Predictors and Effects of Participation in Peer Support: A Prospective Structural Equation Modeling Analysis

Abstract

Background

Peer support provides varied health benefits, but how it achieves these benefits is not well understood.

Purpose

Examine a) predictors of participation in peer support interventions for diabetes management, and b) relationship between participation and glycemic control.

Methods

Seven peer support interventions funded through Peers for Progress provided pre/post data on 1,746 participants’ glycemic control (hemoglobin A1c), contacts with peer supporters as an indicator of participation, health literacy, availability/satisfaction with support for diabetes management from family and clinical team, quality of life (EQ-Index), diabetes distress, depression (PHQ-8), BMI, gender, age, education, and years with diabetes.

Results

Structural equation modeling indicated a) lower levels of available support for diabetes management, higher depression scores, and older age predicted more contacts with peer supporters, and b) more contacts predicted lower levels of final HbA1c as did lower baseline levels of BMI and diabetes distress and fewer years living with diabetes. Parallel effects of contacts on HbA1c, although not statistically significant, were observed among those with baseline HbA1c values > 7.5% or > 9%. Additionally, no, low, moderate, and high contacts showed a significant linear, dose–response relationship with final HbA1c. Baseline and covariate-adjusted, final HbA1c was 8.18% versus 7.86% for those with no versus high contacts.

Conclusions

Peer support reached/benefitted those at greater disadvantage. Less social support for dealing with diabetes and higher PHQ-8 scores predicted greater participation in peer support. Participation in turn predicted lower HbA1c across levels of baseline HbA1c, and in a dose–response relationship across levels of participation.

In seven diverse, international settings, less available support, greater depressed mood, and older age predicted more contacts with peer supporters for diabetes management. More contacts predicted better blood glucose control.

Peer support (PS) provided by individuals with a variety of roles and titles such as “Community Health Workers,” “Lay Health Advisors,” “Promotores de Salud” or, in Thailand, “Village Health Volunteers” provides varied benefits in prevention and health care [1–3]. A 2014 review in the Annual Review of Public Health by Perry et al. [4] identified broad contributions of peer support to meeting basic health needs (e.g., reducing childhood mal/undernutrition), enhancing primary care, health promotion, and disease management.

Central to the prevention and management of chronic disease is sustaining a constellation of behaviors including accessing clinical care on a regular basis, self-monitoring, adopting and sustaining healthy lifestyle habits (e.g., increasing physical activity, eating a healthy diet), regularly monitoring signs and symptoms, taking medications to prevent and control symptoms, coping with possible medication side effects, all while managing the additional stress of having a long-term health condition. A systematic review of articles [5] published between 2000 and 2011 listed 46 papers that addressed PS in promoting sustained, complex health behaviors in cardiovascular disease, diabetes, and other chronic diseases. Across all 46 papers, 83% reported significant clinical, behavioral, or quality-of-life benefits of PS in comparisons to control groups (24 papers, 53%) or in prepost changes within groups (14 papers, 30%). Among the 33 papers reporting on randomized controlled trials, 82% reported benefits in comparison to controls (22 papers, 67%) or in prepost changes (5 papers, 15%). A 2017 expansion of this review including applications of peer support outside of chronic disease management [6] found a similar pattern of results, and a 2016 review [7] also found benefits of peer support in diabetes management.

Current U.S. National Standards for Diabetes Self-Management Education and Support [8] emphasize the importance of ongoing diabetes self-management support (DSM Support) subsequent to diabetes self-management education. As noted in the 2017 standards, “The effectiveness of providing support through diabetes educators, disease- management programs, trained peers, diabetes paraprofessionals, community-based programs, or through the use of technology has also been established” (emphasis added) [8]. Peers can encourage and help individuals “troubleshoot” ongoing disease management and can provide motivational boosts and emotional support in dealing with the distress chronic disease often creates.

Peers for Progress was founded in 2006 to encourage research about and exchange of effective approaches to peer support around the world with a special emphasis on diabetes [9]. It initially funded 14 projects in 9 countries on 6 continents. Across the sites, hemoglobin A1c (HbA1c) as a measure of glycemic control declined from an average of 8.49% to 7.74%, systolic blood pressure from 137.0 mmHg to 133.6 mmHg, and body mass index (BMI) from 31.96 kg/m² to 30.85 kg/m² [10]. Several of the 14 projects have published their findings independently [11–20] as well as through a 2015 supplement of the Annals of Family Medicine [21–25].

The standards for diabetes self-management education and support [8] cited above make clear the importance not just of education for self-management but of ongoing support that is not simply repetition or “booster” offerings of education. This points to the value of ongoing access to peer support, someone to speak with when questions arise, and assistance with obtaining health resources and services. These underscore the importance of engagement and contact with peer supporters. In this article, we examined individual participant characteristics that predicted participation in seven peer support interventions supported by Peers for Progress and then explored the relationship between degree of participation with the peer supporter and changes in glycemic control as measured by HbA1c. Although far from a comprehensive measure of diabetes management, we used HbA1c surely not as a comprehensive characterization of clinical status but as one objective indicator common in diabetes research and care.

Methods

Study Settings

As mentioned previously, Peers for Progress funded 14 projects to build the evidence base for peer support, the results of which have been previously reported [9–19, 21–25]. Among controlled investigations at eight sites, six collected systematic data on numbers of contacts of individual participants with peer supporters. One of these six included two separate interventions. The balance of this paper focuses on participants in the seven intervention conditions of these six projects in order to study the determinants and effects of participation in peer support.

Intervention Sites

• Site 1, Rural Alabama, USA: Peer support provided in a community-based setting for a predominantly African-American, rural population [25]

• Site 2, Ann Arbor and Ypsilanti, Michigan, USA: Peer support provided in community settings for African-Americans [14]

• Site 3, Southwestern Detroit, Michigan, USA: Peer support provided through a clinical setting for Latinos [23]

• Site 4, San Francisco, California, USA: Peer supporters integrated in clinical treatment teams for Latinos, African-Americans, Asians, Asian Americans, and whites [12] in a university medical center out-patient service

• Site 5, Imperial County, California, USA: Peer support provided through a federally qualified health center by telephone or in-person visits for Mexican-origin adults living on the Mexico-US border [22]

• Site 6, Cambridgeshire, England: Peer support for adults receiving care through the National Health Service implemented in three conditions: individual support, group support, or both individual and group support [19, 26]

• Site 7, Hong Kong, Special Administrative Region, China: Peer support provided by telephone [27, 28] to Chinese patients aged 18 to 70 years recruited through three cooperating diabetes centers that provided technologically assisted, team-based integrated assessments and care [29, 30]

Cross-Site Data

Self-report measures

Common measures assessed at baseline and postintervention included a measure of availability and satisfaction with support for diabetes management that was based on measures developed by Tang et al. [31]. Separately for each of “friends and family,” and “your health care team,” it asked participants to rate two statements, “how much support do you get …” and “how satisfied are you with the support you get …” “for dealing with your diabetes” (total of four items, each on 5-point scale). Baseline measures also included a three-item screener for health literacy [32] including how often participants had someone “help you read hospital materials,” how often they had “problems learning about medical conditions because of difficulty understanding written information,” and how confident they were in “filling out health care forms by yourself” (5-point scale). These measures were included in the models given their potential moderating role in examining the relationship between receipt of peer support and glycemic control.

Additional self-report measures included general quality of life measured using the EQ-5D [33]; depression measured using the Patient Health Questionnaire (PHQ)-8 [34]; and diabetes distress measured by four items drawn from the Diabetes Distress Scale [35].

Self-report and/or clinical records provided continuous measures of participant age, years of education, and years with diabetes, each scaled by 5-year increments, and also participant gender, all of which were included in modeling.

Glycemic control

HbA1c reflects the average blood glucose over the previous 3 months. It was obtained directly via venous puncture at the baseline and postintervention data collection visits for five of the seven interventions included [14, 19, 20, 23, 25, 27]. For the other two interventions, medical chart abstraction was used if HbA1c was obtained within one month [22] or within six to nine months prior to the time of assessment [12]; if not available, HbA1c was obtained directly via venous puncture. Hypoglycemia was not assessed in a standardized manner across sites.

Measures of Participation Across Sites

Although a variety of peer support models were tested, the seven interventions collected data on numbers of contacts individuals had with their peer support programs. Number of contacts, then, provided an indicator of participation that could be used across all sites. Because of variability across sites in terms of intervention duration and the numbers of contacts planned, those assigned to PS conditions were coded based on distributions within their sites as in low, moderate, or high tertiles among those participating, or, if data indicated 0 contacts, as none. Numbers of contacts were abstracted from peer supporters’ records and generally reflect only contacts initiated by the peer supporters.

As further presented in the Results, these contact data were also evaluated as dichotomous measures of “no” versus “some” contacts (low, moderate, and high tertiles collapsed). Descriptive statistics based on intervention group and participation group were estimated using random-effects models with a random intercept to control for between site differences using SAS 9.4. Testing of differences in participants versus nonparticipants as well as differences among participants (no, low, moderate, and high contacts) were estimated using this same mixed model.

Data Management

The designs, interventions, and project implementation schedules were directed by the investigators. Upon completion of each of the projects, the data managers at each site transferred their multiple-wave data to Peers for Progress staff at the University of North Carolina-Chapel Hill where the data were verified, cleaned, and merged into a single file using IBM SPSS Statistics Version 22.0 (IBM Corp., Armonk, NY, USA). Scale mean, sum, and index scores were computed as appropriate for each scale to prepare the data for analyses.

Structural Equation Modeling of Participation and Impacts on HbA1c

To model the pathway to participation as well as the impact of participation on final HbA1c values, we used structural equation modeling (SEM). The inclusion of variables in the cross-site, shared database was planned jointly by the investigators [9] and, along with general findings in the field, served as the base for modeling. Modeling examined two key study questions, the predictors of participation in peer support and the relationship between participation and final HbA1c as an indicator of the health-related benefit of peer support. Site, gender, age, years of education, and years with diabetes were included in the pathways to baseline indicators of health literacy, family and clinical team support, BMI, EQ-Index, PHQ-8, and diabetes distress scores, as well as in the pathways to participation and final HbA1c. In addition, health literacy, family and clinical support, BMI, EQ-Index, PHQ-8, and diabetes distress were included in the pathways to participation and final HbA1c. Latent variables were identified to estimate baseline health literacy and baseline support based on the corresponding set of questions specified above. SEM was modeled using MPLUS version 8.2.

Subgroup Analyses

Progression of treatment to include insulin therapy is associated with greater requirements for self-management, side effects such as hypoglycemia, and treatment complexity. Because these might influence the attractiveness, value, or benefits of peer support, analyses also examined the relationship between participation and HbA1c separately among those receiving or not receiving insulin therapy at baseline. Additionally, analyses also examined the relationships with participation for those with elevated baseline levels of HbA1c,

Results

There were 3,738 participants across all 8 sites who were assigned to treatment/intervention or to control/comparison conditions. Of these, 698 were eliminated from two sites that did not record numbers of contacts for each participant along with 14 individuals with type 1 diabetes. To examine determinants and effects of participation in peer support, all subsequent analyses were confined to those who were randomized to intervention and actually offered peer support, leaving a sample of 1,746. Table 1 presents the baseline characteristics (means and percentages) of these adults with type 2 diabetes, using univariate generalized linear mixed models to account for differences among sites. Additionally, individuals with missing information on age, gender, years with diabetes, or education (n = 134) were excluded from multivariate analyses, resulting in an analytic sample of 1,612.

Table 1.

Descriptive Statistics (Means and Percentages) of Baseline Characteristics and of Change in HbA1c by Participation Level and Tests of Differences Between Groups by Participation Level, Adjusting for Differences among Sites^*

	Intervention (N = 1,746)	Nonparticipants (N = 480)	Participants (N = 1,266)	Low participation (N = 486)	Moderate participation (N = 427)	High participation (N = 353)	p-values for between group differences*
							Participants vs. nonparticipants	Differences among participants
% Male	61%	61%	60%	60%	61%	60%	.5667	.3724
Mean (std) Age	57.52 (1.82)	56.49 (1.89)	57.84 (1.84)	57.52 (1.88)	57.62 (1.89)	58.54 (1.91)	.0174	.3369
Mean (std) Years Of Education	10.83 (1.17)	10.82 (1.18)	10.83 (1.18)	11.12 (1.18)	10.57 (1.18)	10.76 (1.18)	.9999	.0392
Mean (std) Years. With Diabetes	12.51 (1.08)	12.34 (1.2)	12.56 (1.09)	11.94 (1.18)	13.35 (1.19)	12.41 (1.22)	.7450	.2230
Mean (std) Social Support Score**	0.02 (0.08)	0.11 (0.09)	–0.01 (0.08)	0.03 (0.09)	–0.06 (0.09)	–0.02 (0.09)	.0065	.2713
Mean(std) Health Literacy Score**	–0.09 (0.13)	–0.13 (0.13)	–0.08 (0.13)	–0.05 (0.13)	–0.12 (0.13)	–0.08 (0.13)	.2803	.3319
Mean (std) BMI (kg/m2)	33.22 (1.19)	33.69 (1.22)	33.07 (1.19)	33.31 (1.21)	32.74 (1.22)	33.16 (1.22)	.1054	.4301
Mean (std) EQ Index Score	0.77 (0.04)	0.77 (0.04)	0.78 (0.04)	0.77 (0.04)	0.78 (0.04)	0.78 (0.04)	.5347	.6144
Mean (std) PHQ-8 Score	5.99 (0.5)	5.67 (0.54)	6.09 (0.5)	6.4 (0.53)	5.91 (0.54)	5.9 (0.55)	.1786	.2590
Mean (std) Diabetes Distress Score	2.34 (0.14)	2.31 (0.15)	2.35 (0.14)	2.41 (0.15)	2.33 (0.15)	2.29 (0.15)	.6150	.3201
Mean (std) Baseline HbA1c (%)	8.32 (0.31)	8.37 (0.32)	8.3 (0.32)	8.31 (0.32)	8.34 (0.32)	8.25 (0.33)	.4279	.7396
Mean (std) Change (from Baseline) in HbA1c (%)	–0.24 (0.13)	–0.04(0.15)	–0.28 (0.13)	–0.15 (0.14)	–0.31 (0.14)	–0.41 (0.14)	.0118	.0581

Descriptive statistics were model-based estimates from a random-effects model with a random intercept to control for differences among sites. Testing of differences in participants versus nonparticipants as well as differences among participants (low, moderate, and high participation) were estimated using this same mixed model.

Identification of Latent Variables

Using robust maximum likelihood estimation, latent variables were identified to estimate social support and health literacy at baseline. Social support was estimated by the four items noted in the Methods assessing amount of and satisfaction with support “for dealing with your diabetes” from each of “family/friends” and the “health care team.” Similarly, health literacy was estimated by three items noted in the Methods regarding “reading hospital materials,” “understanding written information,” and “filling out health care forms.” Analyses indicated a standardized latent variable including satisfaction with available support from both family and the clinical team (loadings from 0.73 to 0.95), and a standardized latent variable for health literacy (loadings from 0.69 to 0.86). Years of education, gender, age, and years with diabetes remained as observed variables.

Relationships Among Variables at Baseline

Supplementary Material 1 includes model coefficients and for relationships among model variables at baseline and between model variables and the 2-level measure of participation (Yes/No) as well as final HbA1c. For example, a 5-year increase in age was associated with a 0.11 unit increase in baseline support. Greater BMI at baseline was associated with male gender and younger age. Greater baseline HbA1c was also associated with younger age as well as longer disease duration. Turning to other measures, higher scores on the health literacy latent variable were associated with male gender and more education. The latent variable reflecting amount of and satisfaction with support from family/friends/care teams was also associated with male gender but with less education and with older age. Turning to distress and depression, higher scores on both the 4-item measure of diabetes distress and the PHQ-8 were associated with male gender and older age while higher scores on the PHQ-8 were also associated with greater education.

Table 2 includes correlations among variables at baseline.

Table 2.

Cross Sectional Pearson Correlations, Significance Levels, and Numbers of Participants Among Variables at Baseline and Pearson Correlations between Variables and Final HbA1c

Variable names	BMI (n=1603)	EQ Index	PHQ-8 Score	Diabetes Distress Score	Baseline A1c	Literacy	Support	Age	Years Education	Years with diabetes
EQ Index Score (n = 1,596)	–0.309 <0.00011590
PHQ-8 Score (n = 1,542)	0.227 <0.00011533	–0.596 <0.00011527
Diabetes Distress Score (n = 1,560)	0.067 0.00791552	–0.254 <0.00011545	0.493 <0.00011494
HbA1c (n = 1,598)	0.014 0.58331592	–0.126 <0.00011585	0.168 <0.00011529	0.292 <0.00011546
Literacy (n = 1,610)	0.096 0.00011601	0.090 0.00031594	–0.202 <0.00011540	–0.247 <0.00011558	–0.112 <0.00011596
Support (n = 1,610)	0.008 0.74091601	0.056 0.02551594	–0.163 <0.00011540	–0.206 <0.00011558	–0.065 0.00981596	0.235 <0.00011610
Age (n = 1,612)	–0.086 0.00061603	–0.035 0.15911596	–0.132 <0.00011542	–0.327 <0.00011560	–0.265 <0.00011598	0.090 0.00031610	0.124 <0.00011610
Years of Education (n = 1,612)	0.160 <0.00011603	–0.032 0.20371596	–0.025 0.32941542	–0.166 <0.00011560	–0.068 0.00671598	0.441 <0.00011610	0.089 0.00041610	0.046 0.06751612
Years with Diabetes (n = 1,612)	–0.001 0.96381603	–0.115 <0.00011596	0.043 0.08921542	0.004 0.87791560	0.131 <0.00011598	–0.051 0.04031610	–0.021 0.40341610	0.190 <0.00011612	–0.126 <0.00011612
Final HbA1c (n = 1,345)	0.083 0.00251336	–0.129 <0.00011334	0.128 <0.00011292	0.233 <0.00011302	0.614 <0.00011335	–0.022 0.41271343	–0.059 0.02951343	–0.219 <0.00011345	0.018 0.51351345	0.115 <0.00011345

Site Differences

As indicated in the Methods, site effect was controlled in all pathways including the pathways related to all baseline values of health literacy, family and clinical support, BMI, EQ-Index, PHQ-8, and diabetes distress score as well as in the pathway to participation and final HbA1c. Because of confounding among site characteristics (nation, clinical setting, health system, urban/rural, participant age and race/ethnicity, and intervention features), identification of meaningful effects of variables associated with site differences is not possible and, so, site differences have not been reported in any analyses of Peers for Progress. Reports of the individual projects [11–20] provide descriptions and evaluations of their interventions and their effects within the settings and populations in which they were applied.

Modeling Participation in Peer Support

Table 3 includes descriptive statistics for contacts for each site. Overall, 1,266 of 1,746 individuals assigned to peer support interventions (72.5%) were indicated to have had at least one peer support contact. Development of models using SEM utilized a binary categorization of participation: any contacts (N = 1,266) versus no recorded participation in peer support services or activities (N = 480).

Table 3.

Descriptive Statistics of Number of Individual Contacts of Participants with Peer Supporters by Site

Site	N	Minimum	Maximum	Mean	Median	Std. Deviation	Std. Error of Mean	Skewness
Site 1	198	0	58	12.11	12.50	8.547	0.607	1.435
Site 2	60	0	17	4.58	4.00	4.518	0.583	0.972
Site 3	54	0	34	10.76	4.00	12.535	1.706	0.616
Site 4	147	0	29	7.02	5.00	7.294	0.602	1.189
Site 5	165	0	24	5.17	4.00	4.825	0.376	1.739
Site 6	810	0	24	2.50	1.00	3.417	0.120	1.801
Site 7	312	0	101	16.65	18.00	12.693	0.719	1.928
Total	1,746	0	101	7.08	4.00	9.190	0.220	2.531

Figure 1 includes the significant pathways that were associated with participation. For the dichotomous indicator of participation, the relationships are expressed as odds ratios, calculated from exponentiated parameter estimates. The latent variable, social support, is denoted as a circle. Health literary, which was also a latent variable was not significantly associated with participation and/or final HbA1c and therefore was not included in this figure although estimates of these associations can be found in Supplementary Material 1. All other variables were measured directly. Supplementary Material 1 also includes model significant and nonsignificant pathway coefficients and standard errors for the 2-Level Participation Model (yes/no).

Fig. 1.

Predictors of 2-Level Participation (Yes/No) and final HbA1c, showing pathways and model coefficients significant at p < .05*, p < .01**, p < .001***. All estimates presented are parameter estimates with the exception of those bolded and italicized that denote odds ratios. Only significant pathways are included. The exception to this is the set of fixed variables for site that are significant in all pathways but for which the direction of association varies across the 7 interventions. The complete set of pathways including those for site is detailed in Supplementary Material 1. As detailed in the text, older age, lower levels of support from family or care team and depressive symptoms (PHQ-8) predicted higher levels of participation in peer support program which in turn predicted better glycemic control with lower final HbA1c. Additionally, female gender, longer disease duration, higher levels of distress, higher BMI and higher HbA1c at baseline each predicted high final HbA1c.

Lower levels on the latent variable reflecting support for dealing with diabetes were associated with a greater likelihood of participation with a one standard deviation higher score on the social support latent variable associated with an odds ratio for participation of 0.82 [95% CI: 0.70–0.96]. Additionally, a 1-point increase in PHQ-8, was associated with a slightly higher odds ratio for participation of 1.04 [95% CI: 1.00, 1.08]. An increase in age was also associated with participation with an odds ratio of 1.14 [95% CI: 1.05–1.23] for each five-year increase in age.

Modeling Impacts of Participation on HbA1c

As also portrayed in Figure 1, any participation in peer support was associated with final HbA1c 0.22 points (standard error = 0.09; 95% CI = –0.40 to –0.04) lower than among nonparticipants after controlling for all the other variables in the model as seen in Supplementary Material 1. Additionally, higher baseline HbA1c, baseline BMI, baseline diabetes distress, and years with diabetes were associated with higher final HbA1c while males and older participants had lower final HbA1c.

Subgroups with Elevated HbA1c

The sample included a number of individuals with HbA1c values within recommended ranges (e.g., 591 or 33.8% ≤ 7.0%, the “target for glucose control” of the Interntaional Diabetes Federation [36]). Subgroup analyses evaluated the relationship between participation and HbA1c among those with higher baseline HbA1c levels. Due to decreased sample sizes, these were not significant, but still showed the same patterns as analyses of the full sample. Among those with baseline HbA1c > 7.5% (n = 822), participation was nonsignificantly associated with a –0.14 point reduction in final HbA1c (95% CI: –0.46, 0.19). Further restricting to those with baseline HbA1c > 9% (n = 365, criterion of poor control of the National Committee for Quality Assurance [37]), participation was nonsignificantly associated with a –0.30 point reduction in final HbA1c (95% CI: –0.94, 0.33).

Subgroups On/Not On Insulin Therapy

As noted in the Methods, insulin therapy is often associated with greater requirements for self-management, side effects such as hypoglycemia, and treatment complexity. For those not on insulin at baseline (n = 1,134), the effect of participation was similar to that within the whole sample. Among those not on insulin, participants had on average a final HbA1c 0.30 points (standard error = 0.10; 95% CI = –0.49 to –0.10) lower than nonparticipants after controlling for the same variables as in the overall analysis.

For those on insulin at baseline (n = 477), the overall effect of participation was not significant. Given the importance of this group, we explored further the relationship between participation and final HbA1c among those on insulin. This analysis used the categorization described in the Methods in which individuals were coded based on distributions within their sites as in low, moderate, or high tertiles of participation, or as no participation for those assigned to peer support but with 0 contacts. Controlling for the same variables as in the overall analysis, we found a borderline significant trend for participation and final HbA1c (p = 0.053) with each increase in participation level associated with a –0.14 (standard error = 0.08, 95% CI = –0.29 to 0.01) decrease in HbA1c. Using post hoc tests, we also explored differences among the four groups (no, low, moderate, and high participation). The adjusted final HbA1c values by level of participation were: no participation—9.22% (n = 97, SE = 0.22); low participation—9.35% (n = 159, SE = 0.20), moderate participation—9.16% (n = 126, SE = 0.24) and high participation—8.84% (n = 95, SE = 0.22). The high participation group differed significantly from the low participation group, 8.84 versus 9.35% (p < .01), but other differences were not significant, due at least in part to tests being underpowered because of modest sample sizes, e.g., 97 and 95 in the no and high participation groups respectively.

Extent of Participation

The SEM analyses modeled participation as a binary variable (some, none). As in the previous examination among those on insulin, further analyses examined the several levels of participation for the entire sample: no participation or low, moderate, or high tertiles of participation. The covariate-adjusted final HbA1c for each of these four levels of participation is portrayed in Figure 2. The overall linear trend is significant (p = .002). The difference between those with no (8.18%) and high participation (7.86%) was 0.32 points. Supplementary Material 1 includes model coefficients and standard errors for the four-level participation model.

Fig. 2.

Estimated final Hemoglobin A1c (HbA1c) measure of blood glucose control (means and standard errors) by level of participation (low, moderate, and high tertiles of participation and no participation) and controlling for baseline HbA1c, site, gender, age, education, years with diabetes, and the latent variables support and literacy from structural equation modeling. Linear trend is significant at p = 0.002.

Discussion

Two important aspects of participation in peer support emerged from these analyses of seven peer support interventions. First, as measured by number of contacts, participation was greater among those reporting greater depression as well as less diabetes support from friends, family, and their health care team, and lower satisfaction with that support. Second, across diverse populations, settings, and models of peer support tested, and after controlling for and modeling the effects other variables (socio-demographic variables, years with diabetes, health literacy, and amount of satisfaction with diabetes support), participation in peer support interventions was significantly associated with improvements in glycemic control as measured by HbA1c. Additionally, final HbA1c varied in a dose–response manner by level of participation, none, low, moderate, or high. This supports a causal relationship between participation in peer support and glycemic control. Those receiving insulin therapy, however, only showed an association between HbA1c and high participation. Whether this association was influenced by the risk or incidence of hypoglycemia was not assessed.

The variety of settings of the seven projects that provided the data analyzed here is both a strength and weakness. That greater participation in the variety of peer support interventions across the seven is related to improved HbA1c suggests that the influence of peer support is robust. A weakness is that the variety of settings, populations, and designs introduces nonequivalence and otherwise complicates finding a simple effect across all seven interventions.

The present emphasis on participation may inform broader understanding of necessary features for peer support to be effective. In some prior reports of unsuccessful peer support interventions, participation was mediocre [38, 39]. There is nothing magical about simply offering peer support. Clearly, we need to understand better how peer support interventions can be constituted or presented so as better to engage individuals who may benefit from them.

One possible key to encouraging participation emerged in a systematic review of peer support interventions to promote breastfeeding. “Proactive” approaches—initiation of contact by the peer supporter, contact both before and after delivery, contact within 72 hours of delivery—were predictive of breastfeeding outcomes after controlling for likely confounders [40]. Another recent review found that engagement is possible and that peer support is effective in reaching those whom health initiatives rarely benefit (e.g., those with high distress, low income, low education, ethnic minority populations) [41]. The present study’s findings that a) greater depression, b) lower levels of baseline diabetes support from family, friends, and the health care team and b) older age all predicted greater likelihood of participation and that c) other socio-demographic characteristics (e.g., low education) were not associated with participation adds weight to the observation that peer support programs are able to reach those who may most need them.

Our findings reinforce the importance of participation found in prior studies. A 2014 review [4] noted “a possible dose-response relationship between CHW [community health worker] exposure and improvements in HAART [highly active antiretroviral therapy for HIV] adherence.” [42] The process evaluation of a successful dyadic support intervention in type 2 diabetes found a dose-effect around those who had more than 8 versus 8 or fewer contacts [43]. Similarly, a study of peer support for Latinos with diabetes in New York found a significant association (p = .04) between HbA1c reduction and number of phone contacts and a borderline association (p = .054) with total contacts among visits, group support, nutritional education and phone contacts [44]. The importance of participation has also been noted in its absence. In their evaluation of perinatal and newborn care promoted through “Lady Health Workers” in Pakistan, Bhutta et al. noted that the intervention was only able to reach 24% of total pregnancies, pointing then to “…the need for attention to issues of program management and coverage … to achieve maximum potential” [45].

As there is nothing magical about simply offering peer support, so too, there is likely nothing magical about simply receiving contacts from peer supporters or participating in it. The mediators of impacts of peer support likely include social modeling as well as social motivational influences such as patient activation and autonomy motivation [46–48]. Along these lines, a cross-site analysis of some of the same programs as included here showed participants’ reports of nondirective support from peer supporters were associated with lower while reports of directive support were associated with higher levels of depression [49]. Peers for Progress has encouraged consideration of 5 key functions of peer support: assistance in daily management, social and emotional support, linkage to clinical and community resources, ongoing support, and, foundational to the others, presence or “being there.” [50, 51] These have been useful in planning and understanding peer support as well as understanding reported failures [5] and recently in a scoping review of peer support for people with schizophrenia [52]. Thus, there are candidates for key characteristics of peer support and mediators of its impacts, but further research regarding these may guide interventions to achieve greater benefits.

As detailed in the Introduction, the settings of the interventions included community-based organizations, clinical diabetes centers, primary care settings, a federally qualified health center, a university-based out-patient clinic, and the National Health Service in the UK. Additionally, interventions included group and individually delivered peer support as well as combinations of these. Conceptualization of how these characteristics of setting and delivery may interact with content of interventions and needs and characteristics of recipients is needed along with research then to evaluate key features of how programs should integrate these many dimensions. Surely illumination of the many possible, complex, and nuanced combinations of these features may be advanced by varied research designs including qualitative evaluation and quality improvement models such as evaluation of rapid change cycles [53].

Considering the variety of interventions and settings, the changes in HbA1c associated with participation are striking. Although far from a comprehensive measure of clinical status, reductions in HbA1c have been shown related to health benefits. Also, when considered as changes across a population such as all offered peer support, they take on greater significance. In the present analyses among all participants, those with no and high participation differed by 0.32 points (8.18%–7.86%). To put this in context, the influential UK Prospective Diabetes Study [54] found that a 1-point reduction in HbA1c (e.g., 9% to 8%) was associated with reductions in microvascular endpoints, amputation/death from peripheral vascular disease, myocardial infarction, stroke and heart failure of 37%, 43%, 14%, 12%, and 16% respectively (p. 408). It also noted that these reductions were generally linear so that, extrapolating, the 0.32% difference observed here would be associated with reductions of 11.84%, 13.76%, 4.48%, 3.84%, and 5.12%, respectively.

Moreover, given concerns that many with diabetes do not receive recommended care, the fact that 72.5% of those offered peer support interventions participated is noteworthy.

At the same time, the dose–response relationship between participation and glycemic control should drive further attention to how peer support merits recognition as a standard component of diabetes self-management education and support services [8]. Far more than a pleasant accoutrement, the present findings as well as the many others cited in the Introduction [1–7, 10–25, 27, 28] make clear that peer support should be part of state-of-the-art diabetes care. To address this fully will require not only further research on peer support but also implementation research on how the general organization of care for diabetes as well as other chronic diseases may best provide it.

Finally, the SEM found that male gender was associated with lower HbA1c independent of the influence of peer support. There are likely many reasons for this, but future work might do well to consider how best to provide support to men and, perhaps, how to address impediments affecting women.

Limitations

As described in the Methods, data were drawn from evaluations of several different interventions implemented with varying populations and in disparate settings, albeit sharing a common functional definition of peer support and the evaluation instruments and indicators reported here. Among questions left unanswered is whether the modality of support, e.g., individual, group, face-to-face, telephonic, or digital, or the care delivery system through which it is organized makes a difference in participation or outcomes. The variety and flexible combinations of settings, modalities, and delivery systems among the programs analyzed here left the present analyses unable to examine differences among them. The program in Cambridgeshire, England however employed a factorial design comparing individual, group, individual plus group, and control. For systolic blood pressure, it found greatest impact among those assigned to group or to individual plus group support versus control [19].

It should also be noted that the present analyses included single indicators of two much more complicated concepts: participation and clinical status. Although number of contacts might reasonably be included as part of a characterization of participation, it is far from a comprehensive indicator. Indeed, research on social support [55] includes many findings suggesting that the perceived availability of support may be as important as its exchange, and a recent paper on peer support included varied observations of the importance of simple presence, “being there” [51]. Turning to HbA1c, many in the world of diabetes and its care would argue for the greater importance of other clinical indicators such as systolic blood pressure or of combinations of lipid, blood pressure, and HbA1c measures to characterize clinical status, as well as indicators of quality of life, including depression or other aspects of emotional status [56].

Conclusions

Varying in many details, conducted in diverse settings, and focusing on diverse populations with diabetes, seven randomized controlled evaluations of peer support showed a) greater improvements in HbA1c among those who were offered peer support and completed at least one contact than among those who were offered peer support but completed no contacts, and b) a positive, dose–response relationship between number of contacts with peer supporters and improvements in HbA1c. Among those offered peer support, participation was predicted by older age, baseline depression, and lower level of baseline satisfaction with support for “dealing with diabetes” from family, friends, and the health care team. These predictors reinforced other findings indicating that peer support may reach those who most need it and who too seldom benefit from clinical and prevention interventions [5, 41]. In light of the greater improvements achieved among those who participated the most, a key challenge will be to continue to hone strategies to increase engagement to attain the greatest benefits from peer support.

Funding

Peers for Progress is a program of the Gillings School of Global Public Health at the University of North Carolina-Chapel Hill and receives support from The Merck Foundation, Novo Nordisk, Sanofi China, the Center for Diabetes Translational Research at the University of Michigan (DK092926, Michele E. Heisler & Gretchen Piatt, Co-Directors). The seven projects that provided data for the present analyses were funded through the American Academy of Family Physicians Foundation with support from the Eli Lilly and Company Foundation.

Conflicts of Interest

Andrea Cherrington—Advisory Board, Bayer HealthCare Pharmaceuticals Inc.; Edwin Fisher—Advisory Board, Eli Lilly, Global Director, Peers for Progress, supported by Merck Foundation, Novo Nordisk, Sanofi China; and the American Academy of Family Physicians Foundation with support from the Eli Lilly and Company Foundation; Tricia Tang—Speaker Fees from Insulet Corp, Dexcom. The authors declare that these relationships did not influence or constrain the development, analyses, or description of the projects and results described.

Peers for Progress Projects and Investigators

American Academy of Family Physicians National Research Network, LA Net, WellMed: Lyndee Knox, PhD; Wilson Pace, MD; America Bracho, MD, MPH

CENEXA – Centro de Endocrinologia Experimental y Aplicada, Universidad Nacional de La Plata (UNLP-CONICET), La Plata, Argentina: Juan José Gagliardino, MD; Charles M. Clark, Jr, MD

Centre for Population Studies and Health Promotion, University of Yaounde, Cameroon: Paschal Kum Awah, PhD; Andre-Pascal Kengne, MD, PhD

Department of Medicine and Therapeutics, Hong Kong Institute of Diabetes and Obesity and Li Ka Shing Institute of Health Sciences, The Chinese University of Hong Kong, Prince of Wales Hospital: Juliana C.N. Chan, MBChB, MD, FRCP; Rebecca Y.M. Wong, MSC; Gary T.C. Ko, MD; Roseanne O. Yeung, MD, MPH; Suky Junmei Yin, PhD

Institute of Metabolic Science, Cambridge University Hospitals, University of Cambridge, Cambridge, England: David Simmons, FRACP, FRCP, MD (Cantab); Jonathan P. Graffy, FRCGP, MD

Mahidol University, Bangkok: Boosaba Sanguanprasit, PhD, MPH; Chanuantong Tanasugarn, DrPH, MPH

Monash University and DiabetesVic, Melbourne, Australia: Brian Oldenburg, PhD; Michaela Riddell, PhD

San Diego State University: Guadalupe X. Ayala, PhD, MPH; Andrea Cherrington, MD, MPH; John P. Elder, PhD, MPH; Lucy Horton, MS, MPH; Leticia Ibarrra, MPH

School of Medicine, University of Alabama at Birmingham, Birmingham, Alabama: Monika M. Safford, MD; Andrea Cherrington, MD, MPH; Susan Andreae, MPH

School of Nursing, University of Wisconsin-Madison, Madison, Wisconsin; Mulago Hospital, Kampala, Uganda: Linda C. Baumann, RN, PhD; Fred Nakwagala, MD

University of California, Los Angeles; Women for Peace, Western Cape, South Africa: Mary Jane Rotheram-Borus, PhD; Margaret Gwegwe, BA; Neal Kaufman, MD, MPH

University of California, San Francisco: Thomas Bodenheimer, MD, MPH; David H. Thom, MD, PhD; Ellen Chen, MD; Amireh Ghorob, MPH; Danielle Hessler, PhD

University of Michigan: Michele Heisler, MD, MPA; Tricia S. Tang, PhD; Martha M. Funnell, MS, RN, CDE; Robin Nwankwo, RD, MPH, CDE

University of Yaounde and Central Hospital, Yaounde, Cameroon: Jean Claude Mbanya, MD, PhD, FRCP; Felix K. Assah, MD, PhD