Discussion
In this large contemporary multiethnic cohort study, the incidence rate of type 2 diabetes was 11.4 per 1000 person-years with variation of the incidence rate across racial/ethnic groups consistent with contemporary analysis in US middle-aged populations.18 ,19 Vigorous PA, exercise PA and faster walking pace are inversely associated with incident type 2 diabetes and leisure sedentary behaviors are positively associated with type 2 diabetes, consistent with previous studies in predominantly NHW cohorts.7 ,20–22 The independent protective associations of vigorous PA and activity score, along with the deleterious associations of leisure sedentary behavior on type 2 diabetes risk remained after full adjustment including BMI. A key finding is that the magnitude of the associations varied by race/ethnicity, and the statistical significance of the associations in the overall cohort is mainly driven by the NHW subgroup, as the AA, HA, and CA groups only achieved p<0.05 among AA for TV watching and HA for activity score. The duality of these findings combined with the greater burden of type 2 diabetes in racial/ethnic minority populations1 suggests other factors that vary by race/ethnicity may play important roles in the development of type 2 diabetes in non-NHW populations including diet,23 social determinants of health,24 ,25 chronic stress/allostatic load,25–27 sleep insufficiency,25 ,28 and inflammation.29–31 While our findings of potential racial/ethnic differences in the association of activity measures with type 2 diabetes incidence should be considered with some caution due to sample size variation in race/ethnic groups, there are some plausible biological mechanisms that may explain these findings. The response of glucose metabolism to PA has been measured in NHW populations with relatively consistent improvements in insulin sensitivity, even without weight loss, although this is debated.32 There are limited data on the glucometabolic response to PA in AA and HA. In the HERITAGE Family Study, a 20-week trial of thrice weekly exercise in AA and NHW, AA had a greater increase in fasting glucose with exercise versus NHW, although there were reductions in fasting insulin in both groups.33 Potential components of improved glucose metabolism with PA include skeletal muscle mitochondrial mass, function, and aerobic capacity.34 ,35 Aerobic capacity was tested in a 6-month trial of aerobic exercise training in AA and NHW postmenopausal women. The NHW women had a greater increase in cardiorespiratory fitness with sustained resting metabolic rate (RMR), whereas in AA women, RMR declined over the course of the trial.36 ,37
Among HA, non-obese, non-diabetic individuals have greater insulin resistance versus NHW, which persists even after accounting for chronic exercise, acute exercise, total and abdominal fat distribution.38 ,39 HA have higher levels of tumor necrosis factor-α (TNF-α) than NHW or AA, with increasing TNF-α being correlated with insulin resistance.40 ,41 The improvement in insulin sensitivity with a short-term diet–exercise intervention is similar in HA and NHW. However, because at the outset HA are less insulin sensitive compared with NHW, the ethnic gap in insulin sensitivity remained.42 Our study reflects these findings, with ethnic minorities having no statistically significant inverse associations with PA and incident type 2 diabetes.
Family history is a known risk factor for type 2 diabetes, functioning through gene–environment interactions.43 In this study, participants with no family history of diabetes had a trend toward greater inverse associations of PA, faster walking pace, and a higher activity score with type 2 diabetes incidence, whereas sedentary behavior increased type 2 diabetes risk irrespective of the family history of diabetes (see online supplementary table S5), although formal effect modification testing was only significant for television viewing time. Thus, sedentary behavior may function independent of gene–environment interactions. Studies of the relationship between PA and gene variants associated with worsened glucose metabolism and type 2 diabetes have been mixed with PA-specific analysis showing no interaction44 or diminished interactions,45 although lifestyle interventions, including diet and exercise with weight loss, have shown positive interactions.46 ,47
Mechanisms of PA, walking pace and sedentary behavior, and incident type 2 diabetes
PA reduces blood glucose acutely and chronically through insulin-dependent and insulin-independent actions including increased GLUT-4 mediated uptake of glucose into muscle; reduced insulin resistance; and reduction in body weight, adiposity, and inflammation, which have been extensively reviewed previously.48 ,49 Walking pace has been cross-sectionally associated with insulin resistance independent of obesity in non-diabetic older men.50 Equivalent energy expenditures from walking or vigorous PA result in comparable magnitudes of reduction in type 2 diabetes risk.20 Sedentary behavior has been associated with adiposity;7 adiposity-associated inflammation;51 and reduced lipoprotein lipase activity, clearance of triglycerides, clearance of an oral glucose load, and glucose-stimulated insulin secretion.52 Thus, sedentary behavior and vigorous PA have opposing effects on glucose metabolism. Lifestyle interventions such as the Diabetes Prevention Program (DPP) not only increase PA but also reduce sedentary behaviors including television watching,53 which is beneficial in improving glucose metabolism. In the lifestyle intervention of the DPP, all ethnicities had similar reductions in type 2 diabetes, which was significantly associated with the degree of weight loss.54 Weight loss is strongly associated with caloric reduction, whereas PA has a greater association with weight loss maintenance.55 The high correlation between weight loss and lower diabetes risk may explain some of our findings, given that AA and HA had higher BMI and waist circumference.
Strengths and limitations
There are several strengths of our study including a socioeconomically diverse, multiethnic US population with over a decade of follow-up, allowing broad generalizability of our findings, as well as use of a validated PA questionnaire with calculation of metabolic equivalents, and documentation of type 2 diabetes with fasting glucose, medication use, and physician diagnosis, unlike some prior studies that rely on self-reported diabetes. Despite the strengths of this study, there are several potential limitations. PA and sedentary behaviors were self-reported, and the time frame queried about was their typical activities within the last month. Therefore, discrepancies with actual activity levels may exist, and depending on the season in which participants were examined, the prior month may not be representative of customary activities. There is also potential discordance in PA measured using self-report versus objective instruments among assorted race/ethnic groups compared with NHWs. AAs self-report less PA,56 ,57 but AA men and women have similar levels of objectively measured PA.57 ,58 HAs self-report lower PA levels compared with NHWs, but have higher objectively measured levels compared with NHWs, which may be due to differences in work PA.59 ,60 Two UK-based diabetes prevention studies found that South Asians and NHW groups undertook similar levels of objectively measured PA despite self-reported estimates being ∼40% lower in the South Asian group.61 Overall, objective measures minimize the differences in moderate–vigorous PA among racial/ethnic groups.59 ,62 In our study, the only large differences in PA measure were for any vigorous activity and greater walking pace among NHWs. Another potential bias is the variation in sample size for the race/ethnic groups with potential power implications for detecting significant race/ethnic interactions in stratified analyses. Because of multiple comparisons, there is the potential for type II error. Thus, we accounted for multiple comparisons using a Bonferroni correction, which did not fundamentally alter the significance of our findings. We were unable to distinguish between type 1 and type 2 diabetes, but incident type 1 diabetes is extremely uncommon in older adults, so we assume a predominance of type 2 diabetes. The distinction of using metformin for prediabetes versus diabetes was not made. It is important to note that although prediabetes is an indication for metformin, not all patients with prediabetes would be eligible for this medication and the medication is rarely prescribed to those eligible in real-world practice. Only about 3.7% of 17 352 insured patients with prediabetes and eligible for metformin during the 2010–2012 period were actually prescribed the medication.63 Such a proportion is remarkably low; it is therefore reasonable to assume the majority of metformin was prescribed for diabetes.
Implications
The strong association between sedentary behavior and type 2 diabetes risk indicates that interventions may benefit from including components that specifically address reducing sedentary behavior. For example, sitting interspersed with short breaks of low-intensity activity (such as walking) or standing may prevent the deleterious effects of prolonged sitting but requires further study.64–66 Lifestyle interventions with PA and caloric intake reduction with subsequent weight loss are shown to be beneficial for type 2 diabetes risk reduction,54 but the independent association of activity measures and type 2 diabetes reduction remains unclear in minority ethnic populations. Given AA and HA have a higher risk of developing type 2 diabetes compared with NHWs, further study of increased PA and reduced sedentary behavior as methods of prevention are warranted to better understand the independent impact of these interventions in US racial/ethnic minority groups.