Effects of Exercise on Glycemic Control and Body Mass in Type 2 Diabetes Mellitus:  A Meta-analysis of Controlled Clinical Trials FREE

For many years, exercise, along with diet and medication, has been considered 1 of the 3 cornerstones of diabetes therapy.¹ Regular physical activity is recommended for patients with type 2 diabetes since it may have beneficial effects on metabolic risk factors for the development of diabetic complications.² The low-cost, nonpharmacological nature of physical activity further enhances its therapeutic appeal.

Two of the major goals of diabetes therapy are to reduce hyperglycemia and body fat. Chronic hyperglycemia is associated with significant long-term complications, particularly damage to the kidneys, eyes, nerves, heart, and blood vessels.³ Obesity, especially abdominal obesity, is associated with insulin resistance, hyperinsulinemia, hyperglycemia, dyslipidemia, and hypertension^4- 5; these abnormalities tend to cluster and are often referred to as the "metabolic syndrome."⁶ Elements of the metabolic syndrome are strong risk factors for cardiovascular disease,^7- 8 and regular exercise in nondiabetic subjects has beneficial effects on virtually all aspects of the syndrome.^4,9- 10

Although there have been numerous small studies on the effects of exercise in patients with type 2 diabetes, their findings have varied. There have been no large studies with adequate statistical power to guide practitioners in recommending exercise plans for their patients with diabetes. Exercise interventions reduced glycosylated hemoglobin (HbA_1c) in some studies,^11- 15 but not in others.^16- 22 Meta-analysis may be especially useful in summarizing and analyzing prior research when the number of subjects per study is small and the results are somewhat conflicting,²³ as is the case for exercise interventions in patients with type 2 diabetes.

The main objective of this study was to systematically review the effect of exercise interventions on glycemic control as represented by HbA_1c and body mass, measured as body weight in kilograms or body mass index (calculated as weight in kilograms divided by the square height in meters) in adults with type 2 diabetes.

Literature searches of computer databases were performed for the period up to and including December 2000 (MEDLINE 1966-2000, EMBASE 1980-2000, Sport Discuss 1949-2000, Health Star 1975-2000, Dissertation Abstracts 1861-2000, and the Cochrane Controlled Trials Register). The reference lists of major textbooks, review articles, and of all included articles identified by the search were hand searched to find other potentially eligible studies. Potential missing articles and unpublished literature were sought from experts. Non-English studies were included.

The computer-based search strategy included common text words and Medical Subject Headings related to exercise and type 2 diabetes. MEDLINE and EMBASE searches were limited to human subjects and used a validated and highly sensitive search filter for randomized controlled trials and nonrandomized controlled clinical trials (CCTs).²⁴ If studies reported data for which it was impossible to discriminate between participants with type 2 diabetes and those with impaired glucose tolerance, we attempted to contact authors for the individual patient data.

We limited the analyses to exercise interventions lasting at least 8 weeks since our main outcome of interest, HbA_1c, reflects average blood glucose concentration from the previous 8 to 12 weeks. An exercise intervention was defined as a predetermined program of physical activity described in terms of type, frequency, intensity, and duration. Studies in which the intervention consisted only of recommending increased physical activity were not included within the analyses since it would be impossible to quantify the exercise intervention and compliance. To be included, compliance with exercise interventions had to be verified by direct supervision or through exercise diaries. Studies that included drug cointerventions were excluded from the analysis.

For the variables of interest, we extracted sample sizes as well as baseline and postintervention means and SDs for the intervention and control groups. The authors of potentially eligible studies were contacted when necessary to resolve ambiguities in reported methods or results and to seek additional information. In some cases,^12,19,25 postintervention SDs were not available. In these instances, we imputed the baseline SD. This was assumed to be valid since the baseline and postintervention SDs were found to be similar within the other trials included in the meta-analysis (for example, the postintervention SDs were on average 0.14 units lower than the baseline SDs). Where necessary, means and measures of dispersion were approximated from figures in the manuscripts using an image scanner (RT6l5CTW [resolution 200 dpi], Compaq Computer Corp, Houston, Tex), as described previously.²⁶

The characteristics of the exercise interventions were extracted, including the type, frequency, duration, intensity, and energy cost. The Compendium of Physical Activities²⁷ was used to estimate the exercise intensity in terms of metabolic equivalents (METs). Exercise volume (total energy expenditure on exercise, in METs per hour) was calculated by multiplying the intensity in METs by total time spent exercising (number of exercise sessions multiplied by duration of each exercise session). The methodological quality of each included trial was assessed using a validated 5-point scale as described by Jadad et al.²⁸ The instrument assigned scores for reported randomization, blinding, and withdrawals. In addition to this quality scoring, we recorded separately whether allocation concealment was adequate, as described by Schulz et al.²⁹

Two of the authors (N.G.B. and E.H.) independently performed the literature search, quality assessment, and data extraction. Any disagreements on inclusion of trials or quality assessment were resolved by discussion with a third author (R.J.S.).

Statistical analysis was performed with Review Manager Software (RevMan 4.1, Cochrane Collaboration, Oxford, England) and JMP Software (Version 3.1.6.2, SAS Institute Inc, Cary, NC). In each study, the effect size for the intervention was calculated by the difference between the means of the exercise and control groups at the end of the intervention. Each mean difference was weighted according to the inverse of its variance, and the average was taken (weighted mean difference [WMD]). When the same outcome was measured by different scales (ie, body mass represented by body weight in some studies and body mass index in others), the mean difference was standardized by dividing it by the within-group SD; the results were then weighted and the average taken (standardized mean difference [SMD]). The WMD or SMD in each study was pooled with a fixed-effects model.³⁰ The χ² test for heterogeneity was performed, and when significant heterogeneity was found, the analysis was redone with a random-effects model.³⁰ The funnel plot technique³¹ was used to detect publication bias.

We performed a meta-regression analysis to explore whether effects of the exercise interventions on HbA_1c were mediated by effects on body mass, by the exercise intensity, or by exercise volume. In all meta-regression models, studies were weighted by sample size. In the first model, the mean difference in end-of-study HbA_1c for individual studies was regressed on the corresponding mean difference in body mass at the same time point. In the second model, we corrected for baseline values by regressing the difference between exercise and control groups' change from baseline in HbA_1c on the corresponding values for body mass change. In the third model, the mean difference in end-of-study HbA_1c was regressed on the exercise intensity (METs). In the fourth model, exercise volume was used (total METs per hour). In the fifth model, body mass, exercise intensity, and exercise volume were entered simultaneously.

The computer searches yielded approximately 2700 potential articles. After the application of the filter for CCTs, the number of potential studies was 1487. The most common reasons for exclusion were: review article only, nonhuman subjects, lack of type 2 diabetic control group, lack of an exercise training intervention, duration of intervention less than 8 weeks, and/or absence of exercise supervision or exercise diaries. Eventually, 14 trials were deemed appropriate for inclusion, but in some cases there were multiple publications from the same trial.^32- 40

Two of the 14 trials presented data for 2 comparisons, therefore 16 comparisons were included (Table 1). One of these trials¹² had a 2 × 2 factorial design in which participants were assigned to 4 groups (exercise, diet, exercise and diet, and control). For this trial, we were able to analyze 2 comparisons: exercise and diet vs diet alone; and exercise alone vs control. The second trial, Vanninen et al,²¹ had data analyzed separately for men and women. In the article by Wing et al,²² the results of 2 separate studies were presented.

Of the studies otherwise meeting inclusion criteria, one study⁴¹ was excluded because the exercise intervention alternated between 3 months of exercise and 3 months without exercise, and another study⁴² was excluded because program participation was not associated with a significant increase in physical activity. Two studies^43- 44 were excluded because we were unable to differentiate between participants with and without diabetes and 3 others^45- 47 were excluded because postintervention HbA_1c and body mass values were not available. In the study by Kaplan et al,¹⁸ the combined exercise and diet group was excluded from the analysis because only the last 5 of the 10 weeks included exercise.

A total of 504 participants were included in the 14 trials. The mean (SD) age of participants in studies for which this information was available was 55.0 (7.2) years, duration of diabetes was 4.3 (4.6) years, and 50% of participants were women (Table 1). The quality of the trials according to the scale described by Jadad et al²⁸ was moderate to low. The mean (SD) score was 1.6 (0.5) out of a possible 5 points. The quality assessment criterion that permitted the greatest discrimination between studies was randomization since the studies obtained similar scores on other methodological quality characteristics. Of the 14 trials, 11 were randomized controlled trials and 3 were CCTs (Table 1). None of the trials was double blind or had adequate allocation concealment. Nine trials had adequately described dropouts,^{11- 16,18- 20} there were no dropouts in 2 of the studies,^17,25 while 3 studies did not comment on dropouts.^21- 22

The compliance to the exercise interventions was relatively high. In 9 studies,^{11- 14,18,20- 22} the mean participation rate was above 80%, 2 articles^15,19 indicated that compliance was good, and 3 studies^16- 17,25 did not comment on compliance. Dietary compliance was assessed in all but 1 study²¹ that prescribed dietary cointerventions, using food diaries or weekly meetings with a dietician. Medication was altered for a small number of participants during 1 study.²¹ In this study, 6 participants started taking glybenclamide and it was not stated how many of these participants had been assigned to the exercise group. More details on the medications taken by participants and dietary cointerventions are provided in Table 1.

The exercise interventions in each study are described in Table 1. The exercise interventions typically prescribed 3 workouts per week, each lasting a mean (SD) of 53 (17) minutes (including 10 minutes of warm-up and cool-down) for 18 (15) weeks. The intensity of the aerobic exercise was moderate and typically consisted of walking or cycling. Two studies^16- 17 used resistance exercise training as an intervention and 1 study²⁰ added resistance training with elastic bands to its aerobic training program. Resistance training was composed of 2 to 3 sets ranging from 10 to 20 repetitions. One study¹⁶ described the initial resistance to be at 50% to 55% of the participants' repetition maximum, while the other did not specify the intensity.¹⁷ Both studies stated that the resistance was progressively increased.^16- 17

The results are presented for 2 types of comparisons. The exercise vs nonexercise control comparisons included studies in which there was no diet cointervention or in which the same diet cointervention was given to both the exercise and control groups. The second set of comparisons was between combined exercise and diet interventions vs nondieting/nonexercising control groups.

Baseline and postintervention HbA_1c values were described in 12 studies (14 comparisons). For the 11 comparisons between exercise and nonexercise control groups there were no significant baseline differences in HbA_1c (WMD, 0.08%; P = .65). As shown in Figure 1, when the postintervention results were pooled, HbA_1c was significantly lower in the exercise groups compared with the control groups (7.65% vs 8.31%; WMD, −0.66%; P<.001). Figure 1 also illustrates the effects on HbA_1cof interventions combining exercise and diet compared with nonexercise, nondiet controls. When diet and exercise were combined, the effect on HbA_1c was similar to the effect of exercise alone (WMD, −0.76%; P = .008).

A sensitivity analysis identified no significant differences between the results from randomized controlled trials and CCTs. The postintervention WMD between exercise and control groups for the 9 randomized comparisons was −0.63% (95% confidence interval [CI], −1.01% to −0.25%; P = .001), whereas the corresponding WMD for the 2 CCTs was −0.75% (95% CI, −1.36% to −0.14%; P = .02). Further subgroup analysis comparing aerobic or resistance training groups with the control group revealed no significant difference. The WMD for aerobic training vs control was −0.67% (95% CI, −1.04% to −0.30%; P<.001) and was −0.64% (95% CI, −1.29% to 0.01%; P = .05) for resistance training vs control.

There was only 1 study²⁰ in which participation was limited to participants with diabetes who were older than 65 years. The age of the participants in this study was much higher than the overall mean (SD) age of the participants in this meta-analysis (69.4 [4.7] years vs 55.0 [7.2] years, respectively). The intervention in this study was not successful in reducing HbA_1c. If this study were excluded, the overall WMD for HbA_1c would have been −0.74% (95% CI, −1.09% to −0.39%).

In all but 2 of 14 trials, the details on the changes in body mass were given in kilograms (Raz et al¹⁴ and Vanninen et al²¹ only presented body mass index values). In the body mass comparisons for the 13 exercise groups vs nonexercise control groups, no significant postintervention differences were found (SMD, 0.06; P = .60; Figure 2). The effect of exercise vs nonexercise control was similar when only the 12 studies that measured body mass in kilograms were considered separately (83.02 kg vs 82.48 kg; WMD, 0.54 [95% CI, −2.91 to 3.99]; P = .76). There were no significant differences between the 2 groups when randomized controlled trials, CCTs, aerobic training studies, and resistance training studies were considered separately. In the studies comparing combined exercise and diet interventions vs control (Figure 2), the postintervention body mass difference also did not reach statistical significance (SMD, −0.20; P = .25).

The average baseline to postintervention changes in body weight were approximately −0.9 kg (P = .70) in the exercise groups, −3.4 kg (P = .11) in the combined exercise and diet groups, −2.5 kg (P = .29) in the diet groups, and 0.8 kg (P = .73) in the control groups.

Abdominal obesity was represented by waist-to-hip ratio or waist circumference. This information was available in 4 studies.^{12- 13,16,19} The postintervention WMDs were −0.02 U (P = .05) for waist-to-hip ratio and −4.53 cm (P<.001) for waist circumference. However, much of this difference could be accounted for by baseline differences in abdominal adiposity favoring the exercise groups (waist-to-hip ratio, −0.01; P = .40; waist circumference, −3.52 cm; P<.001). Only 1 study¹³ in our meta-analysis directly measured abdominal obesity by magnetic resonance imaging. The aerobic training program in that study (55 minutes, 3 times/week, 10 weeks) resulted in a significant reduction in both abdominal subcutaneous adipose tissue (227.3 cm² to 186.7 cm²; P<.05) and visceral adipose tissue (156.1 cm²to 80.4 cm²; P<.05) while no significant reductions were found in the control group. In the same study, waist circumference and waist-to-hip ratio were not significantly reduced (98.4 cm to 97.4 cm and 0.97 to 0.94 U, respectively). Only 2 studies, 1 of aerobic training¹³ and 1 of resistance training,¹⁶ presented data on the sum of skinfolds and neither study found a significant exercise effect on this outcome. There were no significant changes in body fat percentage within the 2 aerobic training studies^13,19 that measured this variable. In the study by Mourier et al,¹³ magnetic resonance imaging indicated that the mid-thigh muscle cross-sectional area was significantly increased after aerobic training (149.3 cm²to 183.5 cm²; P<.05).

The postintervention mean difference in body weight did not predict the postintervention mean difference in HbA_1c (r² = .003; P = .84 for model 1). Even after correcting for baseline values, no significant association between these variables was found (r² = .09; P = .31 for model 2). Exercise intensity (METs) was not associated with the postintervention mean difference in HbA_1c (r² = .07, P = .35 for model 3). Similarly, exercise volume (total MET hours) was not associated with the postintervention mean difference in HbA_1c (r² = .04; P = .51 for model 4). When body mass (SMD), exercise intensity (METs), and exercise volume (total MET hours) were entered simultaneously as independent variables (model 5), a nonsignificant 8% of the variance in HbA_1c was explained (P = .79).

The funnel plot technique³¹ was used to evaluate publication bias. The postintervention WMDs in HbA_1c were plotted against the sample size of the study. The plot did not show any asymmetry, an indication that significant publication bias was not likely. We did not find an unpublished study meeting the inclusion criteria. To statistically pool results from different studies using a fixed-effects model, the values must be relatively homogenous between studies. The results from the various analyses were consistent and homogeneous, however, there was 1 exception. The χ² tests suggested that there was heterogeneity in the baseline HbA_1c values for the combined exercise and diet vs control comparison (Figure 2). This analysis was repeated with a random-effects model; the baseline WMD was reduced to 0.02% (P>.99) and the postintervention WMD was not changed (WMD, −0.76% [95% CI, −1.32 to −0.20]; P = .008).

Postintervention HbA_1c values were significantly reduced in the exercise groups compared with control groups while body mass was not. The postintervention HbA_1c values were 0.66% lower in the exercise groups when compared with nonexercise control groups. A reduction in HbA_1c of this magnitude is clinically significant and close to the difference between conventional and intensive glucose-lowering therapy in the United Kingdom Prospective Diabetes Study (UKPDS). In the UKPDS, subjects receiving intensive treatment with insulin or sulfonylureas had HbA_1c averaging 0.9% below the conventional treatment (7.0% vs 7.9%; P<.001) and had significant reduction in diabetes-related clinical end points (40.9 vs 46 events per 1000 patient-years; P = .03).^48- 50 In UKPDS subjects randomized to intensive glycemic control with metformin, HbA_1c was only 0.6% lower than with conventional treatment but there were risk reductions of 32% (P = .002) for diabetes-related clinical end points and 42% for diabetes-related deaths (P = .02).⁴⁹ The potential importance of good glycemic control for the reduction of cardiovascular disease risk was supported in a recent meta-regression study, which demonstrated an exponential relationship between fasting glucose concentrations and the incidence of cardiovascular events.⁵¹ We speculate that a greater reduction in cardiovascular complications might be anticipated with exercise than with insulin or sulfonylureas in the UKPDS since, unlike these medications, exercise is associated with other cardioprotective benefits^{4,9,32,52- 54} and does not cause weight gain.

The meta-regression results suggest that the differences in HbA_1c found between the exercise groups and control groups after the intervention were not mediated by differences in weight loss, exercise intensity, or exercise volume. The finding that exercise does not need to reduce body weight to have a beneficial impact on glycemic control is clinically important. Exercise training decreases hepatic and muscle insulin resistance and increases glucose disposal through a number of mechanisms that would not necessarily be associated with body weight changes. The mechanisms were extensively reviewed recently by Ivy et al,⁵⁵ and include increased postreceptor insulin signaling,⁵⁶ increased glucose transporter protein and messenger RNA,⁵⁷ increased activity of glycogen synthase⁵⁸ and hexokinase,⁵⁹ decreased release and increased clearance of free fatty acids,⁵⁵ increased muscle glucose delivery due to increased muscle capillary density,^59- 61 and changes in muscle composition favoring increased glucose disposal.^59,62- 63

In the present meta-analysis, the exercise interventions produced no statistically significant reduction in body weight. There are several possible explanations for this. First, the exercise interventions were of relatively short duration and involved only moderate amounts of exercise. Second, exercise participants might have reduced their daily physical activities, partially counterbalancing the increased energy expenditure from the exercise intervention. Third, exercise group subjects might have increased their food intake, or decreased it less than control subjects. However, the studies' dietary intake records did not support this possibility. Fourth, in relatively inactive people, increasing physical activity can result in an increase in lean body mass.^64- 65 Therefore, it is certainly possible that the amount of weight loss did not fully reflect the amount of fat loss. More accurate measures of body composition, such as computed tomography, magnetic resonance imaging, or hydrostatic weighing, would be desirable in future studies to precisely measure changes in body composition.

The effect of exercise on HbA_1c and body mass was estimated from data obtained across different ethnicities (Northern Europeans, Southern Europeans, blacks, Asian, Middle-Easterners), medication status (no medication, oral hypoglycemic agents, insulin therapy), age groups, and dietary interventions. The results are therefore widely generalizable to middle-aged patients with type 2 diabetes. Because only 1 study²⁰ included many participants who were older than 65 years, we cannot be certain that the overall results are generalizable to people older than 65 years. Adherence rates to the exercise programs were relatively high in most studies (mean >80%, where reported). Adherence rates lower than these would presumably result in a lesser impact on HbA_1c.

There is little research on the effects of resistance training (such as weight lifting) in patients with type 2 diabetes; only 2 resistance exercise studies met inclusion criteria for this analysis. Several relevant resistance training studies were excluded from the present analysis because of the absence of an appropriate control group⁶⁶ or the inclusion of nondiabetic participants.^67- 75 In the present meta-analysis, the postintervention WMD for HbA_1c in the resistance training groups vs nonexercise control groups was similar to aerobic training groups vs nonexercise control groups (−0.64% [95% CI, −1.29% to 0.01%] and −0.67% [95% CI, −1.04% to −0.30%], respectively). Well-designed studies on the effects of resistance training and aerobic training are needed to better understand the impact of increasing muscle mass and reducing fat mass (especially visceral fat) on glycemic control and other metabolic abnormalities.

In conclusion, although the individual trials on the effects of exercise in patients with type 2 diabetes have had partially conflicting results, the current meta-analysis suggests that exercise training reduces HbA_1c by approximately 0.66%, an amount that would be expected to reduce the risk of diabetic complications significantly. The studies reviewed in this meta-analysis did not find significantly greater weight loss in the exercise groups compared with the control groups. Therefore, exercise should be viewed as beneficial on its own, not merely as an avenue to weight loss. Future research should include longer interventions with better quantification of body composition changes. In the interim, our analysis using an evidence-based approach adds support to the idea that exercise is a cornerstone of diabetes therapy.