Comparison of the Effect of Endurance, Strength and Endurance-Strength Training on Glucose and Insulin Homeostasis and the Lipid Profile of Overweight and Obese Subjects: A Systematic Review and Meta-Analysis

The most effective type of training to improve cardiometabolic parameters in overweight subjects is unknown. This meta-analysis compared the effect of endurance, strength and combined training on glucose, insulin metabolism and the lipid profile of overweight and obese adults. The Cochrane, PubMed, Scopus and Web of Science databases were searched to identify randomised trials assessing the effect of training intervention on fasting and 2 h glucose and insulin levels, glycated haemoglobin (HbA1c), homeostatic model assessment of insulin resistance (HOMA), C-peptide, total cholesterol (TC), low- (LDL-C) and high-density lipoprotein cholesterol and triglycerides (TG). Forty-six studies were included showing that endurance training more favourably reduced HbA1c (p = 0.044), and LDL-C (p = 0.021) than strength training. Endurance-strength training more effectively decreased glucose (p = 0.002), HbA1c (p = 0.032), HOMA (p = 0.002), TC (p = 0.039), LDL-C (p = 0.046), HDL (p = 0.036) and TG levels (p = 0.025) than strength training. Combined training significantly reduced the HOMA index (p = 0.009) and TG levels (p = 0.039) compared with endurance training. Endurance and endurance-strength training have a more favourable effect on glucose and insulin homeostasis and lipid profile than strength training in overweight and obese adults. However, the results from this meta-analysis should be interpreted cautiously due to significant heterogeneity among included studies.


Introduction
Obesity is a major public health problem associated with many serious health conditions. Recent data showed that almost two billion adults are overweight globally, while more than 670 million are obese [1]. Obesity significantly increases the risk of the development of several diseases, as excessive body weight is associated with elevated blood lipids, lipoproteins, cholesterol and insulin resistance. Consequently, obesity increases the risk of type 2 diabetes mellitus [2], may cause dyslipidaemia [3] and contributes to

Data Item and Data Collection Process
The following data were extracted from the included papers: first author name, publication year, country, region, the number of subjects included and who completed the intervention, characteristics of the studied population, overweight and/or obesity definition used in the study, age and sex of the study participants, intervention characteristics (study design, type of training, training intensity, frequency, time of intervention and supervision), for each outcome pre-and post-intervention values, changes and p-value. The data from the included papers were extracted by two researchers (J.G. & M.J.) and were checked by the third researcher (M.K.). Another investigator (A.M.-B.) converted each parameter to the same units.

Data Analysis
Study participants were categorised using the body mass index (BMI) cut-off values defined by the World Health Organisation (WHO) as overweight (25-29.9 kg/m 2 ) or obese (≥30 kg/m 2 ) [27]. As the review included the Asian population, special cut-off values for this group were used (overweight: 23-27.5 kg/m 2 and obesity: >27.5 kg/m 2 ) [28]. Waist circumferences (WC) and waist-to-hip ratio (WHR) were categorised according to cut-off points defined by the International Diabetes Federation (IDF) and the WHO, a WC of European men and women should not exceed 94 cm and 80 cm, respectively, whereas 90 cm and 80 cm for Asian men and women [29]. According to WHO guidelines, WHR ≥ 0.9 for men and ≥0.85 for women was defined as abdominal obesity. Moreover, 25% of FM was used as a criterium for diagnosing obesity in men and 32% of FM for women, which is in line with the American Council on Exercise recommendation [30].
The American Diabetes Association recommendations were used to assess glucose metabolism. Impaired glucose tolerance was defined as plasma glucose concentrations of 120 min in the OGTT ranging from 7.8 to 11.0 mmol/L, while impaired fasting glucose was

Data Item and Data Collection Process
The following data were extracted from the included papers: first author name, publication year, country, region, the number of subjects included and who completed the intervention, characteristics of the studied population, overweight and/or obesity definition used in the study, age and sex of the study participants, intervention characteristics (study design, type of training, training intensity, frequency, time of intervention and supervision), for each outcome pre-and post-intervention values, changes and p-value. The data from the included papers were extracted by two researchers (J.G. & M.J.) and were checked by the third researcher (M.K.). Another investigator (A.M.-B.) converted each parameter to the same units.

Data Analysis
Study participants were categorised using the body mass index (BMI) cut-off values defined by the World Health Organisation (WHO) as overweight (25-29.9 kg/m 2 ) or obese (≥30 kg/m 2 ) [27]. As the review included the Asian population, special cut-off values for this group were used (overweight: 23-27.5 kg/m 2 and obesity: >27.5 kg/m 2 ) [28]. Waist circumferences (WC) and waist-to-hip ratio (WHR) were categorised according to cut-off points defined by the International Diabetes Federation (IDF) and the WHO, a WC of European men and women should not exceed 94 cm and 80 cm, respectively, whereas 90 cm and 80 cm for Asian men and women [29]. According to WHO guidelines, WHR ≥ 0.9 for men and ≥0.85 for women was defined as abdominal obesity. Moreover, 25% of FM was used as a criterium for diagnosing obesity in men and 32% of FM for women, which is in line with the American Council on Exercise recommendation [30].
The American Diabetes Association recommendations were used to assess glucose metabolism. Impaired glucose tolerance was defined as plasma glucose concentrations of 120 min in the OGTT ranging from 7.8 to 11.0 mmol/L, while impaired fasting glucose was defined as fasting glucose levels from 5.6 to 6.9 mmol/L, normal glucose tolerance was defined as glucose levels at 120 min in the OGTT < 7.8 mmol/L and normal fasting glucose was defined as fasting glucose levels ranging from 3.9 to 5.5 mmol/L. Diabetes mellitus was diagnosed when fasting glucose levels were ≥7.0 mmol/L or glucose levels at 120 min in the OGTT ≥ 11.1 mmol/L or glycated haemoglobin ≥6.5% [31].
Assessment of fasting insulin levels may be performed in numerous ways, and there are no specific reference values. According to the Adult Treatment Panel (ATP) III-Met, insulin resistance is diagnosed if the homeostatic model assessment of the insulin resistance index reaches ≥1.8 [32]. The normal levels of C-peptide were considered to be in a range from 0.9 to 1.8 ng/mL [33].
According to updates to the ATP III of the National Cholesterol Education Program, LDL-C should be <70 mg/dL for patients with a very high risk of cardiovascular disease and <100 mg/dL for those with a high risk of cardiovascular disease. Preferable concentrations of HDL-C are >40 mg/dL for men and >50 mg/dL for women. The levels of TG should not exceed 150 mg/dL and TC levels should remain <200 mg/dL [34].
Methods of selected unit conversion used in the review are presented in Supplementary Table S2. However, the original data were used to perform the meta-analysis, while the tables show the values after unifying the units for easier data interpretation. Moreover, when logarithmic values are presented, data were transformed back to the raw scale.

Risk of Bias in Individual Studies
Two independent researchers (J.G. & M.J.) assessed the risk of bias using the Cochrane risk of bias tool for randomised trials (RoB 2). The following domains were evaluated: bias due to randomisation, bias due to deviations from intended intervention, bias due to missing data, bias due to outcome measurement, and bias due to selection of reported results [35]. Cochrane handbook for systematic reviews of interventions criteria for low risk, some concerns, and high risk of bias was used [25].

Statistical Analysis
Meta-analysis was performed using the Comprehensive Meta-Analysis 3.0 software (Biostat, Inc., Englewood, NJ, USA) and a p < 0.05 was considered statistically significant. If data were presented only in a figure, the GetData Graph Digitizer 2.26.0.20 (S. Fedorov,Russia) software was used to extract the data. Data in the tables are presented as means and standard deviations (SD) or equivalent and data synthesis was undertaken, including a calculation of effect sizes with 95% confidence intervals (CIs). If a standard error or a 95% CI was provided instead of SD, these data were converted to SD according to the instructions presented in the Cochrane guidelines. Similarly, if the studies included two or more groups of the same type of training, the groups were combined into a single group according to the formula provided in the Cochrane guidelines [25]. Additionally, if studies provided the median and range instead of means and SD, the mean was calculated by the method of Hozo et al. [36]. Fixed-effects models were used if no heterogeneity was present, while random-effects models were used for moderate and high heterogeneity. Standardised mean differences (SMDs) for post-intervention (or changes) values were used and forest plots were generated to compare effect sizes across studies. Funnel plots were generated and Begg's and Egger's tests were performed to assess publication bias. Heterogeneity between studies was evaluated using Cochran Q statistics with p < 0.1 indicating significant heterogeneity. The I 2 test was used to measure consistency between studies. According to the Cochrane handbook for systematic review I 2 < 40% suggests a low risk of heterogeneity, 40% to 75% is considered a moderate risk of heterogeneity, and >75% indicates a high risk of heterogeneity [25]. A sensitivity analysis was performed to assess the influence of each study on the overall effect. The sensitivity analysis was also performed by excluding studies with a high risk of bias to determine how the exclusion affects the overall effects. A cumulative meta-analysis was performed to evaluate how the effect changed over time with studies sorted from the oldest to the newest. Subgroup analysis was conducted to compare the effect of studies with short (≤12 weeks) and long (>12 weeks) times of the intervention as well as to assess the effectiveness of combined training with the same and longer duration as endurance and strength training alone. Dupuit et al. [37] included two endurance groups. Therefore, in the subgroup analysis, the group which performed the endurance exercises at the same duration as that in combined group was included.

NI CT
ET: Similar as described for ET ST: Exercises: arm curl, triceps extension, and shoulder press for upper-limb training; squat, leg press, leg curl, leg extension, and calf raise for lower-limb training; and bench press, seated butterfly, lat pulldown, trunk curl, back extension, and dead lift for trunk training. Participants selected 2 types each from the upper and lower limb training options, and 3 from trunk training choices, and thus performed 7 exercises in each training session; 3 sets for each exercise consisting of 10 repetition   3 High-intensity interval training; 4 Mean ± standard deviation; 5 Two weeks for the familiarisation with the training and 8 weeks for the main training; 6 The goal duration/volume of training; 7 Mean ± standard error; 8 The total number of min that needed to be obtained was determined by fitness level, as all subjects were prescribed a specific amount of exercise per unit body weight. Higher fit individuals required less time to expend the prescribed number of calories per week; subjects were encouraged not to exceed 60 min/day; 9 Low-amount moderate-intensity training group; 10 Low-amount vigorous-intensity training group; 11 High-amount vigorous-intensity training group; 12 Four sets of 4 min training with 3 min recovery; 13 High-intensity circuit training; 14 Low-intensity circuit training; 15 Program 1: 40 min, programme 2: 50 min; a-c Studies marked with the same letters were conducted in the same population.

The Effect of Training Intervention on 2 h Glucose Levels
The impact of the intervention on 2 h glucose levels was assessed in two studies. Endurance training was compared with strength training in both studies [62,65] and one study also evaluated the effect of endurance and combined training and resistance and mixed training [62].

The Effect of Training Intervention on 2 h Glucose Levels
The impact of the intervention on 2 h glucose levels was assessed in two studies. Endurance training was compared with strength training in both studies [62,65] and one study also evaluated the effect of endurance and combined training and resistance and mixed training [62].

The Effect of Training Intervention on 2 h Insulin Levels
The effect of training programmes on 2 h insulin concentrations was measured in two studies [62,65]. One study compared endurance with strength training [65] and Donges et al. [62] compared endurance, strength and endurance-strength training.

The Effect of Training Intervention on 2 h Insulin Levels
The effect of training programmes on 2 h insulin concentrations was measured in two studies [62,65]. One study compared endurance with strength training [65] and Donges et al. [62] compared endurance, strength and endurance-strength training.

The Effect of Training Intervention on 2 h Glucose Levels
The impact of the intervention on 2 h glucose levels was assessed in two studies. Endurance training was compared with strength training in both studies [62,65] and one study also evaluated the effect of endurance and combined training and resistance and mixed training [62].

The Effect of Training Intervention on 2 h Insulin Levels
The effect of training programmes on 2 h insulin concentrations was measured in two studies [62,65]. One study compared endurance with strength training [65] and Donges et al. [62] compared endurance, strength and endurance-strength training.

The Effect of Training Intervention on C-Peptide Levels
C-peptide levels were measured in two studies [22,68], Sukala et al. [68] compared endurance and strength training, while Stensvold et al. [22] assessed the effect of endurance, strength and combined training.
C-peptide levels were measured in two studies [22,68], Sukala et al. [68] compared endurance and strength training, while Stensvold et al. [22] assessed the effect of endurance, strength and combined training.

The Effect of Training Intervention on Lipid Metabolism
The effect of training programmes on lipid profile is presented in Table 5.  [22,68]. CI-confidence interval; ET-endurance training; ST-strength training; Std-standard; Std diff-standard differences.

The Effect of Training Intervention on Lipid Metabolism
The effect of training programmes on lipid profile is presented in Table 5.
Moreover, the effectiveness of combined training with the same and longer duration as endurance and strength training alone was compared. Interestingly, more efficiency of combined training than strength training was found in studies in which endurancestrength training had a similar duration of volume than strength training alone (glucose (random-effects model, SMD: 1.264, 95% CI: 0.532, 1.997, p = 0.001, Figure S49B) and HOMA (random-effects model, SMD: 1.475, 95% CI: 0.517, 2.433, p = 0.003, Figure S52B). Moreover, the comparisons of the effect of strength and combined training on HbA1c, TC and LDL-C levels were based on the studies in which both types of activity had the same duration and the analyses also showed that endurance-strength training was more effective than strength training (HbA1c: SMD: 1.320, 95% CI: 0.114, 2.525, p = 0.032, Figure 4C; TC: SMD: 1.185, 95% CI: 0.060, 2.309, p = 0.039, Figure 9C; LDL-C: SMD: 1. 655, 95% CI: 0.032, 3.278, p = 0.046, Figure 10C). Additionally, endurance-strength training was more effective than endurance training alone in decreasing the HOMA index (random-effects model, SMD: 0.415, 95% CI: 0.127, 0.703, p = 0.005, Figure S52A) for studies in the similar duration of both programme.

Discussion
Herein, it is reported that endurance training is more effective in reducing HbA1c and LDL-C levels than strength training, endurance-strength training more effectively decreases glucose, HbA1c, HOMA, TC, LDL-C, HDL-C and TG concentrations than strength training and combined training significantly more reducing HOMA index and TG levels than endurance training. The findings agree with the current physical activity guidelines, which recommend mostly endurance or endurance training combined with strength training for obese subjects at risk of cardiovascular disease [76][77][78].
The recent network meta-analysis of Batrakoulis et al. [23] compared the efficacy of five different exercise modalities (continuous endurance training, interval training, resistance training, combined training and hybrid-type training) on cardiometabolic parameters in overweight and obese subjects and found that hybrid-type training was the most effective for reducing fasting glucose concentrations and increasing HDL-C levels, combined training was the most effective in reducing fasting insulin concentrations, HOMA-IR index and LDL-C levels, and interval training exhibited the highest probability of reducing HbA1c and TG levels. Moreover, subgroup analysis showed that the effects of combined training are more pronounced in males, while hybrid-type training induces more cardiometabolic benefits in females. However, it should be highlighted that the meta-analysis included only studies performed on overweight and obese subjects without comorbidities which may explain the differences in results obtained in this study. Another meta-analysis compared the effect of endurance, strength and combined training in subjects with type 2 diabetes and found that endurance training had a clear but small benefit for TC levels in comparison to strength training, while combined training compared with endurance training was most effective in reducing fasting glucose and HDL-C levels. These findings are mostly in line with these results, but the authors observed no differences between training programmes in the effect of components of the lipid profile of markers of glucose and insulin metabolism [79]. In a meta-analysis, Liang et al. [10] also examined the effects of aerobic, resistance, and combined exercise on metabolic syndrome parameters and cardiovascular risk factors to identify the most effective way of improving metabolic syndrome and preventing cardiovascular disease. The combined exercise was most effective at controlling glucose and TG levels, but there was no statistically significant difference in TC, LDL-C, HDL-C and insulin levels among the exercise groups. Based on the surface under the cumulative ranking curve (SUCRA), combined exercise was also the best for improving insulin and TC levels, resistance exercise was most effective at ameliorating LDL-C levels and aerobic exercise was optimal for improving HDL-C levels.
The mechanism that explains the differences in the effect of endurance, strength and combined training programmes on cardiometabolic parameters has not been clarified. One of the explanations for the more beneficial effect of combined training compared with endurance or strength training is that in several studies, albeit not all, the duration of a single training session was longer compared with endurance or strength training alone [13,15,20,37,48,59,72]. In the Studies of a Targeted Risk Reduction Intervention through Defined Exercise (STRRIDE) study, the participants in the combination groups exercised for approximately double the time of the aerobic and resistance training groups [13,15,20,59]. Martins et al. [48] reported that combined training was two times longer than endurance training. Similarly, in the study performed by Hara et al. [72], combined training consisted of exercises performed in the endurance and strength groups. Therefore, it is not clear if the marked beneficial combination training effects on some markers are due to the greater volume of exercises or a mechanistic synergy of the two exercise modes. However, the results of the subgroup analysis showed that combined training was more effective in decreasing glucose levels and HOMA index than strength training and also than endurance training in decreasing the HOMA index for studies in which combined training had a similar duration as endurance and strength training alone. Nevertheless, it should be highlighted that there was only one study that compared the effect of strength and endurance-strength training on glucose concentrations and HOMA index in which combined training was longer than strength training alone [20].
It was hypothesised that the greater effectiveness of one type of training over another could be related to a higher reduction in body weight and improved body composition. Decreasing visceral FM can particularly affect cardiometabolic parameters and it is well known that abdominal obesity is highly correlated with impaired glycaemic control and lipid profile due to increased visceral fat accumulation [80]. Several studies included in this meta-analysis reported that endurance [16,62] or endurance-strength training [66] was more effective in decreasing body weight compared with strength training. Additionally, Mohammed Rahimi et al. [46] observed that the reduction in %FM and WC in the combined group was significantly greater than in the endurance group, while Tayebi et al. [58] reported that the decrease in %FM in the endurance group was significantly higher than in the strength group, and in endurance-strength training was more pronounced than for both endurance and strength groups. Batrakoulis et al. [23], in the network meta-analysis, found that combined training had the highest probability of being ranked best compared with other exercise types in reducing body weight and the highest likelihood of lowering FM in obese subjects without comorbidities. Liang et al. [10], in another meta-analysis that included studies performed on subjects with a high risk of metabolic syndrome, showed that aerobic, resistance and combined exercise groups achieved significant effects on body fat. However, aerobic exercise was superior to resistance exercise regarding BMI but there was no statistically significant difference in weight and WC among the exercise groups. Notwithstanding, according to the SUCRA results, combined exercise is best for improving weight and WC, while resistance exercise was most effective at ameliorating body fat. Morze et al. [81], in a network meta-analysis performed on subjects with obesity, noted that aerobic training was ranked best for improving body weight, BMI and WC and combined training for improving FM and equally to resistance training for improving free fat mass. By contrast, Yarizadeh et al. [82], in their meta-analysis, compared the effect of aerobic, resistance and combined exercise modalities on subcutaneous abdominal fat and reported that aerobic exercise was shown to produce greater efficacy in decreasing this parameter.
Reduced caloric intake is a crucial factor influencing weight loss and improvement of cardiometabolic parameters. However, in this meta-analysis, only studies in which subjects were instructed not to change their dietary habits during the intervention were included. Indeed, several studies reported no differences in energy values and/or macronutrient distribution in diet between values obtained before and after the intervention period in all study groups [43,49,53]. However, Ho et al. [66] mentioned that when comparing withingroup changes, the aerobic and resistance groups had significantly lower daily energy intake at week 12 compared with baseline, but there were no significant differences in total energy intake between groups.
Another mechanism that may explain differences in the effect of endurance, strength and endurance-strength training may be related to differences in energy expenditure between training types. It is important to point out that strength training results in a significantly lower caloric expenditure than a similar amount of time spent in vigorous endurance training. Davidson et al. [83] estimated that the typical strength programme expended 45% of maximal VO 2 , while 75% of maximal VO 2 was used in the aerobic programme, with 67% more calories likely to be expended in the endurance programme [15]. However, resistance exercise has also been demonstrated to increase basal energy expenditure by increasing muscle volume [84]. Unfortunately, only few studies included in the meta-analysis provided information about energy expenditure by each training programme [13,15,20,50,53,59], and the week energy expenditure was the same in all groups only in one study [50].
Higher adherence to endurance or endurance-strength training than to strength training could partly explain the better effect of the first two training types noted in this meta-analysis. Unfortunately, adherence to the study intervention was reported only in single studies [13,15,20]. AbouAssi et al. [20] showed that participants in the endurance training group were more adherent to the aerobic regimen compared with participants in the combined group. No other group differences in adherence were observed. Bateman et al. [13] found that adherence was slightly lower for each portion of the combined group than for either endurance or strength training. However, the total time accumulated for the combined group remained almost double that of the other exercise groups. Future studies must direct greater attention toward exercise adherence.
It has been demonstrated that more favourable changes in response to training usually occur in subjects with more pronounced disorders at baseline and baseline differences between groups may have an important effect on the obtained results. In the meta-analysis, studies that recruited subjects with and without obesity-related comorbidities were included. Most studies did not have differences at baseline in analysed parameters between groups. Nevertheless, some differences at baseline between groups for study outcomes were reported by Alvarez et al. [50], Oh et al. [53], Hara et al. [72], Venojarvi et al. [65] and Donges et al. [62], which may have some effect on the study results and the metaanalysis findings.
One of the mechanisms by which physical activity can decrease the risk of cardiovascular diseases is the anti-inflammatory effect of exercise [85]. Weight gain may lead to the overproduction of pro-inflammatory cytokines involved in the pathogenesis of cardiometabolic disorders [86]. Therefore, a reduction in low-grade inflammation may accompany improved cardiometabolic markers [87]. Endurance, strength and combined training may affect inflammatory parameters differently. A recent meta-analysis reported that endurance training is more beneficial in reducing C-reactive protein, interleukin 6, and visfatin concentrations in overweight and obese adults than in strength training. Additionally, a combined training programme was significantly more beneficial in lowering tumour necrosis factor α levels compared with a strength training programme [88]. The differences between the effects of particular types of training on inflammatory parameters may be explained by the promotion of other specific cardiovascular and neuromuscular adaptations [89].
The intervention time could also impact the results of the meta-analysis; therefore, a subgroup analysis was performed dividing the studies into two groups with short (≤12 weeks) and long (>12 weeks) intervention periods, showing that for short intervention times, combined training was more effective than strength training in reducing glucose levels and HOMA index and endurance training in decreasing glucose levels and HOMA index. These results can be explained by the difficulty in maintaining high adherence levels in longer intervention studies and could be related to decreased motivation and an increased drop-out rate. However, for the long-term intervention, endurance-strength training more effectively decreased insulin and TG levels than the endurance programme. Additionally, combined training more effectively decreased TC and LDL-C levels than strength training but the observation was performed based on the results of one study. Surprisingly, there were no differences between studies with short and long-term intervention in effect on HbA1c, but HbA1c does not change rapidly and the marker estimates the average glucose levels over the past three months [90]. This is one of the first meta-analyses to compare the effect of endurance, strength and combined training on glucose, insulin and lipid metabolism in overweight and obese subjects (with and without comorbidities) who did not receive dietary intervention or advice. Different criteria were used to define overweight and obesity for different populations and parameters, such as BMI, WC, WHR or %FM, allowing more studies to be included in the analysis. The other strengths of this meta-analysis include the detailed characteristics of the study populations and interventions. Moreover, the effectiveness of combined training with similar and longer duration than endurance and strength training alone was compared.
Nonetheless, this study has several limitations. Firstly, meta-regression and network meta-analysis were not performed. Secondly, there was significant heterogeneity among the included studies despite the strict inclusion and exclusion criteria. In addition, subgroup analysis was not performed regarding sex, age, the health status of participants, intensity and frequency of training. Furthermore, the effect of training programmes on anthropometric parameters and body composition was not assessed as it was comprehensively presented in the recent meta-analysis [81]. Additionally, the use of a strict definition of endurance, strength and combined training in the study protocol was not possible and the effect of each training type on a control group was not compared. The meta-analysis included both studies in which the duration of a single training session was the same in all groups and studies in which combined training had a longer duration than endurance or strength training alone. Therefore, the obtained results may vary in different parameters because of the different duration of a single training session for each type of exercise.

Conclusions
Endurance and endurance-strength training have a more favourable effect on glucose and insulin homeostasis as well as lipid profile than strength training in overweight and obese adults, with the intervention duration having a significant impact on the obtained results. Moreover, combined training seems to have a more promising effect than endurance training. However, the results from this meta-analysis should be interpreted cautiously due to significant heterogeneity among included studies. Additionally, more studies are needed to assess the impact of training intervention on 2 h glucose and insulin levels and C-peptide.