts of “what works for whom and when.” Thus, they argue for exploring what works for whom and in what contexts rather than simply determining that something does work. Hawe et al. (2004) suggest that the researchers’ job does not stop there but process evaluations and the use of mixed methods are necessary to build a complete picture.

These critiques have relevance to researching the relationship of peer support to health care and health outcomes.

Current designs of RCTs in peer support discount variables that may play a part in the success or failure of a program at the outset. Therefore they do not figure in the results. Randomization, which is meant to produce intervention and control groups that are identical at the outset of a research project, may involve purposive or convenience samples and be based on clinical and demographic factors (diagnosis, age, medical interventions, and clinical results). They will ignore personalities, beliefs and behaviors, motivations, and expectations for participating in peer support and the contexts such as impact on significant others (e.g., partners and family) all of which will impact on results.

Hawe et al. (2004) suggest that, instead of trying to create laboratory conditions, it would be better to start by examining all the variables present to understand the complexity of a system to understand how an intervention will work. Similarly, the U.K. Medical Research Council (2019) recommends a thorough understanding of the contexts and the various parts of a complex intervention, as well as starting from a theoretical base. This gives coherence to the intervention and allows for replication. Piloting is important as well as building in a process evaluation. It is only after much preparatory work that a researcher considers which research design to adopt. Other recommendations include mixed methods studies to ensure that both quantitative and qualitative data are captured. Interviews of significant people to the study may uncover additional important variables to factor into the study. Bonnell et al. (2012) critique the use of RCTs in complex public health interventions arguing they may be modified to become “realist trials” that account for how components of interventions interact and the circumstances that assist interventions to work. “Realist trials” aim to demonstrate how components work in a trial and explore what aspects contribute to health changes. They use both qualitative and quantitative methods and contribute toward determining “the validity of program theory rather than only examining ‘what works’ to better inform policy and practice in the long-term” (Bonnell et al., 2012, p. 1).

In practical terms, this would include multiple groups receiving interventions as well as a group acting as a control and then comparing the results. It would also include examining more immediate impacts such as improvement in knowledge or self-efficacy rather than changes to health behaviors which might be the goal. This has the advantage of identifying effective “ingredients” of an intervention. Another inclusion might be conducting research across different contexts, using consistent measures to assess how the intervention results vary with context.

The Value of Considering Peer Support as a Complex Intervention

These critiques of RCTs and the U.K. Medical Research Council recommendations suggest that peer support is best viewed as a complex intervention. Complexity can be portrayed as a system of multiple components that are dynamic, sometimes emergent processes and people interacting with one another (Braithwaite et al., 2017; Hawe, 2015; Mannell & Davis, 2019). Complex systems are portrayed as alternating between stability and chaos and therefore inherently unpredictable and this has implications for research into them (Khan et al., 2018).

Peer support for people who have experienced illness is itself a complex system operating within the complex contexts of health care systems and the broader social structures. Members of a support group bring their various beliefs, judgments, experiences, personalities, social relationships, and health statuses into the group, making it a replica of the broader complexity.

Hawe (2015) considers that applying complexity thinking will change how interventions are designed and evaluated. One of those changes in thinking relates to evaluating “real-world” interventions such as behavior change (Schoen et al., 2020). Greenhalgh and Papoutsi (2018) agree with Khan et al. (2018) that by embracing complexity rather than attempting to control for it, as do reductionist methods, interventions will be tailored to work within local contexts while recognizing change takes time.

Furthermore, applying reductionist methods such as the RCT to complex systems may relegate possibly important variables to background “error” so they are not evaluated for their influence on the outcomes. In addition, reductionist methods in an RCT may overestimate the role of the intervention (Hawe, 2015). Greenhalgh and Papoutsi (2018) point out that dependence and interdependence of components in complex interventions make it difficult to identify the “active ingredient” required in RCTs. This has led to recognition of the value of mixed methods and process evaluations, or a “paradigm shift.”

In 2006, an international group examined four approaches to complex systems which began with establishing a theoretical framework, observation of interactions within the system, an intermediate RCT, followed by assessing the project for broader adoption and any potential adverse impacts (Boon et al., 2007).

Paul et al. (2007) adopted the U.K. Medical Research Council 2001 five-phase framework to test complex interventions. In their study of peer support in diabetes self-management, they argued that this framework allowed them to explore and pilot the key components of peer support before proceeding to a cluster RCT. This article discussed the first three phases only, with Phases 1 and 2 using qualitative work which assisted in designing a flexible program where peer supporter training was designed in association with those being trained and the inclusion of process evaluation. Results of the process evaluation would then inform the implementation phase. This evaluation was another RCT, conducted over a 2-year period and resulted in little change in the clinical indicators, though participants reported valuing the groups (Smith et al., 2011). The authors note the limitations of conducting a clustered randomized controlled trial in general practice, including difficulty of blinding and poor attendance of participants. They also report there was a decrease in well-being among participants at one point.

More recently, Deaton and Cartwright (2017) critiqued the use of the RCT where social context is ignored. Complexity theory and the poor RCT results already mentioned suggests that the role of peer support should be researched in a far wider context than the health system (Rooghenas et al., 2019). Deaton and Cartwright (2017) argue that social context is important regarding the relevance of an RCT to wider populations and support that view that observational studies are important in ensuring context is understood. On this basis, we argue it is important to understand peer support from the participants’ points of view, their motivations, and what they derive from participation.

This includes the point at which people decide to participate. While an illness will impact individuals differently, the impact may be a life-changing moment in each person’s life. Frank in The Wounded Storyteller (2013) discussed the effect that an illness can have on a person’s identity and he portrays the need to tell one’s own story as a quest toward healing oneself and arriving at a new point where the illness is integrated into one’s identity. Personal journeys are highly variable and acknowledge that life-changing events may be stressful and affect physical health and functioning (Rios et al., 2014). Seeking social support including peer support may be one part of these personal journeys toward new identities and will influence health outcomes.

Social support in health (e.g., Wilkinson & Marmot, 2003) also plays an important role in individuals’ abilities to participate in peer support. Social connectedness (being married, having family, friends, neighbors, community memberships, and employment) contributes to better health outcomes (Walker, 2010). Health outcomes variations may be influenced, for example, by partner and family attitudes to peer support and its effects, and hence be important influences to understand.

Kingod et al. (2017, p. 95) effectively summarize the interplay between the personal, social, and health system:

Through their online interactions, individuals with chronic illness animated illness associated identities, sought and provided social support and connectivity, shared experiential knowledge only available from those living with particular chronic illnesses, while also mobilizing collective voices for the purposes of promoting otherwise neglected perspectives concerning life with chronic illness.

A key finding is that the longing for mutual solidarity and emotional support in relation to the day-to-day management of illness motivates people with a chronic illness to seek advice and inspiration among peers with the same condition (Kingod et al., 2017). This suggests that research aimed at exploring how peer support improves knowledge, self-efficacy, engagement in one’s own health care, and its role in personal change would reveal the effectiveness of peer support in health care.

Consider Other Measures and Methodologies

While we do not see that the RCT will be abandoned in the foreseeable future, these measures can be applied in other research designs, for example, longitudinal studies where change can be measured over time within a cohort or panel (Caruana et al., 2015; Tse et al., 2017; Young et al., 2005).

Longitudinal studies can also be qualitative. Tse et al. (2017) collected views on peer support from health professionals, co-workers, and participants over 12 months, which demonstrated that peer support improved relationships between health professionals and their patients.

In some instances, longitudinal studies can be as equally limited as many RCTs. For example, Batenburg and Das (2014) ran a longitudinal study with a small number of breast cancer patients that had two reporting points: one at the beginning and then at 6 months. In this study, numbers declined over the period studied and there was little information on the intervention itself.

Mixed methods research combining qualitative and quantitative methods is a comprehensive approach to collecting data on health outcomes as an initial qualitative collection can inform about the context and meanings that surround a health issue and inform the collection of quantitative data. Or qualitative collections can precede quantitative methods. Thomson and Balaam (2019) collected data on neonatal peer support using an online survey followed up by interviews of a purposive sample to inform on the management and supervision of peer support workers in their wider context. Another study using mixed methods supplemented an RCT with semi-structured interviews comparing peer support with other service supports for people with mental health issues (Wrobleski et al., 2015). Another used focus groups as well as quantitative data collections to analyze peer support for psychological problems (Aschbrenner et al., 2016). In addition, an RCT was supplemented with a final interview program for peer support for people with brain injury, showing that other factors beyond the RCT needed investigation (Munce et al., 2019). These studies demonstrate that quantitative and qualitative data collection can be used to overcomes one of the chief criticisms of RCTs by Deaton and Cartwright (2017) that many lack an understanding of the context of outcomes of interventions.

Process evaluations, both quantitative and qualitative, are important to pursue as they provide detailed analysis of the content of peer support programs, including training of peer supporters, delivery of the curriculum, the responses of the participants and documentation of adverse events, plus where the strengths and limitations of the program occur. A process evaluation needs to be designed at the outset and to operate either continuously or at certain points identified as delivering outcomes. Those conducting the process evaluation should be separate from those delivering an intervention.

Given that participants may have different goals, motivations, and be on differing journeys, measuring peer support needs reconsideration. The question becomes whether to measure the components participants consider their own goals or to track changes in health behaviors and associated behaviors within a peer support group or program. If it is to be changes in health behaviors, clinical outcomes such as HbA1C or using the 6 Minute Walking Test (6MWT) or weight loss, the question to next arise relates to a realistic timeframe for specific participants.

As we have seen, changes in clinical outcomes may require long timeframes to demonstrate effectiveness, while clinical measures may not capture the full complexity of the intervention and the context. Here, Patient Activation Measures (PAM) have capacity to measure the complex interrelationships with support groups. Rather than capturing clinical outcomes, their application can measure behavior change that indicates progress toward potentially effective health outcomes (Hibbart & Gilburt, 2014).

In addition, measures of self-efficacy and health literacy especially if performed pre and post intervention might be better indications of a person’s progress in a journey toward improved health outcomes (Bandura, 1977). Depression and anxiety scales will also yield valuable information on participants’ ability to accept new information and make changes.