For this systematic review, searches were conducted in the electronic databases MEDLINE/PubMed (1966 – April/2016), CINAHL (1981 – April/2016), SPORTDiscus (1985 – April/2016), LILACS (1986 – April/2016), and SciELO (1998 – April/2016). There were no restrictions to language or time of publication, and the search strategies for each database (presented on the Table of Appendix A) took into account their specific descriptors through the Medical Subject Headings (MeSH), CINAHL Subject Headings, and Health Sciences Descriptors (DeCS). In addition, we screened from the reference lists of the eligible trials.

The articles selected for a more specific analysis included randomized controlled clinical trials published as full papers, studies that involved patients with type 2 DM over 19 years old, and articles which compared resistance exercise against aerobic exercise through a structured protocol with detailed description of both modalities.

The following were considered as primary outcomes: long-term glycemic control assessed by glycated hemoglobin (HbA1c) or short-term glycemic control assessed through postprandial or fasting blood glucose (FBG) concentration; cardiorespiratory fitness through maximal oxygen consumption (VO2max); and body mass index (BMI). As secondary outcomes, we considered blood lipid profile [total cholesterol, high-density lipoprotein cholesterol (HDL), low-density lipoprotein cholesterol (LDL), triglycerides] and the presence of adverse effects such as episodes of hypoglycemia, muscle fatigue, and mortality.

The following types of documents were excluded: letters, editorials, extended abstracts, studies that included patients with ulceration, skin lesions and/or rheumatic disease, the presence of chronic conditions, as well as patients with gestational diabetes and type 1 DM. Studies which involved prior or on-going insulin therapy, change in the use of corticosteroids, oral hypoglycemic drugs, or any kind of hypoglycemic diet two months before the start of the exercise protocol were also excluded.

The searches, data collection, and content analysis of the selected studies were performed by two independent reviewers (CSN and KA), and the differences were discussed by a third evaluator (SRAM) using an eligibility criteria of a previously established protocol for the elaboration of the systematic review.

The quality of evidence for the outcomes HbA1, VO2max, BMI, total cholesterol, HDL, LDL, and triglycerides was assessed using the GRADE system and presented through the Summary of Findings Table (Table 1). The GRADE proposal classifies the level of evidence as high, moderate, low, or very low considering five factors that can affect the quality of the evidence of the outcomes of a clinical trial: design, risk of bias, inconsistency, indirectness, and imprecision.11 A review of the evidence for each factor followed the subsequent classification: no (no reduction in points), serious (reduction of 1 point), and very serious (reduction of 2 points),12 being scored by reviewers according to the interference biases detected in these items.

For the specific GRADE item ‘risk of bias’, the Cochrane Collaboration's risk of bias tool was used, which takes into consideration the items: randomization, allocation concealment, blinding, loss to follow up, selective outcome reporting, and early stopping of trials. The established guidelines to assess the risk of bias were high, low, or unclear.13 Risk was considered high when the item was not fulfilled, when the method by which the item was carried out was not reported, or when the method was not valid.14 Low risk of bias was considered when the item was fulfilled and reported inadequately,15 and unclear risk of bias when the available information was insufficient for a high or low risk of bias account is made, when information about the conduct was sufficient, but the risk of bias was really unknown, or when the analyzed item did not apply to the study in question.16

The statistical analysis was performed using the Review Manager Software (RevMan) version 5.3 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014). The data were typed independently by two reviewers and subsequently compared. A meta-analysis was performed of all primary outcomes and most of the pre-defined secondary ones. A subgroup analysis of the studies was carried out according to the duration of the protocol, from eight to 20 weeks and 21–48 weeks. The homogeneity of the studies was determined through the test of heterogeneity. Studies were considered homogeneous when p assumed a value greater than 0.05 and the heterogeneity index (I2) was classified as low (heterogeneity values up to 30%). A meta-analysis of random effect was used considering that the studies were not functionally equivalent and that the effect size varied across populations.

A total of 336 diabetic patients of both genders were evaluated. The studies were performed in Brazil, Canada, South Korea, Greece, India, Iran, Italy, and the USA. The predominant ethnicity was non-Hispanic white. The studies involved participants with an average age per group between 48 and 58 years and time of diagnosis of diabetes of one year or longer. The main criteria for inclusion consisted of sedentary individuals with DM type 2 for more than six months and who presented HbA1c between 6.5 and 11%. The reasons for exclusion and more details about the features of the studies are arranged in Table 2.

In relation to the aerobic exercises, walking was the predominant form, followed by exercise on a bike ergometer and cycle ergometer. While in the resistance exercises, the most widely used was resistance training of the major muscle groups. The protocols ranged from eight to 48 weeks lasting 20–60min/day, two to five times per week (Table 3). Of the eight studies included, one17 (published in19,21,23,26) did not report comparisons between resistance and aerobic groups.

Four studies showed increased risk of bias18,22,25,31 for not describing the process of randomization, allocation, and if there was blinding of participants, trainers, and assessors (presented in the Figure of Appendix A). Articles related to the same group as represented by (1) Arora et al.18 and Shenoy et al.20; (2) Bacchi et al.28 and Bacchi et al.32; (3) Jorge et al.25 and De Oliveira et al.29; and (4) Sigal et al.17 (in the following publications19,21,23,24,26) only differed with regard to the bias of incomplete results and selective description.

Assessor blinding was conducted in only four studies17,27,28,30 (in five publications19,21,23,24,26). As for the bias of incomplete data, two articles17,22 (in one publication26) did not describe whether there were losses or exclusions. However, those that did report incomplete data described them mostly as a result of lack of attendance, presence of negative results in laboratory and/or diagnostic tests, time constraints, or personal reasons not favoring the risk of incomplete data bias. None of the authors described that they performed an intention-to-treat analysis, and this was not observed in the articles. Another prominent point refers to the sample size calculation, which was not performed in any of the included studies.

It was possible to perform five meta-analyses involving the outcomes HbA1c, VO2max, BMI, HDL, LDL, triglycerides, and total cholesterol, according to subgroups (eight to 20 weeks and 21–48 weeks) proposed in the Methods section. In relation to glycated hemoglobin (HbA1c), there was very low-quality evidence (downgraded due to risk of bias, inconsistency, indirectness, and imprecision) of lack of difference between the aerobic and resistance groups in both subgroups [seven studies17,18,22,25,27,30,31; 276 adults; difference in average % of 0.10; 95% CI: −0.15 to 0.34; random effect] (Fig. 2). One study28 showed the mean difference between the baseline and final values for each group with the confidence interval (aerobic group: MD: −0.4; 95% CI: −0.61 to −0.18; resistance group: MD: −0.35; 95% CI: −0.59 to −0.10).

Figure 2.

Meta-analysis graph (forest plot) for comparative analysis of HbA1c (A), VO2max (B), body mass index (C), HDL (D), LDL (E), triglyceride levels (F), and total cholesterol levels (G) after resistance (experimental) or aerobic (control) exercise.

Regarding VO2max, there was very low quality evidence (downgraded due to risk of bias, inconsistency, indirectness, and imprecision) of an increase of 2.86 (ml/kgmin) in favor of resistance exercise [five studies17,25,27,30,31; 247 adults; average difference in ml/kgmin of −2.86; 95% CI: −3.90 to −1.81; random effect] (Fig. 2). One study28 showed the mean difference between the baseline and final for each group with the confidence interval (aerobic group: MD: 4.0; 95% CI: 2.7–5.3; resistance group: MD: 2.1; 95% CI: 0.6–3.5).

There was no difference in relation to BMI between the aerobic and resistance groups, from very low-quality evidence (downgraded due to risk of bias, inconsistency, indirectness, and imprecision) [seven studies17,18,22,25,27,30,31; 293 adults; average difference in kg/m2 0.75; 95% CI: −0.20 to 1.69; random effect] (Fig. 2). One study28 showed the mean difference between the baseline and final values for each group with the confidence interval (aerobic group: MD: −0.76; 95% CI: −1.1 to −0.4; resistance group: MD: −0.54; 95% CI: −0.85 to −0.22).

HDL values did not differ between the aerobic and resistance groups, from very low-quality evidence (downgraded due to risk of bias, inconsistency, indirectness, and imprecision) [five studies17,18,25,27,30,31; 267 adults; average difference in ml/kgmin of −0.36; 95% CI: −0.78 to 0.05; random effect] (Fig. 2). One study28 showed the mean difference between the baseline and final values for each group with the confidence interval (aerobic group: MD: 2.9; 95% CI: −0.28 to 6.1; resistance group: MD: 1.3; 95% CI: −1.1 to 3.8).

With regard to LDL values, there was very low-quality evidence (downgraded due to risk of bias, inconsistency, indirectness, and imprecision) of no difference between aerobic and resistance groups [five studies17,25,27,30,31; 248 adults; average difference in mg/dl of −2.71; 95% CI: −13.15 to 7.73; random effect] (Fig. 2). One study28 showed the mean difference between the baseline and final values for each group with the confidence interval (aerobic group: MD: 1.8; 95% CI: −9.9 to 13.5; resistance group: MD: 2.3; 95% CI: −4.5 to 9.2).

The rate of triglycerides, from a very low-quality evidence (downgraded due to risk of bias, inconsistency, indirectness, and imprecision), did not differ between the aerobic and resistance groups [six studies17,18,25,27,30,31; 267 adults; average difference in mg/dl of 2.37; 95% CI: −12.96 to 17.70; random effect] (Fig. 2). One study28 showed the mean difference between the baseline and final for each group with the confidence interval (aerobic group: MD: −27.8; 95% CI: −57.5 to 1.7; resistance group: MD: −23.9; 95% CI: −49.5 to 1.6).

As for total cholesterol, there was very low-quality evidence (downgraded due to risk of bias, inconsistency, indirectness and imprecision) of no difference between the groups [five studies17,18,27,30,31; 243 adults; mean difference in mg/dl of −4.43; 95% CI: −14.74 to 5.87; random effect] (Fig. 2). One study28 showed the mean difference between the baseline and final values for each group with the confidence interval (aerobic group: MD: −0.8; 95% CI: −15.8 to 14.1; resistance group: MD: −0.7; 95% CI: −8.5 to 7.1).

For all the outcomes mentioned above, one study28 was not included in the meta-analysis because it reported the data by mean change and confidence interval. It was not possible to calculate the standard deviation of the final values because the sample size varied from the baseline.

Only five studies17,27,28,30,31 mentioned the presence of adverse effects during the design of the study. Sigal et al.17 did not report such effects. However, the others reported transient hypoglycemia during physical training, hypoglycemic events, musculoskeletal pain, and need for insulin therapy during the study. Only one of the studies17 reported that the data were not different between the groups (RR=0.94; 95% CI: 0.25–3.58).