The effect of peer support on HbA1c levels in middle-aged and elderly patients with type 2 diabetes: a systematic review and meta-analysis

Search strategy

This meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Moher et al., 2009). Randomized controlled trials (RCTS) were searched in Cochrane library, PubMed, EMbase, China Biology Medicine Disc (CBM), China National Knowledge Infrastructure (CNKI), Wanfang Database and VIP database by using subject words and keywords, supplemented by manual retrieval. An initial search was performed based on the participants, comparison, intervention, and outcomes (PICO) framework and key terms. We also conducted an additional search by screening the reference lists from the selected literature. The search terms and search strategy are shown in Table S1.

Inclusion and exclusion criteria

In this study, we included RCTs that compared peer support intervention with otherwise similar care in adults with diabetes that measured HbA1c level as a primary or secondary outcome.

The inclusion criteria were randomized controlled trials (RCTs) involving middle-aged and elderly patients with type 2 diabetes mellitus (T2DM), defined as adults aged over 40 years based on definitions of middle age provided by the Encyclopaedia Britannica and Wikipedia. This lower age cut-off was selected to ensure comprehensive inclusion of individuals in the early transitional phase of middle age who are at risk for type 2 diabetes. Participants were also required to be in good mental health (defined as the absence of diagnosed psychiatric disorders or cognitive impairments, as reported by the primary study authors) and capable of self-care. Patients with severe diabetes complications, other serious heart, brain, lung and kidney diseases, such as stroke, myocardial infarction, malignant tumors and gestational diabetes mellitus were excluded. If the control group was given routine nursing intervention, and the experimental group was given peer support intervention (including telephone follow-up, group teaching, peer discussion, etc.), and the diabetes with comprehensive health education knowledge, good communication, and voluntary dissemination of type 2 diabetes health knowledge were selected as peer supporters, we included these studies. All included RCTs reported obtaining ethical approval and informed consent from participants, consistent with standard research practices for randomized controlled trials.

Extraction of data and study selection

Two researchers (JL, HZ) independently screened the title and abstract of the literature, and then further read the literature to extract the correct data if it met the inclusion and exclusion criteria. In the event of disagreements between two researchers, a third party was invited to participate in the discussion to facilitate resolution. The individuals responsible for arbitrating these differences were Jialong Chen and He Zhang. Utilizing their expertise and impartial perspective, they guided the team through a collaborative process aimed at reconciling viewpoints and reaching a consensus. This approach ensured that the decision-making process was robust, transparent, and in line with the best practices of scientific inquiry. The extraction content included: ① Basic information of the included literature: the first author, the title and the time; ② Basic characteristics of the subjects: sample size, age, gender, etc.; ③ Enrollment criteria for peers and training for peers intervention measures; ④ Number of contact hours per session, duration of the intervention; ⑤ Frequency of contact, mode of delivery; ⑥ Control measures; ⑦ Research design; ⑧ Primary outcome measures. Two authors independently (Luo J and Zhang H) extracted data from the included articles into a structured table. Study characteristics are shown in Table S2.

Bias and quality assessment

The quality of the included studies was evaluated by two researchers using the Cochrane Collaboration’s Risk of Bias tool. The assessment tool includes seven items: Methods of randomization (selection bias), concealment of allocation (selection bias), blinding of patients and trial personnel (implementation bias), blinding of outcome assessors (implementation bias), incomplete outcome data (follow-up bias), selective reporting (reporting bias), and other sources of bias were excluded. The risk of bias for each domain was categorized as high, moderate, low (Higgins et al., 2011).

Design

This review was registered with the International Prospective Systematic Review Registry (PROSPERO) under the registration number CRD42023462231. The PRISMA checklist for systematic reviews and meta-analyses is also attached to Table S3 of this review.

Data synthesis and statistical analysis

Our meta-analysis was executed leveraging Review Manager 5.4 and Stata version 17 to ensure rigorous and comprehensive data synthesis. When assessing outcomes standardized on equivalent scales, we calculated the mean difference (MD) supplemented by its 95% confidence interval (95% CI). In instances where studies omitted standard deviations (SDs), we employed established statistical methods: SDs were inferred from the available data, extrapolated from P values and CIs in alignment with the Cochrane Collaboration’s guidelines, or imputed drawing from baseline metrics (Higgins et al., 2024; Wan et al., 2014).

The primary outcome, consistent with the established markers of diabetes management, centered on the variation in HbA1c levels from baseline to the study endpoint. We meticulously computed the mean difference between initial and concluding HbA1c levels for both intervention and control cohorts, ensuring the precision of the accompanying SD for each computed difference. This yielded a pooled MD in HbA1c fluctuations, bolstered by a 95% CI (DerSimonian & Laird, 1986).

To validate the consistency across studies, we harnessed Cochran’s Q and I² statistics. We identified significant heterogeneity at either a P-value threshold of < 0.10 or an I² statistic surpassing 50%. Given the inherent differences in study populations, intervention settings, and peer support modalities across the included studies, we adopted a random-effects model for all analyses. This approach aligns with methodological guidance in the Cochrane Handbook and recent meta-analyses in the field, which recommend using a random-effects model to account for between-study heterogeneity in complex behavioral interventions (Higgins et al., 2024). To address potential publication bias, we constructed funnel plots and conducted Egger’s test. A p-value of less than 0.10 in the Egger’s test was interpreted as indicative of publication bias. Additionally, we performed a leave-one-out sensitivity analysis, sequentially omitting one study at a time and recalculating the pooled mean difference (MD). This approach allowed us to assess whether any single study disproportionately influenced the overall effect size, thereby confirming the robustness of our meta-analytic results (Egger et al., 1997; Godavitarne et al., 2018).

Recognizing the diverse facets that influence diabetes outcomes, we incorporated subgroup analyses. This stratification was rooted in factors demonstrably impactful in diabetes care and research: participant age (categorized as >60 years or ≤60 years), intervention duration (either ≤6 months or extended >6 months), the intervention setting (distinguishing formal medical institutions from informal care settings), the modality of intervention delivery (ranging from individualized sessions, group sessions, to hybrid approaches), and the consultation frequency (quantified as low, moderate, or high based on monthly contacts).

Study selection

Figure 1 shows the PRISMA flowchart of the article selection process. A total of 3,395 references were retrieved, 288 duplicate references were removed, 1,955 references were excluded if their duration did not meet the requirements, 1,053 references were excluded after carefully reviewing the title and abstract, and 88 articles were excluded according to the inclusion and exclusion criteria after reading the full text of the articles during the second screening of the literature. Finally, 11 articles were included in this study (Andreae et al., 2021; Castillo-Hernandez et al., 2021; Chen et al., 2021; Debussche et al., 2018; Ju et al., 2018; Long et al., 2020; Peimani et al., 2018; Pienaar, Reid & Nel, 2021; Presley et al., 2020; Tang et al., 2022; Zhao, Yu & Zhang, 2019).

Quality of the included studies

The results of the risk of bias analysis are shown in Fig. 2. Overall, of the 11 included studies, eight studies adequately described randomization sequence generation. Six studies did not describe allocation concealment, whereas six studies did not explain the blinding of personnel and participants, and one of these studies described a high risk of bias. Six studies did not describe blinding to the outcome assessment. There was a low risk of attrition bias due to incomplete outcome data in only one study. All 11 studies had a low risk of bias for selective reporting and other bias. With respect to statistical power, most studies reported formal power and sample size calculations, typically using an 80% power threshold. However, three studies—Chen et al. (2021), Presley et al. (2020), and Zhao, Yu & Zhang (2019)—did not report power or sample size calculations. The overall evidence was originally evaluated as high quality. However, owing to the heterogeneity of the study findings, the quality of evidence was downgraded from high to moderate.

Impact of peer support on HbA1c levels

Figure 3 shows the effect size and 95% CIs of the included studies. Eleven studies evaluated the effects of peer support teaching and routine nursing education on HbAlc levels in patients with T2DM. Because of the high heterogeneity of the studies, a random effects model was employed to evaluate possible differences in HbA1c levels between the control and intervention groups. The overall pooled effect of peer support interventions on HbA1c levels was an MD of −0.20 (95% CI [−0.37 to −0.02]; p = 0.03), and it was statistically significant, favoring peer support over usual care. High statistical heterogeneity was found among the studies in terms of changes in HbA1c levels (I² = 47%, p = 0.04).

Subgroup analyses

Subgroup analyses were executed post hoc to discern potential variations attributed to the distinct characteristics intrinsic to each study, with specific findings detailed in Table 1. (See Figs. S1, S2, S3, S4, and S5).

Our findings delineate differential responses in HbA1c reduction contingent upon participant age. Compared with their younger counterparts (MD −0.20 [−0.37 to −0.02]), participants older than 60 years exhibited a more pronounced reduction (MD −0.30 [−0.85 to 0.25]) in HbA1c levels.

The institutional context of intervention delivery was a significant determinant of outcome variability. Interventions administered within formal medical institutions were associated with an amplified reduction in HbA1c levels (MD −0.32 [−0.72 to 0.08]) relative to those orchestrated in informal settings (MD −0.16 [−0.30 to −0.01]).

The modality of educational delivery also affected HbA1c outcomes. Participants who engaged in group sessions or experienced a hybrid of group and individual sessions manifested a discernible decline in HbA1c levels (MD −0.37 [−0.67 to −0.06] and MD −0.17 [−0.60 to 0.26], respectively). In contrast, exclusively individualized educational interventions did not yield a statistically significant decrease (MD −0.12 [−0.33 to 0.09]).

With regard to contact frequency, higher-frequency contacts were more efficacious (MD −0.32 [−0.55 to −0.1]) in mitigating HbA1c levels than moderate-frequency contacts were (MD −0.03 [−0.18 to 0.12]).

In evaluating the temporal dimension of interventions, our analysis revealed no substantial differential outcomes between shorter-term interventions (≤6 months, MD −0.18 [−0.47 to 0.1]) and their longer-term (>6 months) counterparts (MD −0.22 [−0.45 to 0.02]) in terms of comparative efficacy between the peer support cohort and the control cohort.

Publication bias and sensitivity analyses

We explored the possibility of publication bias for the 11 included studies. Sensitivity analyses were performed to test the robustness of the results. In the sensitivity analyses, the pooled results for HbA1c levels did not significantly change after any single study was excluded. There was no publication bias based on Egger’s test (p = 0.24) or the shape of the funnel plot (Fig. S6).