Peer Support Interventions for Adults With Diabetes: A Meta-Analysis of Hemoglobin A_1c Outcomes

We truly appreciate the detailed comments on our article.

Our goal was to assess the effect of peer support interventions on HbA1C.[1]

To preserve construct validity of our research question, we included studies in which the only difference between intervention and control group was a peer-delivered intervention. We defined baseline care as "otherwise similar care except for the peer-delivered interventions." Nothing more could be done to improve the internal validity when answering our primary research question.

Addressing the concern about the Long et. al. 2012 paper, the results of this study are not an outlier in our analysis and do not skew results. Standard residuals for individual studies showed that Long et al. had a standard residual of 1.98 whereas Dale et al. had standard residual of -2.08 in our analysis. Analysis after omitting the Long et al. study changes SMD from 0.121 to 0.100 and continues to be statistically significant.

We focused on pertinent results and discussions to keep our article word count acceptable. It does seem logical that higher baseline HbA1C will lead to larger effect sizes. We provided the scatter plots in appendices for further reader interpretations. Cost-effectiveness could not be assessed based on available data.

Most medications reduce HbA1C by 1% while DPP4 inhibitors reduce HbA1C by approximately 0.5%.[2] We acknowledged that the effect size of our meta-analysis is small, however, keeping in mind the pooled baseline HbA1C was 8%. American Diabetes Association (ADA) recommends target HbA1c of <7% with goals of <8% for certain patients with comorbidities, medication adverse effects and long-standing diabetes where general goal is difficult to attain.[3] Further studies in patients with higher baseline HbA1C are needed to better understand the effect of peer support interventions on HbA1C improvement.

Our results have an I2 of 60.66% which describes the percentage of the variability in effect estimates that is due to heterogeneity rather than chance.[4] Studies brought together in a systematic review will always differ in some characteristics such as participant and intervention/control characteristics. Systematic reviews may compare varying doses of various medications from the same class of drugs with another class of drugs.[5] Variations in interventions will be more common for interventions targeting self-management or behavioral changes.[6] We attempted to reduce the methodological diversity by including RCTs. We utilized a concrete definition of peer support excluding studies where peer support was provided by community health workers or bilingual clinic employees. The only difference between groups in each study was the peer-delivered intervention which was consistent with our research questions.

References

1. Patil SJ, Ruppar T, Koopman RJ, et al. Peer Support Interventions for Adults With Diabetes: A Meta-Analysis of Hemoglobin A1c Outcomes. Ann Fam Med. 2016;14(6):540-551.
2. Bennett WL, Maruthur NM, Singh S, et al. Comparative effectiveness and safety of medications for type 2 diabetes: an update including new drugs and 2-drug combinations. Ann Intern Med. 2011;154(9):602-613.
3. Standards of Medical Care in Diabetes--2017: Summary of Revisions. Diabetes Care. 2016;40(Supplement 1):S4.
4. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327(7414):557-560.
5. Kew KM, Evans DJ, Allison DE, Boyter AC. Long-acting muscarinic antagonists (LAMA) added to inhaled corticosteroids (ICS) versus addition of long-acting beta2-agonists (LABA) for adults with asthma. Cochrane Database Syst Rev. 2015;(6).
6. Foster G, Taylor SJ, Eldridge S, Ramsay J, Griffiths CJ. Self-management education programmes by lay leaders for people with chronic conditions. Cochrane Database Syst Rev. 2007;(4).

Competing interests: None declared

There is a growing body of literature that suggests diabetes self- management education can improve HbA1c levels. In 2008, the World Health Organization (WHO) reported that peer support is a promising approach to improving and sustaining diabetes self-management behaviors. However, RCTs using peer support interventions have shown mixed results. The aim of this 2016 meta-analysis was to study the effectiveness of peer support interventions on improving glycemic control (HbA1c) in adults with diabetes, as compared to counterparts who received otherwise similar care.

This study uses the AAFP Peers for Progress definition of peer support: any peer-delivered interventions designed to improve self- management or health behaviors, or to provide emotional or social support with the goal of improving overall health. This definition started an interesting discussion: too often doctor's schedules are over-booked, leaving little time for in-depth discussions between the doctor and patient about education and self-management. In recent years there has been a push to establish a multi-disciplinary health profession-centered teams to aid in delivery of health interventions and education. By focusing on peer-delivered education, this study helped point out an obvious but salient fact: expanding the multi-disciplinary team to include non-health professionals. This addition can help augment the effects of self-management education on HbA1c levels and can help sustain the benefits in the long-term. Inclusion criteria for this paper included studies in which both the intervention group and the control group received similar baseline education care from health care professionals. However, the authors do not define "baseline care". Studies ranged from US to Europe, Canada, China to Argentina. Baseline care may be different across the studies both within the US and internationally skewing the internal validity of this paper, however, the international and comprehensive nature of the study gives strength to the external validity.

After searching in multiple databases for both English and non- English articles published from 1960 to 2015, over 400 citations were collected. Ultimately, only 17 articles were eligible for inclusion in the meta-analysis (14 RCT, and 3 cluster RCTs). In the results section, the authors report that with a random effects model, the pooled effect of peer -support interventions on HbA1c levels resulted in a standard mean difference (SMD) of .121. This translates to an improvement in HbA1c levels of .24% (95% CI, 0.026-.217; P=.01). When they did a sensitivity analysis, excluding the 3 cluster RCTs, there was a small effect of .27%. While this effect was statistically significant, the clinical significance is questionable.

The discussion focused on the data presented in Figure 2, where the effects of peer-intervention on HbA1c were reported. There seems to be a single data point that skews the SMD, the Long et. al. 2012 paper. In this study authors reported a standard mean difference of .661, which was nearly double the highest difference recorded in the other articles. It was also the only paper studied in which the change from baseline was >0.5. The focus of the paper was on peer mentoring and financial incentives on improving glucose control in African-American veterans. This raised some very interesting questions: are financial motivations a more robust method for incentivizing glycemic control? What is unique about African Americans that allowed for such a large change from the baseline? Conversely, was it the fact that the group was composed of veterans who forged a unique bond through the military that facilitated this peer- support mediated change? None of these were touched on by the authors, and there was no comment made about this outlier.

These questions are not trivial, as they can offer insights on sensitive and appropriate means to deliver health-education, especially in light of the sub-group analysis done in the paper. The authors report that the pooled effect of peer-support interventions, in a subgroup of 5 studies with predominantly Hispanic participants, showed a statistically and clinically significant change in baseline HbA1c levels. A subgroup analysis of 3 papers with predominantly African-American participants, showed a similar effect on baseline HbA1c as seen in the Hispanic studies, but it was not statistically significant. This is reported in stark contrast to the effect of peer-support in a subgroup of 9 studies with predominantly non-minority patients, as the pooled effect was shockingly negative, -0.002. The between-group difference (between minority studies and non-minority studies) was statistically significant with a p-value of .001. These results suggest that there is a statistically significant difference in the effect of peer-support on changing HbA1c in minority vs non-minority groups. What is the cause of this difference? Why are minorities groups more receptive to peer-support driven interventions, than non-minority groups? The authors suggest that glycemic control and knowledge of diabetes are improved when culturally appropriate health education is provided to people with diabetes who belong to ethnic minority groups. As providers, this fact is extremely important. While health-literacy remains a focus of many public health initiatives, only about 12% of adults have proficient health-literacy. There is a huge potential for benefit if the medical community can combine peer-support and culturally appropriate health-education. While non-minority patients do not respond as well to peer-support groups, minority patients do, forcing different management decisions based on ethnicity and race. Non- minority groups may respond more favorably to health-professional driven interventions, while this study shows minority groups respond much better to peer-support driven interventions

Another part of the paper that sparked intense debate was concerning the self-efficacy outcomes. The results showed no interaction between the duration of peer training (in hours) andmean participant baseline HbA1c level, or duration of observation and change in HbA1c levels between the intervention and control group. This was surprising. While studying the data in the appendix, it was clear that there was no effect of duration of peer training (in hours) on SDM, as the slope of the regression analysis was nearly perfectly 0. However, there seemed to be an interaction between baseline HbA1c and SDM, as the slope of the regression analysis was positive (.657). While this was not clinically significant (p=0.09), more information on this topic was desired. Change in baseline HbA1c levels should be theoretically easier for those with high baseline HbA1c, as there is more potential for change. A more detailed explanation for the phenomena was not provided. Lastly, we investigated the author's report that there was no effect on duration of observation and SDM. However, from the graph it is clear that there are far more papers that observed the patients for 1 month, as opposed to only three studies that observed patients for longer than one month.

This was a very informative paper, but there were limitations and biases. First, this was a meta-analysis with weak internal validity, due to the heterogeneity of papers studied. However, it is the first step in studying the effect of peer health education and provides a guideline for future research in this field. Another limitation is the modest change in HbA1c observed after peer-support interventions. A change of <1% in HbA1c is not clinically significant. Moreover, the effect may have been skewed by one outlier (Long 2012). Additional limitations included the fact that there was no control for the type of diabetes, or for the type of medication regimens used. This is important, as the educational materials and methods of peer-support interventions are completely different based on type of diabetes, control goals and medications used or previously failed on. One last limitation was that there was no comment on cost-effectiveness of peer-support interventions.

In conclusion, this paper was important for a few reasons. It highlighted the importance of support groups for health education, and brought a new type of intervention to the forefront: peer-support groups. The study also showed that the effect of peer-support intervention on HbA1c lowering was more profound in minority groups, specifically Hispanic minorities. Due to the negligible effect on A1c, more research must be done before our group would feel comfortable prescribing this to patients. There is a need for studies with long term follow-up, as this was absent in the selection criteria. Long-term studies in specific subgroup could have a larger and more clinically significant effect on A1c levels.

Competing interests: None declared

We truly appreciate the expert comments and insights from Dr. David H. Thom. We did plan to study the effect of the 'dose' of peer support on HbA1C improvements; however, the number and intensity of contacts was reported in diverse formats across studies hence could not be summarized into a common metric for quantitative analysis. Therefore, we decided to include the information on the number and intensity of contacts as an 'attendance rates' column in the table of study descriptions (Table 1).[1]

We are likewise enthusiastic about the findings of no interaction of peer supporter training duration on improvements in HbA1C. We hope this finding will encourage peer support program implementation in low-resource settings.

REFERENCES

1. Patil SJ, Ruppar T, Koopman RJ et al. Peer Support Interventions for Adults With Diabetes: A Meta-Analysis of Hemoglobin A1c Outcomes. The Annals of Family Medicine. 2016;14(6):540-551.

Competing interests: None declared