Effects of different exercise types on vascular endothelial function in individuals with abnormal glycaemic control: a systematic review and network meta-analysis

Registration

This systematic review and NMA were performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Hutton et al., 2015). The study protocol has been registered in the International Prospective Register of Systematic Reviews under ID: CRD42025643867.

Literature search strategy

A comprehensive literature search was performed across multiple databases, including PubMed, Web of Science, Cochrane Library, Embase, and EBSCO, to identify randomized controlled trial (RCT) examining the impact of exercise on brachial artery FMD in individuals with impaired glycemic control, specifically those with prediabetes or T2DM. The search period spanned from the inception of each database to January 2025. In addition, Google Scholar was searched, and the reference lists of pertinent reviews and meta-analyses were manually screened for additional relevant studies. The search utilized Medical Subject Headings (MeSH) terms in PubMed and Cochrane Library, as well as Emtree terms in Embase. The strategy involved the use of key search terms, including “Exercise” OR “Type 2 Diabetes Mellitus” AND “Endothelial Function.” Detailed information on the full search strategy can be found in Appendix S1.

Eligibility criteria

Eligibility criteria were established based on the Population, Intervention, Comparator, Outcomes, and Study (PICOS) framework (Amir-Behghadami & Janati, 2020).

Inclusion criteria

1. Population: Adults aged 18 years or older with abnormal glycemic control, specifically those with prediabetes or T2DM. Prediabetes was defined as meeting at least one of the following criteria for baseline glucose parameters: fasting glucose between 5.6 and 6.9 mmol/L, 2-hour glucose between 7.8 and 11.0 mmol/L, or HbA1c between 5.7% and 6.4% (American Diabetes Association Professional Practice Committee, 2022). T2DM was defined by at least one of the following criteria: fasting glucose ≥7.0 mmol/L, 2-hour glucose ≥11.1 mmol/L, HbA1c ≥6.5%, or random plasma glucose ≥11.1 mmol/L (American Diabetes Association Professional Practice Committee, 2022).

2. Intervention: Exercise, defined as planned, structured, and repetitive physical activity aimed at improving or maintaining physical fitness (Caspersen, Powell & Christenson, 1985). This includes interventions such as aerobic continuous exercise (ACE), aerobic interval exercise (AIE), resistance exercise (RE), mind-body exercise (MBE), and combined exercise (CE), as outlined in Table 1 (Martínez-Vizcaíno et al., 2022). The experimental group underwent structured exercise training for at least 6 weeks.

3. Comparator: Comparisons could be made between the exercise group and a control group, or among different exercise modalities. The control group adhered to their routine physical activities, received health education, and performed stretching exercises, or maintained their usual daily lifestyle, while the experimental group underwent a structured exercise training program in addition to the control regimen.

4. The sole outcome measure in this review was FMD, which serves as a marker of endothelial function and vascular health. Only studies that assessed endothelial function through brachial artery FMD were included. The FMD measurement was performed by calculating the absolute diameter change (peak diameter–baseline diameter), with subsequent conversion to percentage values using the standard equation: %FMD = (absolute FMD/baseline diameter) ×100.

5. Study design: Only RCT, including cluster-randomized and crossover designs, were included.

Exclusion criteria

Studies were excluded if they met any of the following criteria:

1. Focused on pregnant women, individuals with type 1 diabetes, or patients with acute medical conditions (e.g., those presenting in emergency departments).

2. Were duplicates, literature reviews, letters to the editor, conference abstracts, animal studies, or low-quality literature.

3. Had unavailable full-text articles or incomplete study data.

4. Did not report vascular function outcomes relevant to this review.

5. Combined exercise interventions with diet, insulin, or other pharmacological treatments.

Study selection and data extraction

The process of screening titles, abstracts, and full texts was carried out independently by two authors (Zhongxiang Li and Shengyao Luo), following the established inclusion and exclusion criteria. Any discrepancies were resolved through discussions with Dan Wang.

The data extracted from the included studies comprised several key pieces of information: the first author, year of publication, and participant demographics (including the number of individuals in both experimental and control groups, age, and gender). Information on the exercise intervention was also recorded, which included details such as the type of exercise, its intensity, duration, frequency, and session length. Brachial artery FMD outcomes were the primary measure of interest. In instances where certain data were missing, the corresponding authors were contacted via email in an effort to retrieve the required information.

Risk of bias assessment

The risk of bias was evaluated using version 2 of the Cochrane Risk of Bias (RoB 2) tool, which introduces several important advancements for assessing bias in randomized trials (Sterne et al., 2019). This tool encompasses five key domains: (1) the randomization process, (2) deviations from intended interventions, particularly allocation concealment, (3) handling of missing outcome data, (4) outcome measurement, and (5) selection of reported results. Using robvis tool to create risk-of-bias plots (McGuinness & Higgins, 2021).

Data synthesis and statistical analysis

The effect was estimated in this NMA by combining the changes observed before and after the intervention in both the experimental and control groups. The calculation method of the standard deviation (SD) of the change was based on the formula provided by the Cochrane Handbook for Systematic Reviews of Interventions (version 6.3) (Higgins et al., 2022) (the formula is: SD_change=$\sqrt{S{D}_{baseline}^2+S{D}_{final}^2-\left(2\times Corr\times S{D}_{baseline}\times S{D}_{final}\right)}$). In this formulate, a correlation coefficient (Corr) of 0.5 was assumed, which is commonly used when the actual correlation is not reported in the original studies (Follmann et al., 1992). The sensitivity of results to this assumption was tested by varying Corr between 0.3 and 0.7 in a sensitivity analysis (see Appendix S2).

Data were analyzed using META through Stata 16.0 software (StataCorp, College Station, TX, USA), with outcome indicators treated as continuous variables. For brachial artery FMD, mean difference (MD) and their corresponding 95% confidence intervals (CI) were calculated, with baseline adjustments made at α = 0.05. Heterogeneity was assessed using the Q-test and I² statistic. Publication bias was assessed using Egger’s test and funnel plots (including a network funnel plot for NMA), with symmetry as the evaluation criterion. A random-effects multivariate NMA was conducted using a frequency framework with STATA 16.0 software (Salanti, 2012), following the most recent PRISMA NMA guidelines (Hutton et al., 2015). A network diagram was created to visually depict the relationships between different intervention strategies. The lines connecting the nodes represent direct comparisons between the interventions, with the size of the nodes and the thickness of the lines reflecting the number of studies. Furthermore, a network contribution plot was generated to evaluate the contributions of each direct comparison.

Transitivity assumptions were evaluated through the inclusion criteria of individual studies, by assessing whether all participants in the network could be randomly assigned to any intervention, and by applying consistency models (Rücker & Schwarzer, 2015). As a key assumption in NMA, transitivity asserts that indirect comparisons are valid approximations of unobserved direct comparisons (Salanti, 2012) and that effect modifiers are uniformly distributed across studies (Jansen & Naci, 2013). To evaluate consistency in each closed loop, inconsistency factors (IFs) with 95% CIs were calculated, with consistency indicated when the lower limit of the 95% CI equals 0 (Chaimani et al., 2013). The inconsistency model was applied to test for inconsistency, while the consistency model was used when no significant inconsistency was detected (P > 0.05) (Shim et al., 2017). Node-splitting analysis was conducted to assess local inconsistency, and the results were deemed reliable with P > 0.05. The surface under the cumulative ranking curve (SUCRA) was applied to rank and compare the effectiveness of various exercise interventions (Salanti, Ades & Ioannidis, 2011). SUCRA values range from 0 to 100, with higher values indicating more favorable effects of exercise interventions (Mbuagbaw et al., 2017).

Literature selection

Figure 1 outlines the process of literature search and selection. A total of 2,320 articles were initially retrieved from five databases, with an additional 25 articles identified through Google Scholar. In addition, 13 relevant articles were sourced from reference lists. After removing 737 duplicates, 1,621 articles remained for the screening phase. Of these, 1,556 were excluded based on the review of titles and abstracts. Following a full-text review, 48 additional articles were excluded, resulting in 17 articles that were ultimately included in the NMA.

Literature characteristics

The characteristics of the studies included in this research are detailed in Appendix S3. A total of 797 individuals with abnormal blood glucose control were included in this study (Prediabetes: 76; T2DM: 721). The average age ranged from 33 to 70.5 years. Four studies (Abdi, Tadibi & Sheikholeslami-Vatani, 2021; Choi et al., 2012; Ku et al., 2009; Kwon et al., 2011) exclusively included female T2DM patients, while Liu, Li & Liu (2007) did not specify the gender of the participants. In the study by Gainey et al. (2016), 81.8%–83.3% of the participants were female, whereas in Desch et al. (2010), only 22%–34% of the participants were female. In the remaining studies, the gender ratio of male to female participants was approximately 1:1. Based on the type of exercise intervention, 186 individuals were assigned to ACE, 63 to AIE, 30 to RE, 109 to CE, 52 to MBE, and 357 to control group (CON). The exercise intensity in the ACE group was relatively consistent, with a range of approximately 50%–75% HRmax. The intensity in the AIE group also showed little variation, with exercise intensity ranging from 80%–95% HRmax, and active rest ranging from 50%–60% HRmax. The exercise intensity in the RE, CE, and MBE groups was similarly consistent. With the exception of Rech et al. (2019) which involved light stretching, no other studies in the control group included exercise interventions. The duration of exercise interventions in all 17 studies was ≥8 weeks, with over half (70.59%) of the studies having an intervention duration of 12 weeks. All 17 studies included supervised training.

Risk of bias assessment results

The results of the bias risk assessment are provided in Appendix S4. In the D1 (randomization process) domain, 14 studies were assessed as having a low risk of bias. However, three studies described the use of random allocation methods but failed to provide specific details on the random sequence generation process. In the D2 (deviation from intended interventions) domain, all studies adhered to the intended interventions, resulting in a low risk of deviation. In the D3 (missing outcome data) domain, six studies had unclear reporting of missing data, indicating a potential risk of bias. Additionally, a dropout rate exceeding 20% was considered high risk, and three studies were classified as high risk based on this criterion (Chen et al., 2024). In the D4 (outcome measurement) domain, all studies employed standardized measurement methods, resulting in a low risk of bias. Finally, in the D5 (selection of reported results) domain, 12 studies raised some concerns due to the lack of mention of RCT protocol registration.

Network meta-analysis

Figure 2 illustrates the NMA plots of eligible studies assessing the impact of various exercise categories on brachial artery FMD. The size of the nodes corresponds to the sample size within each exercise category, while the thickness of the lines connecting exercise types reflects the number of studies evaluating that particular comparison. The most frequently utilized intervention was ACE, whereas MBE was the least common.

Appendix S5 outlines the contributions of direct and indirect comparisons to the NMA, as well as the number of studies for each direct comparison. The consistency of brachial artery FMD was evaluated using loop-specific heterogeneity estimates, the inconsistency model, and node-splitting analysis (Appendix S6). Loop-specific heterogeneity estimates revealed high consistency within each closed loop for brachial artery FMD. The inconsistency model showed no significant discrepancies between direct and indirect comparisons (P > 0.05). Furthermore, node-splitting analysis indicated no significant differences between direct and indirect evidence (P > 0.05), supporting the reliability of the results.

Forest plots for the eligible comparisons of brachial artery FMD, along with their 95% CI, are provided in Appendix S7. Seventeen studies assessed brachial artery FMD (n = 797) (Abdi, Tadibi & Sheikholeslami-Vatani, 2021; Afousi et al., 2018; Choi et al., 2012; Cox et al., 2024; Davoodi et al., 2022; Desch et al., 2010; Gainey et al., 2016; Gibbs et al., 2012; Ku et al., 2009; Kwon et al., 2011; Li, 2009; Liu, Li & Liu, 2007; Mitranun et al., 2014; Okada et al., 2010; Ploydang et al., 2023; Rech et al., 2019; Wycherley et al., 2008). The pooled estimates from the NMA of brachial artery FMD are presented in Table 2. AIE (MD = 2.23%, 95% CI [1.09%–3.37%], MBE (MD = 1.97%, 95% CI [0.60%–3.33%], P < 0.05), CE (MD = 1.17%, 95% CI [0.13%–2.21%], P < 0.05), and ACE (MD = 1.20%, 95% CI [0.52%–1.87%], P < 0.05), P < 0.05) all demonstrated significant improvements in brachial artery FMD compared to the control group. Table 3 and Appendix S8 present the SUCRA values for all interventions, showing that AIE (SUCRA = 89.0), MBE (SUCRA = 80.1), CE (SUCRA = 51.0), ACE (SUCRA = 50.9), and RE (SUCRA = 20.1) were all more effective than the control group (SUCRA = 9.2) in improving brachial artery FMD in individuals with impaired glycaemic control.

Appendix S9 presents the funnel plots for brachial artery FMD, which were used to assess potential publication bias and small-sample effects in the NMA. The results showed that a limited number of data points displayed scattered distributions, which may be attributed to small sample sizes or other underlying biases. Egger’s test further indicated significant publication bias (P = 0.0045). In sensitivity analysis, after excluding (Wycherley et al., 2008; Desch et al., 2010), the result of Egger’s test was no longer significant (P = 0.0726). The SUCRA rankings were as follows: AIE (88.9), MBE (88.1), ACE (54.5), CE (41.9), RE (21.3), and CON (6.4), indicating the robustness of the NMA results.