Peer support of complex health behaviors in prevention and disease management: a review with emphasis on diabetes

Peer support (PS) provided by “community health workers,” “lay health advisors,” “promotores,” “patient navigators,” “peer supporters,” and individuals with a number of other titles has been shown to play influential roles in health and the health care delivery system [1–4]. In particular, PS strategies can encourage appropriate regular care, can provide practical and emotional support for complex behaviors that are critical to staying healthy, and can help individuals cope with the stressors chronic diseases and conditions so often entail [5–15]. Calls-to-action and formal policy recommendations for the implementation of PS approaches [16–19] include numerous provision for community health workers in the US Affordable Care Act and a strong emphasis on Community Health Workers in the World Health Organization’s Global Health Workforce Alliance [20]. Programmatic and methodological challenges nevertheless limit what we know about PS approaches’ impact on health. Among these challenges, PS often takes on many definitions, roles, and forms [21–24] making it difficult to summarize or consolidate evidence across studies.

As detailed in the Methods and Results, a number of reviews have examined PS programs. However, most of these have focused on a specific health problem or challenge (e.g., promoting breastfeeding), or modality (e.g., telephone support). As detailed in this paper, we identified 24 reviews of PS interventions, of which 21 were focused on PS in a specific problem area of prevention or care, or a specific modality. The three that examined PS more broadly included that by Swider [1] published in 2002. That by Viswanathan and colleagues [2] limited its focus to PS through community health worker interventions to “create a bridge between community members, especially hard-to-reach populations, and the health care system” (p. 793). It found “moderate” evidence in impacts on knowledge, health behaviors, utilization, and cost/cost effectiveness. The third, by Gibbons and Tyus [3] limited its focus to US-based programs for those traditionally lacking access to care. It reported “efficacy in enhancing outcomes” across mammography, cervical cancer screening, and a variety of other health/prevention objectives. A fourth, more recent review by Perry and his colleagues in the 2014 Annual Review of Public Health [25] identified contributions of community health workers to basic health needs in low-income countries (e.g., reducing childhood undernutrition), to primary care and health promotion in middle income countries, and to disease management in the United States and other countries with developed economies.

In this present review, we sought to identify reports of PS interventions from around the world, addressing a wide variety of prevention and health objectives entailing sustained and complex behavior change (e.g., promoting diabetes management, but not flu shots), and using a broad definition of PS entailing assistance and encouragement for those behaviors as well as linkage to appropriate care. We then extended that review in two ways. We included a review of reviews of peer support across the same broad range of complex behavior change, and we updated the review through the end of 2015 focusing on diabetes prevention and management, diabetes being both a major source of global disease burden and a model for chronic disease prevention and care in general.

Categorization of designs

Studies were categorized according to their design as: (a) Randomized Controlled Trials (RCTs) (including cluster randomized designs), (b) Other Controlled Designs (including quasi-experimental and multiple baseline designs), (c) Within Groups, Pre-Post designs, and (d) Other designs such as analyses of uncontrolled program records, non-equivalent comparison groups, or comparisons to those offered a program but who chose not to join. Allocation was conservative. For example, a comparison to “those who requested service and were not contacted” [40] was characterized as quasi-experimental in its published paper but was categorized in this review as “Other.”

Categorization of measures employed

A number of studies provided challenges to this coding of measures. In cases in which the proper coding of a study’s outcome measure was unclear, the study was coded in the ‘weakest’ type, i.e., Non-Standardized weaker than Standardized weaker than Objective. For example, a study of impacts on breastfeeding utilized computer-based administrative data from the Women’s Infants and Children’s program in Michigan [40]. On the one hand, this might be coded as Objective because the data were drawn from an administrative database. However, because the source of the data in that administrative database appeared to be case workers’ notes, the study was coded as Non-Standardized.

Categorization of outcomes

Outcomes were categorized as: Significant Between Group differences (SBG) or Significant Within Group or pre-post differences (SWG) indicating effects of PS, Nonsignificant (NS), or Counter indicating negative impacts of PS. Based on these categories, studies were then categorized according to their strongest findings. That is, a paper with one Significant Between Group difference and one Significant Within Group difference would be placed in the former category. Because of the variety of designs and multidimensional nature of many evaluations, it was not possible to categorize studies according to their primary or hypothesized outcomes.

Review of reviews

In addition to the original research papers identified through the search described above, reviews of PS interventions were captured using parallel search strategies and from the references of the research studies identified. Many of these reviews disaggregated studies into specific categories such as by type of outcome, e.g., knowledge, health behaviors, clinical measures, quality of life. For each of these categories of each review, the percentages of studies showing a significant effect of PS was documented.

Study selection

The literature search for the initial review identified 1442 peer reviewed articles. The process of selection of relevant articles is displayed in Fig. 1. In the first iteration of article selection, 1160 articles were excluded because their titles and/or abstracts indicated that they were not research/intervention papers. Of the remaining 282 articles, 230 were excluded based on the specifics of these programs (as described in the Methods and displayed in Fig. 1). This selection process yielded 52 articles describing the effect of PS programs on complex behavioral health outcomes. An additional 13 articles were identified from the reference sections of these 52 papers, yielding a final sample of 65.

Study characteristics

Details of all 65 studies are available in Additional file 1, “Details of Studies from Systematic Review of Peer Support.”

The 65 studies in the final sample represented eight countries: 34 studies from the United States, seven from Canada, four from each of Bangladesh, England, Pakistan, and Scotland, and one from each of Australia, Brazil, Denmark, Ireland, Mozambique, New Zealand, South Africa, and Uganda. Fifty three were from World Bank designated high-income countries and 12 from low-income, low-middle, and high-middle-income countries.

Synthesis of results

Extended review of peer support in diabetes management

Summary of evidence

Across all 65 investigations included in the initial review, 54 or 83.1% found evidence of effects of PS. These include 31 between-group findings favoring PS out of 43 (72.1%) RCT or other controlled trials using objective or standardized outcomes and another 5 (11.6%) reporting significant within group changes. Thus, among RCTs and other controlled trials using objective or standardized measures 83.7% showed significant effects of PS. Within the several categories of design and type of outcome measure, the percentages showing effects of PS ranged from 71.4 to 100%. Similarly, across types of health problems, the percentage of studies reporting significant results for PS ranged from 66.7% for Addictions (Drug, Alcohol, Cigarette Smoking) to 90% for prevention/management of cardiovascular disease. Among 24 review papers, 19 provided quantifiable findings including a median of 64.5% of papers across individual reviews reporting effects of PS. Finally, among the 30 studies identified through December, 2015 that evaluated peer support with diabetes, 86.7% reported significant between or within group differences for peer support. Of the 24 diabetes studies using objective or standardized measures in RCT or Other Controlled Designs, 83.3% reported significant between or within group differences for peer support. The average reduction in HbA1c was 0.76 points among those 23 reporting such results. The evidence for PS is broad.

Analysis of interventions with nonsignificant effects

Lack of acceptance of the PS intervention on the part of those for whom it was intended appears to have been a feature of nine of the studies (Graffy 2004; Hunkeler 2000; May 2006; Muirhead 2006; Simoni 2007; Smith 2011; Vilhauer 2010; Palmas 2015; Simmons 2015) [38, 58, 59, 67, 74, 75, 77–79] Two of these (Hunkeler, 2000; May 2006) [38, 74] added PS to interventions which already offered appreciable support. A third study among HIV-positive adults suggests that concerns for confidentiality may limit acceptance of PS (Simoni, 2007) [78]. In a fourth, lack of acceptance of the goals of the study may have accounted for the poor acceptance of PS for breastfeeding when offered to women among whom half “…simply did not want to breastfeed…” (Muirhead 2006, p. 196) [75]. This may have been exacerbated by lack of acceptance of PS by professionals in the setting among whom the authors speculated “…some may be unwilling to accept lay people being involved in the care of women” (p. 196). Failure with those who are not interested in breastfeeding was also noted in the review of Ingram, 2010. [51] A fifth (Vilhauer 2010) [79] appears to have demonstrated that mailings announcing the availability of an online support resource are not effective; 900 mailings yielded only 31 participants.

Five of the nine studies illustrating Lack of Acceptance raise the issues of frequency of contact and how PS is presented and promoted. In Smith 2011 [58], PS classes were offered less than bimonthly (9 over 24 months) and participants were discouraged from contacting the peer supporters between meetings. Instead of using peers to engage those who did not attend, project staff contacted them. Additionally, the meetings appear to have focused more on discussion of a series of topics than on exchange among participants or their individualized goals for self management. In Graffy 2004 [77], PS for breastfeeding after the mother returned home from hospital was left to be initiated by the mothers. Similarly, in Muirhead 2006 “…Women in the PS group who did not commence breastfeeding or who stopped while still in hospital received no PS postnatally….”(p. 196) [75]. In the Palmas 2014 study, “…in over half of the intervention group, the CHWs were not able to deliver any of the planned one-on-one or small group sessions and only able to contact participants by phone” (p. 968). Extending this line of reasoning, adjustment for number of contacts led to a borderline (p = 0.054) effect for the CHW intervention (p. 967) [76].

The study of Simmons and colleagues [67] perhaps points to challenges of scope and complexity that may explain low participation. It deployed 127 peer support facilitators in group, individual, and combined group and individual arms of a 2 X 2 factorial, cluster randomized design with 1299 randomized participants drawn from three counties and “… 62 general practices, a hospital clinic and Diabetes UK members.” One nurse served as the principal study manager. “…Only 61.4% (592/977) of intervention participants attended an actual peer support session.”

Possible harm of unmoderated PS emerged in several interventions for individuals beset by substantial health problems or stress. This includes serious mental illness (schizophrenia spectrum or affective disorder, Kaplan, 2011) [80] or cancer, as in the papers reporting interventions for those with newly diagnosed (Salzer, 2010) [82] and metastatic diseases (Vilhauer, 2010) [79]. It may also have been a problem in the Nicholas 2007 [81] intervention for parents of technology assisted children with lung disease who were assigned to dyads to exchange support. Additionally, several of the reviews (Hoey 2008, Campbell et al., 2004) [47, 83] suggested that unmoderated PS for those with high stress may be unproductive. Earlier findings of Helgeson [84] as well as two RCTs included in the review by Campbell [47] showed negative or mixed results of peer-led support groups. In contrast, however, PS interventions moderated by professionals have shown striking effects on quality of life and, in some cases, survival among women with metastatic disease (co-lead by psychiatrists or social workers with a counselor with breast cancer in remission) [85–87] and among newly diagnosed breast cancer patients (delivered by clinical psychologists) [88, 89]. It appears that key combinations of seriousness of problem, emphasis on emotions, peer-to-peer support, and roles of professionals need to be elucidated as they may yield either positive or adverse effects.

Two studies with diabetes failed to show overall benefits of PS versus control conditions, but nevertheless showed important benefits among subsamples. In the Chan 2014 study [70], all participants received standardized clinical care that included initial assessments, reports on medication adherence, self-monitoring, weight, diet, physical activity for patients and treatment intensification recommendations for physicians, and periodic reports to patients providing status updates and recommendations [90]. The addition of PS did not improve clinical indicators. However approximately 20% of the sample were above norms for depression, anxiety, and stress and also accounted for highly disproportionate rates of hospitalization. In this 20%, peer support not only reduced distress substantially, but also lowered rates of hospitalization to the levels of individuals not similarly distressed. The second diabetes study showing effects of peer support in a subsample was that of Simmons 2015 [67]. In spite of modest participation rates, noted above, assignment to group PS (with or without individual PS) was associated with significantly (p = 0.008) greater reductions in systolic blood pressure than other conditions.

Summarizing, the null or negative findings identified appear to be attributable to methodological problems, identifiable problems in how PS was implemented or promoted to those intended, or populations for whom unmoderated PS may be inappropriate. Further, in spite of lack of overall differences between PS and controls, several studies document apparent important benefits within subsets of participants.

Cost effectiveness

Although cost savings and related outcomes have not been widely evaluated in the studies reviewed, a number of papers in the field have documented cost-effectiveness of PS [15, 29, 69, 91–96]. The review of Hunt 2011 [97] identified cost savings through reductions in emergency care and reductions in hospitalizations and Medicaid costs among adults with diabetes.

The paper by Forchuk and colleagues (2005) [98] reported remarkable cost savings through an intervention to improve quality of life and post discharge status of individuals with chronic mental illness in Canada. In addition to one year of PS post discharge, the intervention included continuity of contact between in-patient and post-discharge community staff. “At a rate of $632.30 CDN [Canadian dollars] per day cost for a bed in a psychiatric hospital, the people in the intervention group consumed $12,212,242 CDN less in hospital services than the control group, prior to discharge.” (p. 562). However, the $4400 CDN per person reduction in hospital and emergency services was only of borderline significance (p = 0.09) as variance in hospital cost data compromised sensitivity of analyses. Nevertheless, the effect of the post-discharge program on pre-discharge costs was suggested by ward staff reports; “…with the support that the [program] provided they felt comfortable discharging clients earlier…” (p. 562). Clearly, cost effectiveness and related analyses are important areas for growth of research on PS.

One of the studies identified through the extended review of peer support in diabetes also led to a cost-effectiveness analysis [99]. It used the differences attributable to peer support in an initial study (Prezio, 2013, 2014) [63, 100] and the Archimedes Model to estimate incidence of diabetes complications over twenty years. This led to an estimated incremental cost in the PS condition of $355 per quality adjusted life year gained. For reference, $50,000 per quality adjusted life year is often viewed as a benchmark for good value [91].

Reach of PS

As in the promotion of breastfeeding of Graffy 2004 [77] in which no further contact with new mothers who had left the hospital would occur unless they contacted the peer supporters, simply inviting individuals to contact peer supporters may be insufficient to bring about such contact. More proactive approaches to offering PS may be necessary, especially among underserved groups whose experiences with the health care system may leave them understandably suspicious of new interventions. For example, a combination of persistent but low-demand contact established working relationships with “Asthma Coaches” among 89.6% of low-income, predominantly African American single mothers of children who had been hospitalized for their asthma [13]. In Pakistan, the successful “Lady Health Worker” intervention for post-partum depression [101] mitigated the impact of factors such as lack of financial empowerment on depression which, in the absence of intervention, sharply differentiated those becoming depressed [102].

In this regard, it is perhaps of significance that, of the 12 studies identified in the present review that were conducted in World Bank designated low-income economies (Bangladesh – 4, Mozambique – 1, Uganda – 1), lower-middle-income economies (Pakistan – 4), and upper-middle-income economies (Brazil – 1, South Africa – 1), 10/12 (83.3%) were RCTs in contrast to 38/54 (70.4%) of those from high-income economies. Nevertheless, all 12 of those from the low-income, low-middle, and upper-middle-income economies showed significant within- or between-group differences favoring PS, in contrast to 42/54 (77.8%) of those from high-income economies. It appears that PS can be well implemented and effective in under-resourced settings.

Peer support in diabetes management

In the extended review in diabetes, a similar proportion of studies using RCT or Other Controlled Designs and objective or standardized outcomes reported significant effects of peer support (20 of 24 or 83.3%) as in the initial, broader review (36 of 43 or 83.7%). It is noteworthy also that the average reduction of HbA1c observed over the 23 studies reporting these data was 0.76 points. This is well above a benchmark 0.5 points generally considered to be clinically meaningful.

Two recent reviews have also examined peer support in diabetes management. Palmas and colleagues [59] restricted their review to RCTs evaluating changes in HbA1c in comparisons of community health workers versus usual care over a period of at least 12 months. They identified nine studies meeting these criteria with a standardized mean difference in change in HbA1c for intervention versus usual care of 0.21 (95% confidence interval = 0.11–0.32). In another 2015 review, Zhang and colleagues [103] examined 20 RCTs and found a pooled effect on HbA1c of −0.16 points (95% confidence interval of −0.007 to −0.25 points, p < 0.001). This review examined differences according to several categories of peer support providers. The greatest difference between changes in intervention and control groups (−0.49 points, 95% confidence interval − 0.12 to −0.86 to) was for interventions “provided by patients themselves to help each other or to share experience together in a group, usually with no specific leader during the intervention,” however, there were only two studies in this category [56, 104]. Substantial effects (−0.35 points, 95% confidence interval = −0.16 to −0.54) were also found for interventions “provided by nonprofessionals like community health workers, medical assistants, or community lay workers who had similar background or shared similar local culture with patients.” Smaller effects (−0.08 points, 95% confidence interval = −0.03 to −0.18) were found for interventions “led by one or several peer coaches, peer leaders, peer educators, peer supporters or peer mentors who were usually also patients but had received relevant training.” It should be noted that the effects on HbA1c in these two meta-analyses reflect the difference in changes between peer support and control conditions, while the 0.76 point mean difference identified in the present review only reflects the pre-post change within the condition receiving PS.

Limitations

Because of the aforementioned variety of a) designs, b) health problems addressed, c) pertinent outcomes, and d) intervention strategies, the present review has not graded studies according to general criteria of quality of research design or risk of bias beyond the categorizations of type of design (RCT, Other Controlled, Within-Group Pre-Post, and Other), and type of outcome (objective, standardized, nonstandardized). Also, it has not included meta-analytic characterization of results of studies reviewed. Continued review of research on PS will benefit from such approaches. These limits were accepted in the interest of capturing the variety of PS programs and outcomes around the world.

It is important to recall the restriction of the present review to interventions applying PS to complex health behaviors needing to be sustained over time and thus the exclusion of interventions focused on isolated or single behaviors, such as undergoing mammography or other screening procedures. This restriction was not based on any supposed lack of importance of these other applications of PS, but only on recognition that PS interventions and factors key to their success may be very different in the two arenas. Evidence indicates effects of PS in screening, [28, 105, 106] effects that complement those identified here.

The initial broad review of articles and reviews only through July 15, 2011 is a shortcoming of the paper. In addition to the extended review of papers in diabetes through 2015, subsequent papers and reviews have continued to document the effects of peer support in a manner similar to those noted here (e.g., [25, 29, 107]).