Physical Activity of Moderate Intensity and Risk of Type 2 Diabetes: A systematic review

We searched EMBASE and Medline through March 2006 for prospective cohort and cross-sectional studies investigating the association between moderately intense physical activity and incidence and prevalence of type 2 diabetes. The search terms “physical activity,” “exercise,” and “walking” were used in combination with “noninsulin dependent diabetes mellitus,” “NIDDM,” and “type 2 diabetes,” and references lists of retrieved articles were examined. Physical activity of moderate intensity was defined as requiring a metabolic equivalent task (MET) score of 3.0–6.0 (11). One MET corresponds to the energy expenditure during rest (quiet sitting). A typical activity of moderate intensity is “brisk” walking at 5.6 km/h (3.5 miles/h) on a flat surface requiring 3.8 MET (12). Other common activities of moderate intensity include playing golf, leisure bicycling at <16 km/h (10 miles/h), and gardening (12).

The EMBASE and Medline searches were performed independently by two investigators (C.Y.J. and R.P.L.), yielding 491 and 488 articles, respectively. Two additional relevant articles were found by examination of reference lists of retrieved articles. Studies were excluded if they did not involve human subjects, did not present age-adjusted estimates, involved study populations overlapping with other studies, or combined moderately intense activity with strenuous or light activity. Ten prospective cohort studies (13–22) and six cross-sectional studies (23–28) were eligible. Because of the large heterogeneity in exposure and outcome measures for the cross-sectional studies, a meta-analysis was only conducted for the cohort studies.

Information on study design, participant characteristics, definition of moderately intense physical activity, adjustments for potential confounders, and estimates of associations were abstracted independently by two investigators (C.Y.J. and R.P.L.). In all studies, estimates of relative risks (RRs) with information about their variance were reported. For studies that reported more than one level of moderately intense activity, the RR for the highest as compared with the lowest level of activity was abstracted.

We used STATA version 9.1 (STATA, College Station, TX) for all statistical analyses. Summary measures were calculated using random-effects models, which allow each of the studies to estimate a different effect size (29). P values for heterogeneity of study results were calculated using Cochran’s Q test (30). When available, we separately analyzed BMI-unadjusted and -adjusted results from the original studies to assess the role of physical activity independent of its association with weight. To examine sources of heterogeneity, we conducted metaregression analysis with the log RR of studies as dependent variable and country (U.S./other) and sex (male/female) as independent variables. We assessed potential publication bias using funnel plots, plots of RRs versus precision, and the Begg (31) and Egger (32) tests.

We identified 10 cohort studies on physical activity of moderate intensity and risk of type 2 diabetes, including a total of 301,221 participants and 9,367 incident cases. Table 1 shows the characteristics of the included studies. Fig. 1 shows the results for moderately intense physical activity and risk of type 2 diabetes with and without adjustment for BMI. For one study only BMI-unadjusted (13) and for one study only BMI-adjusted estimates were presented (22). The summary RR of type 2 diabetes without BMI adjustment was 0.69 (95% CI 0.58–0.83) for the highest as compared with the lowest category of moderate-intensity physical activity. The BMI-adjusted RR of diabetes was 0.83 (0.76–0.90). The P value for heterogeneity in study results was <0.001 for the BMI-unadjusted and 0.24 for the BMI-adjusted estimates.

In five of the cohort studies the specific role of walking was examined (13,15,19,21,22). Figure 2 shows the results for walking and risk of type 2 diabetes. The BMI-unadjusted summary RR comparing the highest with the lowest walking level was 0.70 (95% CI 0.58–0.84), and the BMI-adjusted RR was 0.83 (0.75–0.91). P values for heterogeneity were 0.08 for the BMI-unadjusted and 0.68 for the BMI-adjusted estimates. The reported amount of walking for the reference category was minimal, and the amount in the highest category was at least 10 MET h/week (15,22), which is equivalent to ∼2.5 h/week of brisk walking. Amounts of walking of 2–3 h/week (21), 2.1–3.8 MET h/week (15), and 5.1–10.0 MET h/week (22) have also been associated with a significantly lower risk of type 2 diabetes as compared with being sedentary.

All but two studies (16,22) considered confounding by vigorous physical activity by adjusting for vigorous activity or excluding participants that engaged in vigorous activities. The studies that did not consider potential confounding by vigorous activity did not consistently report stronger or weaker associations than the other studies.

The Egger and Begg tests provided no evidence for publication bias for the BMI-unadjusted (P = 0.63 and P = 0.84, respectively) and BMI-adjusted (P = 0.83 and P = 0.40) association between moderately intense physical activity and risk of type 2 diabetes.

The inverse association between moderately intense physical activity and type 2 diabetes was observed in populations from the U.S., Finland, and the U.K. (Table 1). In the Japanese study by Okada et al. (17), no inverse association was observed, but this may have been due to the moderate-intensity activity definition that was restricted to weekends, included relatively light-intensity activities, and did not consider duration. We conducted a metaregression analysis for the association between moderate-intensity physical activity and did not find significant differences in results between U.S. and non-U.S. populations (P = 0.20). The summary RR for the U.S. studies was 0.68 (95% CI 0.56–0.81), while that of the non-U.S. studies was 0.75 (0.56–1.00). After excluding the study by Okada et al. (17), the RR for non-U.S. studies was 0.66 (0.54–0.80) and the P value for difference by region 0.96. For the BMI-adjusted RRs, the P values for difference by region was 0.07 before and 0.75 after exclusion of the study by Okada et al. We also evaluated potential differences by sex. The BMI-unadjusted association between moderate-intensity physical activity and diabetes risk was significantly stronger for female (RR 0.58 [95% CI 0.51–0.65]) than for male (0.82 [0.70–0.96]) cohorts (P = 0.04). However, after excluding the study by Okada et al., the RR for male cohorts became 0.77 (95% CI 0.67–0.88) and the difference with the female cohorts no longer statistically significant (P = 0.36). For the BMI-adjusted RRs, the P values for differences between male and female cohorts was 0.17 before and 0.81 after exclusion of the study by Okada et al.

Six cross-sectional studies that examined the association between moderately intense activities and type 2 diabetes or impaired glucose tolerance (Table 2) were identified. Results from two Dutch studies suggested inverse associations for gardening and bicycling (which tends to be of easy pace in this population) but not for walking (24,26). In a French study, participation in moderately intense household activities was inversely associated with type 2 diabetes (25). Two other studies were consistent with an inverse association between moderately intense activities and type 2 diabetes, but CIs were wide (23,27). In an Australian study, 2.5 h/week of moderately intense activities as compared with less time spent on these activities tended to be associated with a lower prevalence of abnormal glucose metabolism in women but not in men (28).

In our meta-analysis of 10 prospective cohort studies, a substantial inverse association was observed between physical activity of moderate intensity and risk of type 2 diabetes. Those who regularly engaged in physical activity of moderate intensity had ∼30% lower risk of type 2 diabetes as compared with sedentary individuals. A similar decrease in diabetes risk was observed when we specifically examined regular walking. After adjustment for BMI, the reduction in diabetes risk remained substantial (17%) for both regular moderately intense activity and walking.

Significant inverse associations were observed in both men and women and in both U.S. and Northern European cohorts. Most studies were conducted in predominantly white populations. No significant association between moderate-intensity physical activity and type 2 diabetes was observed in the two studies that reported results for other ethnic groups, but this may have been due to the “light” definition of activity (17) and limited statistical power as a result of lower numbers for nonwhites (22). In the Diabetes Prevention Program, the lifestyle intervention that focused on both diet and physical activity resulted in a similar reduction in incidence of type 2 diabetes in participants of African-American, Hispanic, American-Indian, and Asian ethnicity as compared with whites (6,33).

Results from cross-sectional studies were generally consistent with an inverse association between moderately intense physical activity and type 2 diabetes. Results were more heterogeneous than for the prospective studies, possibly as a result of the smaller study sizes and differences in the activity classification. In two cross-sectional studies of Dutch elderly, no association was observed for walking, whereas other moderately intense activities were inversely association with type 2 diabetes (24,26). This lack of association for walking may have reflected a lower average walking pace in these elderly populations. A brisk usual walking pace was associated with a substantially lower risk of type 2 diabetes compared with an easy pace in two U.S. studies (15,19).

Because our meta-analysis included only observational studies, it is possible that the summary estimates were influenced by confounding and other biases. Although all studies adjusted for age, not all adjusted for known or suspected diabetes risk factors such as dietary factors, alcohol consumption, cigarette smoking, and waist-to-hip ratio. The inverse association was similar in studies with the most complete adjustment for confounding, but we cannot fully exclude residual confounding by unmeasured or imprecisely measured diabetes risk factors. Lack of adjustment for light-intensity physical activity or sedentary activities may also have resulted in residual confounding (15,19). Furthermore, the included studies mostly focused on leisure time physical activity, but commuting and occupational activities can also contribute importantly to the accumulation of moderately intense physical activity for the reduction of diabetes risk (20,34).

The prospective design of the cohort studies and low loss to follow-up in most studies reduced the likelihood of selection bias. Some misclassification of moderate-intensity physical activity probably occurred, but it seems unlikely that this misclassification differed by future diabetes outcome, and it can thus be expected to have biased estimates of associations toward the null. In the studies that relied upon self-report of a physician’s diagnosis of diabetes, differences in detection of diabetes associated with physical activity could have led to diagnostic bias. This is unlikely given that several cross-sectional studies measured glucose concentrations in all participants, and restriction to symptomatic cases did not substantially affect the results in a cohort study (15). We did not find any evidence of publication bias based on the Egger and Begg test and the funnel plot. However, the power of these tests is known to be limited (35), and we cannot fully exclude the possibility that publication bias has affected our results.

We found a significant inverse association between moderately intense physical activity and type 2 diabetes that persisted after adjustment for BMI. In line with this finding, biological mechanisms have been identified for beneficial effects of physical activity on glucose metabolism independent of body fatness. Exercise has been shown to increase insulin-stimulated glycogen synthesis through an increased rate of insulin-stimulated glucose transport by GLUT4 glucose transporters and increased glycogen synthase activity (36). In addition, elevated capillary proliferation in muscles, increased muscle mass, and a higher proportion of more insulin-sensitive types of muscle fibers may contribute to beneficial effects of physical activity on insulin sensitivity (37).

No large randomized trials have specifically investigated the effect of increasing physical activity of moderate intensity on incidence of type 2 diabetes. A randomized trial of moderate-intensity physical activity in individuals with a family history of diabetes did not find a significant reduction in incidence of type 2 diabetes after 2 years, but compliance with the program was poor and the number of participants small (n = 37 in the exercise program) (38). In a 2-year randomized controlled trial in 179 individuals with type 2 diabetes, counseling to achieve 10 MET h/ week of moderately intense physical activity resulted in significantly reduced body weight and fasting glucose and A1C concentrations (39). In a post hoc analysis, a significant reduction in A1C was observed for participants that increased 11–20 MET h/week, but greater beneficial effects were observed in those who increased their activity with 21–30 MET h/week (39). Several randomized controlled trials tested the effects of brisk walking on glucose metabolism in individuals without diabetes (40–42). Increased walking resulted in reduced plasma insulin (40,41) or glucose (42) concentrations after 12–24 weeks.

A Chinese trial in individuals with impaired glucose tolerance found that those randomized at the clinic level to a combination of moderate and vigorous activities had a 47% reduction in incidence of type 2 diabetes as compared with the control group (7). A post hoc analysis of the Finnish Diabetes Prevention Study examined the association between walking during follow-up and incidence of type 2 diabetes. After adjustment for other activities, dietary factors, and BMI, at least 2.5 h/week of walking for exercise was associated with a 63% lower risk of type 2 diabetes as compared with <1 h/week (43). In addition, an increase in walking intensity was associated with a lower risk of type 2 diabetes.

Our systematic review indicates that regular participation in moderately intense physical activity is associated with a substantially lower risk of type 2 diabetes. The association was partly independent of BMI, suggesting that moderate-intensity physical activity can reduce the risk of type 2 diabetes even in those who do not achieve weight loss. Findings from several prospective studies (15,21,22,43) indicate that 30 min or more of daily moderate-intensity activity, as recommended in multiple U.S. guidelines (11,44), can substantially reduce the risk of type 2 diabetes as compared with being sedentary. Moderately intense activity as defined in guidelines (3.0–6.0 MET h) (11) includes walking at brisk pace but not walking at an easy or casual pace (12), and walking at brisk pace also seems preferable for the prevention of type 2 diabetes (15,19,43). Further studies are needed to define more specifically what combinations and duration and pace are optimal for reducing the risk of type 2 diabetes. However, given that only 31% of adults in the U.S. currently meet the general physical activity recommendations (45), efforts to prevent type 2 diabetes should strongly emphasize the benefit of moderately intense physical activities and encourage wider participation in these activities.