Clinical outcomes and glycaemic responses to different aerobic exercise training intensities in type II diabetes: a systematic review and meta-analysis

Studies were identified through a MEDLINE search strategy (1985 to Sept 1, 2016), Cochrane Controlled Trials Registry (1966 to Sept 1, 2016), CINAHL, SPORTDiscus and Science Citation Index. The search strategy included a mix of MeSH and free text terms for the key concepts related to exercise training, high intensity interval exercise, peak VO₂, type II diabetes, glycaemic control, insulin sensitivity and insulin resistance for clinical trials of exercise training in people with type II diabetes. Studies were included if patients exhibited a diagnosis of type II diabetes. Searches were limited to prospective randomized, controlled trials of aerobic exercise training in humans, aged >18 years, lasting 2 weeks or more, supervised and unsupervised program were included. No restrictions were placed on the year, or language, of publication. Reference lists of papers and latest editions of relevant journals which were not available online were scrutinised for new references. Full articles were read and assessed by three reviewers (NS, AG and EC) for relevance and study eligibility. Disagreements on methodology were resolved by discussion, a fourth reviewer (FG) adjudicated over any disputes. Study authors were contacted and requested to provide further data if required.

Included studies were randomized controlled trials of exercise training, of 6 weeks or longer, supervised or unsupervised, in people with type II diabetes. Although some outcomes such as HbA1C% are unlikely to change in less than 12 weeks, peak VO₂ etc. will change in this time period [11, 12]. Studies of type I diabetes were excluded to maintain a homogeneous study population and the type I patients may have made up disproportionately fewer total patients. Resistance training studies were excluded as they tend not to use or measure peak VO₂ as an outcome measure and there are too few isolated resistance studies to warrant an analysis. As a result resistance study data were not included in our analysis. All published studies included in this systematic review were comparisons between exercise study groups and control. Reviewers categorized the studies into four groups based on exercise intensity. The categorization were based on the position stand by Exercise and Sport Science Australia [10]. The measures used to classify exercise intensity were percentage of Heart Rate Maximum (%HR_max), Heart Rate Reserve (%HRR), Peak Oxygen Uptake (%VO₂ Peak) and Borg scale score [10].

In addition to the records identified through database searching, reference lists of identified records were scrutinized. Only the principal study with the greatest number of subjects were included where multiple publications existed from the same dataset. After initial screening over-lapping, duplicates, duplicate data and irrelevant articles such as editorials and discussion papers that did not match the inclusion criteria were removed. We excluded studies where the control group received additional intervention, non-relevant studies; studies using non-aerobic exercise training and those reporting only acute exercise testing responses. We excluded studies from specific analyses if incomplete data was reported and the authors did not respond to our requests to provide missing data.

We recorded the following data; percentage change in HbA1C%, Homeostatic model assessment of insulin resistance (HOMA-IR), lean body mass, BMI, body composition, peak VO₂ (only where this was measured directly during peak exercise), fasting glucose and insulin at baseline and post exercise. We also recorded exercise training frequency, intensity, duration per-session, length of exercise program.

From extracted data we calculated patient-hours of exercise training, percentage change in outcome measures.

We assessed study quality with regard to: eligibility criteria specified, random allocation of participants, allocation concealed, similarity groups at baseline, assessors blinded, outcome measures assessed in 85% of participants and intention to treat analysis. The study quality was assessed according to the validated Tool for the assEssment of Study qualiTy and reporting in EXercise (TESTEX). TESTEX is a study quality and reporting assessment tool, designed specifically for use in exercise training studies. which has a maximum score of 15 [13]. The main point of difference in TESTEX is that there are accommodations for: Activity monitoring in control groups to measure crossover to exercise by sedentary control patients; Assessment of the existence and method of activity monitoring in both exercise intervention and sedentary controls; Assessment of whether the relative exercise intensity remained constant and therefore potentially avoided de-training as participants initially adapt to new exercise programs; Assessment of whether periodic evidence-based adjustment of exercise intensity is reported exercise volume and exercise expenditure Information on all exercise characteristics (intensity, duration, frequency and mode) is provided to calculate exercise volume and exercise energy expenditure.

This tool is a 15-point scale (5 points for study quality and 10 points for reporting) and addresses previously unmentioned quality assessment criteria specific to exercise training studies. Two reviewers NS and FG conducted the risk of bias assessment, PG was consulted of discrepancies occurred. No minimum TESTEX score was required for a study to be included in the analysis.

A mixed-effects meta-analysis model was used to estimate the effect of exercise versus control for the outcomes of interest while controlling for the repeated measures arising in multi-arm studies and over time for the same study. The metafor [14] package within R statistical software was used to conduct the meta-analyses and create forest plots [15].

Continuous outcomes were reported as mean difference between exercise and control and/or relative change from baseline scores depending on the data extracted with 95% confidence intervals (CI).

We used a 5% level of significance to report differences between intensity and control in each of the outcome measures. Egger bias tests were employed if at least 10 studies were included for a given outcome [16]. Heterogeneity is presented as the estimated between studies variability, τ² with 95% CI. A multivariate generalization of the I² statistic is also provided.

Twenty intervention groups provided data on HbA1C%. Results indicated that there was a significant reduction in the exercise versus control groups HbA1C%, MD: −0.69%, 95% CI −1.09, −0.30; p value = 0.0005 (Fig. 2). Cochran’s Q-test indicates significant heterogeneity between studies (Q = 440, df = 28, p value <0.0001; between studies variability: τ² = 0.61, 95% CI 0.31, 1.36; I² = 89.8%). An Egger bias test indicated no evidence of funnel plot asymmetry (p value = 0.710). There was one significant moderator effect; exercise program duration (weeks). For every additional week of follow-up HbA1C% reduces between 0.009 and 0.043%, p = 0.002.

Seven studies provided data on HOMA-IR. Results indicated a significant improvement in HOMA-IR in exercise participants versus control MD: −1.02, 95% CI −1.77, −0.28; p value = 0.007 (Fig. 3). Cochran’s Q-test indicated significant heterogeneity between studies (Q = 113, df = 11, p value <0.0001; between studies variability: τ² = 0.72, 95% CI 0.22, 3.54; I² = 83.7%). There were too few studies to allow investigation of moderator effects or to perform an Egger bias test.

Eight intervention groups provided data on insulin. Results suggested a significant reduction in serum insulin in the exercise participants versus control MD: −10.39 IU, 95% CI −17.25, −3.53; p value = 0.003 (see Additional file 1). Cochran’s Q-test indicates significant heterogeneity between studies (Q = 304, df = 10, p value <0.0001; between studies variability: τ² = 136.8, 95% CI 41.0, 646.0; I² = 94.5%). There were too few studies to allow investigation of moderator effects or to perform an Egger bias test.

Eighteen studies provide data on fasting blood glucose. Results indicated that there was a significant reduction in serum glucose in exercise participants versus control MD: −12.53 g.DL⁻¹, 95% CI −18.94, −6.23; p value <0.0001 (see Additional file 1). Cochran’s Q-test indicates significant heterogeneity between studies (Q = 522, df = 23, p value <0.0001; between studies variability: τ² = 136.1, 95% CI 56.8, 361.9; I² = 94.2%). An Egger bias test indicated no evidence of funnel plot asymmetry (p value = 0.862). There were no significant moderator effects.

Nineteen intervention groups were analysed for body mass index (BMI) which suggested significantly reduced BMI in exercise versus control participants MD—1.56 kg m^−2, 95% CI −2.41, −0.71; p value = 0.0003 (see Additional file 1). Cochran’s Q-test showed no evidence of heterogeneity between studies (Q = 27, df = 26, p value = 0.39; between studies variability: τ² = 0.78. 95% CI 0.00; 4.71; I² = 36.4%). An Egger bias test showed no evidence of funnel plot asymmetry (p value = 0.884). There were no significant moderator effects.

Six studies were included in the analysis. Results for lean body mass indicated no difference between exercise and control groups MD: −0.44 kg 95% CI −1.19, 0.31; p value = 0.246 (see Additional file 1). Cochran’s Q-test indicates no evidence of heterogeneity between studies (Q = 1.28, df = 7, p value = 0.991; between study variability: τ² = 0.00, 95% CI 0.00, 8.00; I² = 0.00%). There were too few studies to test for moderator effects or funnel plot asymmetry.

Six intervention groups provided data on fat mass. Results indicate that there was no difference between exercise and control MD: −0.47 kg 95% CI −1.54, 0.61; p value = 0.396 (see Additional file 1). Cochran’s Q-test indicates no evidence of heterogeneity between studies (Q = 5, df = 7, p value = 0.703; between studies variability: τ² = 0.00, 95% CI 0.00, 10.0; I² = 0.00%). There were too few studies to be able to test for moderator effects or funnel plot asymmetry.

Twelve intervention groups provided data on peak VO₂ (ml/kg/min). Results show a significant improvement in exercise participants versus control MD: 3.40 ml/kg/min, 95% CI 1.65, 5.15; p value = 0.0001 (see Fig. 4). Cochran’s Q-test indicates significant heterogeneity between studies (Q = 79, df = 13, p value <0.0001; between studies variability: τ² = 6.61, 95% CI 1.72, 23.2; I² = 85.3%). There was no evidence of funnel plot asymmetry (Egger bias test p value = 0.815). There was a significant moderator effect; participants undertaking vigorous exercise had significantly higher peak VO₂ max (between 0.64 and 5.98 ml/kg/min more) than participants undertaking low/moderate exercise (p = 0.015).

Median TESTEX score was 7. Several aspects of study design were conducted poorly on more than 50% (13) studies; the method of randomization was only clearly stated in 9/27 studies; group allocation was only concealed from assessors in 5 studies; assessor blinding was only employed in 4 studies; intention to treat analysis was only done in 3 studies; physical activity monitoring of controls was only performed in 1 study; exercise intensity was periodically reviewed in only 5 studies (see Additional file 1: Table S2).

Most of the meta-analyses exhibited heterogeneity that was not substantially reduced through a systematic attempt to identify reasons for heterogeneity by grouping studies according to similarities in outcome reporting or meta-regression.

The exercise training programs varied greatly between studies with respect to exercise intensity, duration, frequency and modality. We accounted for duration via the statistical model used and the other variables via meta-regression however, for the most part the meta-regressions were not significant. Despite the heterogeneous study designs, the Egger bias tests suggested no evidence of funnel plot asymmetry suggesting minimal risk of publication bias.

Few included studies accurately quantified the volume of incidental and structured physical activity, this has been performed previously in people with T1DM [52].

Measures of lean, and fat, body mass would have shed more light onto the role that body composition plays in improving glycaemic control through exercise. We would like to have conducted more moderator variable analyses but limited extracted data precluded this. We were only able to consider program duration, and high/vigorous versus low/moderate exercise intensity sub-analyses.

Finally, whether comorbidity (i.e. concomitant cardiac disease, chronic obstructive pulmonary disease, etc.) impact on exercise-induced improvement of peak VO₂ and glycaemic profile according to exercise training modalities remains to be elucidated.