In 2005, an estimated 20.8 million Americans (7.0% of the total population) had diabetes-14.6 million with diagnosed diabetes and 6.2 million with undiagnosed diabetes (8). The number for diagnosed diabetes represents more than a doubling from 1980 (5.8 million) (9). Continued rapid growth is expected in the future as an estimated one in three Americans born in 2000 are expected to develop diabetes during their lifetimes (9), whereas the prevalence of diagnosed diabetes is predicted to increase approximately 165% between 2000 and 2050, from 11 million (4.0% of the population) to 29 million (7.2%) (4). Contemporary prevalence estimates reveal that approximately one in five adults 65 yr and older have diabetes. Prevalence is also higher by race or ethnicity. African American, Hispanic, and American Indian/Alaska Native adults have a diabetes prevalence almost two to three times greater than white adults (8,9).
Diabetes was the sixth leading cause of death in the United States in 2002 (8); information from death certificates indicates it was the underlying cause of 73,249 deaths and a contributing cause in an additional 224,092. The disease is very serious when uncontrolled because of the damage it causes to end organs (8). For example, diabetes can cause cardiovascular diseases (CVD), such as coronary heart disease and stroke, which account for approximately 65% of deaths among people with diabetes. Persons with diabetes are likely to develop an array of other complications. Diabetes is the leading cause of new cases of blindness among Americans aged 20-74 yr (8), and in 2002, almost 44,000 Americans with diabetes developed kidney failure (8). Amputation of a leg, foot, or toe is very common (annual average of 82,000 such procedures on persons with diabetes in 2002). In addition, diabetes increases the overall risk of physical disability among older adults (13). Not surprisingly, the disease is very costly. In 2002, diabetes was responsible for an estimated $132 billion in costs-$92 billion in direct medical costs and $40 billion in indirect costs (e.g., work loss, disability, premature death).
Because of the tremendous personal, social, and economic burdens imposed by diabetes, we must seek to prevent the disease whenever possible. In the first Surgeon General's Report on Physical Activity and Health, we note this promising quote regarding the potential to prevent diabetes: "Physical activity reduces the risk of premature mortality in general, and of coronary heart disease, hypertension, colon cancer, and diabetes mellitus in particular" (31). In addition, we can point to a variety of organizations that have issued position statements describing the role of physical activity in preventing and treating diabetes (1,2,5,28). Most of these pronouncements, however, have not referred specifically to walking, the most common form of physical activity (31).
Because this review is intended to describe the public health impact of walking on diabetes, we focus on epidemiology. In this article, we will offer some definitions of diabetes, physical activity, and walking; review the evidence for drawing causal inferences and issuing position statements pertaining to physical activity and diabetes as a way of framing the data on walking and diabetes; offer some insights into the dose-response relationship between walking and diabetes; discuss confounding variables and biases in the association of walking with diabetes; and provide some recommendations for future research.
DEFINITIONS OF DIABETES AND PREDIABETES
Diabetes arises because of limited production of insulin or a reduced capacity to use this hormone; insulin potentiates the transfer of glucose into cells to be converted to energy (8,9). Diabetes is defined by a blood glucose concentration of >126 mg·dL−1 after an overnight fast or >200 mg·dL−1 after a 2-h oral glucose tolerance test. Oftentimes symptoms of diabetes (polyuria, polydipsia, and unexplained weight loss) in the presence of a casual plasma glucose value of >200 mg·dL−1 can identify those with diabetes. One of the two main types, type 1 diabetes, most often occurs during childhood or adolescence. This type of diabetes occurs through autoimmune destruction of beta cells in the liver, making insulin unavailable. Type 2 most often appears in adults 40 yr and older and is associated with obesity and physical inactivity. Type 2 diabetes arises when cells become resistant to the effects of insulin and when there is a relative deficiency of insulin. Type 2 diabetes accounts for 90-95% of all cases, and thus this article will be limited to that form of diabetes.
In addition to the almost 21 million Americans with diabetes, an estimated 41 million adults aged 40-74 yr have prediabetes (8). These women and men have an elevated blood sugar concentration that is not so high as to be termed diabetes. People with prediabetes are at high risk of developing diabetes and have been having one or two major kinds of impaired glucose metabolism: impaired fasting glucose, defined by a blood glucose concentration of 100-125 mg·dL−1 after an overnight fast; and impaired glucose tolerance (IGT), indicated by a blood glucose value of 140-199 mg·dL−1 after a 2-h oral glucose tolerance test.
In recent years, there has been considerable interest in the concept of a "metabolic syndrome," whose purported underlying physiological basis is insulin resistance and which is usually defined operationally as a combination of multiple risk factors for CVD (e.g., hypertriglyceridemia, low levels of high-density lipoprotein cholesterol, high blood pressure), diabetes, and excessive adiposity (18). Although this syndrome has captured the attention of many in the fields of diabetes and CVD, a very recent position statement of the American Diabetes Association (ADA) indicates that until clear research emerges to ensure that this syndrome is relevant, clinicians should simply evaluate and treat all existing CVD risk factors and diabetes without diagnosing the syndrome itself (18). Accordingly, this article will not address this topic.
DEFINITIONS OF PHYSICAL ACTIVITY AND WALKING
Physical activity is defined "as any bodily movement produced by skeletal muscles that results in energy expenditure" (7); this expenditure can be measured in kilocalories. Physical activity in daily life can be categorized into occupational, sports, conditioning, transportation, household, or other activities. Exercise is a subset of physical activity that is planned, structured, and repetitive and has as a final or intermediate objective of the improvement or maintenance of physical fitness. Physical fitness is a set of attributes that one possesses or achieves and is generally related to either health or skill (7). Walking, a specific mode of physical activity, has been defined as "an act or instance of going on foot especially for exercise or pleasure" (23).
Whether physiologic or epidemiologic evidence is being considered, to understand the relationship between physical activity (or walking) and diabetes, one must consider a variety of factors: (i) it is important to measure physical activity accurately; (ii) at present, we are unable to accurately measure walking during all types of physical activity; (iii) we must recognize the many ways in which physical activity data may be expressed as a summary score; and (iv) we must understand the relationship between relative and absolute individual intensity levels (6). Ideally, all forms of walking should be assessed, but doing so is not easy. Walking may entail a large or small part of work (e.g., postal delivery), sports (race walking), conditioning (to enhance cardiorespiratory endurance), transportation (to/from school or shopping), household duties (carrying loads of clothing to and from a laundry area), recreation (hiking), or other activities. Unfortunately, these types of walking are not measured equally well. Assessing separate and conjoint amounts of walking is complicated, which has the effect of complicating the study of things such as dose-response relationships. Fortunately, it is possible to use walking to define fairly precise amounts and intensities of a stimulus in physiological studies, although epidemiologic studies have been able to assess walking via self-report (14,16,17,29,34).
After they have gathered data from the constituent parts of daily or weekly life experiences (leisure, occupational, household, etc.), researchers may calculate a summary score from the duration (e.g., hours or minutes per session), frequency (sessions per week, month, or year), and estimated intensity (degree of effort or metabolic cost) of the activities performed (20). Calibrating intensity may involve assessing the perceived effort of physical activity, which can be judged by a participant's perceived increases in breathing or heart rate (14) or by the production of sweat (16). It is also possible to assess the intensity of a usual walking pace using descriptive terms (e.g., slow, brisk, fast) that are tied to listed walking speeds (16,17,29) or by comparing the pace to that of others of similar age and sex (i.e., faster, the same speed, slower) (3). In addition, one may assign intensity codes to reported physical activities that are known to vary in intensity when performed by the average person (20). Intensity is often expressed as multiples of a MET (metabolic equivalent), which represents the metabolic rate of a person at rest and is set at 3.5 mL oxygen consumed·kg−1 body mass·min−1, or approximately 1 kcal·kg−1·h−1 (7).
One way of calculating summary estimates of energy expenditure per week is to multiply the hours per week of participation in various activities by their assigned intensity codes (in METs) to yield MET-hours per week (16,17,29,34) or MET-minutes per week, but this does not consider the participant's body weight. If that is known, one can calculate kilocalories per week (kcal·kg−1·wk−1). A simpler summary score would be time spent in physical activity (e.g., hours or minutes per week) (14).
CAUSAL INFERENCES FOR PHYSICAL ACTIVITY AND DIABETES
There are limited epidemiologic data that ties walking to either the prevention of incident diabetes or to improved mortality outcomes in persons who already have the disease. Accordingly, it is informative to explore how documents such as the First Surgeon General's Report on Physical Activity and Health (31) have concluded that physical activity, in general, might have the potential to reduce the risk of incident diabetes. We can do so by using six criteria for making a causal inference as used by Powell et al. (25) in a 1987 article to argue that physical activity reduces the risk of coronary heart disease as used in the Surgeon General's Report, while also highlighting more recent evidence that further supports each criterion: (i) There are plausible, coherent biologic mechanisms by which physical activity may affect diabetes, as such activity may decrease insulin resistance and increase glucose disposal (via postreceptor insulin signaling, enhanced enzyme activity, better removal of free fatty acids, better delivery of glucose to cells, and changes in muscle composition) (28). In addition, one can assume that physical activity will help forestall some of the major sequelae of diabetes, such as CVD. For example, a recent consensus conference noted that people who exercise may produce at least short-term improvements in total fat (but not abdominal fat) (27), that daily, moderate-intensity activity (but not high-intensity activity) may produce beneficial effects on blood pressure (11), and that physical activity at least 3 d·wk−1, lasting 30 min per occasion that is of at least moderate intensity may improve high-density lipoprotein cholesterol (21). (ii) The strength of the association between physical activity and the incidence of diabetes has also been summarized in a review by Hamman (15) of more than 14 prospective observational studies from around the world that included data from Malta, England, Finland, Japan, Sweden, and the United States. All but one of the studies showed reduced risks (21% to 60%; median, approximately 30%) for incident diabetes associated with physical activity. This magnitude of effect for physical activity is what one often finds in diseases having multiple causes (25). (iii) Consistency in the evidence for a link between physical activity and the incidence of type 2 diabetes is also found in those 14 studies whose data represent many different populations from many different countries throughout the world (15). (iv) Temporality was demonstrated by observational studies in the review by Hamman (15) that assessed the level of physical activity before finding cases of incident diabetes (15). (v) The biologic gradient, or dose-response effect, for physical activity was demonstrated in the variety of studies in which increasing amounts of such activity produced greater reductions in incident diabetes (15) (Fig. 1 as an example). (vi) Experimental evidence exists at least for Chinese (24), Finnish (30), American (19), and Indian (26) studies of high-risk adults (persons with IGT) who used physical activity and diet to reduce risk of incident diabetes via modest weight loss. In fact, the Da Qing (China) Study (24) found similar reductions for exercise alone as for diet alone. As such, the experimental evidence for physical activity and diabetes has been better than that used in reaching the inference that physical activity reduces risk of coronary heart disease (25) or other chronic diseases (31).
Because the criteria for concluding that physical activity has a positive causal effect on diabetes have been met, it is reasonable that several professional groups have developed guidelines recommending physical activity for preventing, treating, and controlling diabetes: the American College of Sports Medicine (1), ADA (2,28), and the Canadian Diabetes Association (5). Across the three groups, the recommendations for physical activity are that it be performed >3 d·wk−1, be of at least moderate intensity, last 20-60 min·d−1, and total 150 min·wk−1 or more. At least one group, the ADA (28), recommends 4 h·wk−1 or more to achieve greater reductions in risk of CVD and 7 h·wk−1 or more of physical activity when the goal is to maintain substantial weight losses (>13 kg).
This brief review of the evidence for drawing a causal inference about physical activity and diabetes, combined with the recommendations from leading professional groups, provides a context to review the limited epidemiologic data on walking and diabetes.
WALKING AND THE INCIDENCE OF DIABETES
Evidence for the association between walking and the incidence of diabetes comes from only two prospective studies of women, whose measures of walking exposure differed (Table 1). A study by Weinstein et al. (34) assessed, in 1992, the weekly hours spent in walking as reported by 37,828 women, aged 45 yr and older, who were participating in the Women's Health Study. These women were observed, on average, for approximately 6.9 yr, and the authors identified 1361 incident cases of diabetes among them. The authors used five categories of weekly hours spent walking-0, <1.0, 1.0-1.5, 2.0-3.0, and >4.0-whose gaps between some of the categories resulted primarily from the need for correspondence with the study's questionnaire. The analyses adjusted for total energy expenditure in kilocalories spent per week in all activities other than walking. Additional adjustment was made for age, body mass index (BMI), smoking status, family history of diabetes, use of alcohol, hormone therapy, hypertension, high cholesterol, dietary factors, and treatment group. For the five categories described, the associated hazard ratios (after adjustment) were 1.0 (referent), 0.95, 0.87, 0.66, and 0.89, respectively. Only the hazard ratio of 0.66 for 2.0-3.0 h·wk−1 (reflective of a 34% reduced risk of developing diabetes) was statistically significant, however.
The second study, by Hu et al. (16), examined quintiles of energy expended weekly in walking and in self-reported walking pace among 70,102 women aged 40-65 yr who were observed from 1986 to 1994 in the Nurses' Health Study. The authors restricted their analyses to women who did not perform vigorous physical activity. After adjustment for age, BMI, cigarette smoking, period of data collection, menopausal status, parental history of diabetes, consumption of alcohol, history of hypertension, and history of hypercholesterolemia, the authors reported that for each increasing quintile of energy expended through walking (lowest quintile as the referent), the relative risk of incident diabetes was 0.95, 0.80, 0.81, and 0.74, respectively. Only the relative risks for the third and fifth quintiles differed significantly from the referent, but across all quintiles, the linear trend was significant (P = 0.01). The median energy expenditure (MET·h·wk−1) for the various quintiles was <0.5, 1.7, 3.0, 7.5, and 20.0, respectively. Thus, if a brisk walk requires 3.0 METs of expended energy, the weekly time spent in the third and fifth quintiles might be approximately 1.0 and 6.7 h·wk−1, respectively, to produce the observed decreased risks of incident diabetes of 20% and 26%. Again, for the third and fifth quintiles, the time spent (h·wk−1) for fast walking (4.0 METs) might be 0.75 and 5.0, whereas for very fast walking (5.0 METs), the corresponding weekly times might be 0.6 and 4.0, respectively. Hu et al. (16), using the same sample of women, also explored a different walking index, namely, self-reported usual walking pace, and linked it to the risk of incident diabetes. Using a similar analytic approach as mentioned, with similar adjustments, but also adjusting for total time spent in walking, the authors found that women who reported a normal walking pace (approximately 2.0-3.0 mph) or brisk or very brisk walking (>3.0 mph), when they were compared with women reporting an easy walking pace (<2.0 mph), had lower relative risks of incident diabetes of 0.86 and 0.59, respectively. Only the fastest walking pace, however, yielded a significant result amounting to a 41% reduced risk.
WALKING AND OUTCOMES FOR DISEASE AMONG PERSONS WITH DIABETES
Evidence for the association between walking and outcomes of disease among persons with diabetes is drawn from only four studies, which used different indices ofwalking exposure (3,14,17,29) (Table 1). Of these four studies, three (3,14,29) examined associations with all-cause or total mortality. The study by Tanasescu et al. (29) linked quintiles of energy expenditure (MET·h·wk−1) with total mortality among 2803 men aged 30 yr and older who were participating in the Health Professionals Study. The authors adjusted for multiple possible confounders: age; BMI; smoking; intake of alcohol; family history of myocardial infarction; use of vitamin E supplements; duration of diabetes; medication for diabetes; quintiles of dietary intake of trans-polyunsaturated fat, saturated fat, fiber, and folate; history of angina and coronary artery bypass graft; baseline presence of hypertension and high serum cholesterol; and vigorous activity. Versus the first quintile (referent), the relative risk of all-cause mortality for quintiles two through five was 0.99, 0.96, 1.08, and 0.60, respectively. Only the fifth quintile differed significantly from the referent, and its contribution made possible a significant linear trend (P = 0.004) in the presence of these otherwise predominantly null findings. Accordingly, one may conclude that there was a meaningful (and, in this case, sizable) decreased risk of total mortality only among men who expended a rather substantial amount of energy through walking. The amount of walking in this quintile amounted to >16.5 MET·h·wk−1 or approximately >5.3 h·wk−1 of brisk walking, if one assumes that might require at least 3.0 METs of energy expenditure. For faster walking efforts, for example, fast walking (at 4.0 METs) or very fast walking (at 5.0 METs), the time spent weekly would be 4.1 and 3.3 h·wk−1, respectively.
A second study, this one by Gregg et al. (14), linked self-reported weekly hours spent walking with all-cause mortality among 2896 men and women aged 18 yr or older who participated in the 1990 or 1991 National Health Interview Survey. Decedent status was ascertained in a follow-up study (14). For the six categories of weekly hours spent walking, which were 0, 0-0.9, 1.0-1.9, 2.0-2.9, 3.0-3.9, and >4.0, the hazard ratios were 1.0 (referent), 1.0, 0.86, 0.61, 0.46, and 0.72, respectively, after adjustment for age, BMI, smoking, sex, race, self-rated health, approaches to losing weight, hospitalizations, hypertension, and use of antihypertensives, physician visits, limitations caused by CVD and cancer, and level of functional limitation. All death rates were also age-adjusted to the US population with diabetes. Although this study found both a somewhat progressive decrease in the hazard ratio, as time spent walking increased, and a significant linear trend (P = 0.004), only the two categories of walking between 2.0 and 3.9 h·wk−1 were tied to a significantly reduced risk of all-cause mortality. Given our very limited ability to compare this study with the results of Tanasescu et al. (29), one might suggest that as much as 2-5 h·wk−1 or more of walking might result in reductions of total mortality of anywhere from 40% to 55% among persons with diabetes.
The study by Gregg et al. (14) used a second index, namely, the perceived intensity of walking, which was measured by self-reported increases in breathing or heart rate during walking and examined the association with CVD mortality. Versus those who reported not walking (the referent), walkers who reported no increase in breathing or heart rate or some, moderate, or large increases in these rates had hazard ratios of 0.95, 0.69, 0.57, and 0.93, with the middle two categories significantly different from the referent. The authors offered as a possible reason for the failure of the most apparently intense (by self-report) form of walking to produce a significantly different hazard ratio the chance that persons reporting large increases in breathing or heart rate had some underlying coronary ischemia or other undetected cardiovascular or lung disease (14). In fact, after excluding the first 2 yr of follow-up experience for these men and women as a means of eliminating those with undetected morbidity, the authors found that the amount of risk reduction forthis group actually increased (14). It is important to recognize also that, when the authors controlled for physical activity participation other than walking and excluded those persons who had participated in moderate or vigorous activity or who had performed >2.0 h·wk−1 of nonwalking, there was no appreciable effect on the hazard ratios.
A third study, by Batty et al. (3), also examined the intensity of walking, but in this case, the authors used a simple self-comparison of regular walking pace to the pace of other, similar men. Among 6408 men aged 40-64 yr, 352 had either diabetes or IGT (indicative of extremely high risk for developing diabetes). The authors' analytic procedure excluded deaths during the first 5 yr of follow-up and involved full adjustment of the rate ratios (comparing risk for mortality) for age, BMI, smoking, social grade, systolic blood pressure, cholesterol, forced expiratory volume in 1 s, and disease at study entry (bronchitis, dyspnea, intermittent claudication, ischemia, or physician-diagnosed heart problems or high blood pressure, and unexplained weight loss in the preceding year). The authors found that, when compared to the reference group of men who believed their walking pace to be faster than that of other men, there were higher rate ratios for all-cause mortality among men who thought their walking pace was the same (1.11) or slower (2.28) than others. Only the latter was significantly different from the referent group, however. When we divided each ratio rate by 2.28 to make comparisons to other studies in this review, which used the least active study group as a reference point, we estimated that this might correspond to a rate ratio of 0.44. Assuming our recalculated risk is reasonably accurate, the associated risk reduction for faster walkers (vs slower walkers) would be approximately 56%, which would most closely correspond to the 43% lower risk for persons reporting moderate increases in breathing or heart rate (vs those reporting no walking) seen in the study by Gregg et al. (14).
Three studies of persons with diabetes linked walking to CVD mortality (14), mortality from coronary heart disease (3), or fatal and nonfatal CVD events (17). In one, Gregg etal. (14), using a similar analytic strategy and statistical adjustments to that described in the above review of their article, and found a significantly lower hazard ratio of 0.47 for deaths coded for underlying CVD for men and women who reported walking of 3.0-3.9 h·wk−1 (nonwalkers as referent), a 53% decrease in risk. Hu et al. (17), in a study reported 2 yr after their previous study (16) described in the previous paragraphs, studied 5125 women, aged 30-55 yr, who participated in the Nurses' Health Study and assessed their walking from 1976 to 1992 to form quartiles of energy expended (MET·h·wk−1). The authors identified incident cardiovascular events, including fatal and nonfatal myocardial infarction and stroke, occurring after 1980 and through the end of follow-up in 1994. After adjustment for age (in 5-yr categories), BMI, cigarette smoking, period of data collection, menopausal status, parental history of myocardial infarction before the age of 60 yr, use of multivitamins, supplementation with vitamin E, consumption of alcohol, history of hypertension, history of hypercholesterolemia, and use of aspirin, and after excluding women who exercised vigorously, the authors found that, relative to the first quartile of walking exposure, those in the second, third, and fourth quartiles had relative risks of cardiovascular events of 0.85, 0.63, and 0.56, respectively. The linear trend was significant (P = 0.03), but only the highest quartile approached significance versus the referent [relative risk = 0.56, 95% confidence interval (CI) = 0.31-1.00]. The estimated median energy expenditure for this group was 20.7 MET·h·wk−1. Assuming that 3.0 METs is the energy expended for a brisk walk, the highest quartile might require 6.9 h·wk−1 to decrease their risk of CVD events by approximately 40%. After the convention used above to interpret the results of Tanasescu et al. (29), the weekly time spent in faster walking (h·wk−1) might correspond to 5.2 for fast walking (at 4.0 METs) and 4.1 for very fast walking (at 5.0 METs).
Batty et al. (3), following their approach of excluding deaths in the first 5 yr of follow-up and making adjustments similar to those they made for total mortality, linked self-reported walking pace to mortality specific to coronary heart disease in their sample of middle-aged men. Using men who believed they walked faster than other men as the referent, the authors found that men who reported the same or a slower pace had significantly higher rate ratios of 2.18 and 4.25, respectively. When we again inverted the rate ratios to make the slower walking pace the referent group, as was done in other studies in this review, we estimated that the ratios for men reporting a faster walking pace or the same walking pace would be 0.46 and 0.24 with corresponding risk reductions of 54% and 76%, respectively.
IMPLICATIONS FOR PUBLIC HEALTH
The three studies that linked different walking indices to all-cause or total mortality studies (3,14,29) yielded findings that were generally similar to those produced from the three studies that linked walking to various indices of CVD mortality (3,14) or events (17). From the first group, one might conclude that 2.0 to 5.3 h·wk−1 of walking (or more) might yield reductions in mortality of anywhere from approximately 40% to 54%, and if walking is at a moderate or even faster pace, mortality reductions of anywhere from approximately 40% to 60% might be obtained. From the three studies linking walking indices with CVD mortality or events, one might conclude that at least 3.0 to 7.0 h·wk−1 or more of walking might yield reductions in mortality of anywhere from approximately 45% to 55%. Fast-paced walking, however, might yield reductions in mortality from coronary heart disease by as much as 76% (3). As weekly time allows the best conceptualization of dose-response as a basis for issuing recommendations for walking behavior, from the few studies that have examined either all-cause or CVD mortality or fatal and nonfatal CVD events, approximately 2 to 3 h·wk−1 of brisk walking might be the minimum worth encouraging among persons with diabetes. Hence, we note that public health recommendations for physical activity and health outcomes in diabetes seem generally also to apply to walking as a subpart of physical activity, whether for incident diabetes or for disease outcomes among persons with the disease.
ISSUES OF CONFOUNDING IN EPIDEMIOLOGIC STUDIES
It is important to consider the effect of confounding on the association between walking exposure and incident diabetes or mortality in persons with diabetes. By definition, confounding may distort the true nature of the association (6). The six studies described in the previous sections (3,14,16,17,29,34) all adjusted for age and BMI. Accounting for the influence of age, for example, on the relationship between walking and diabetes is important because total amounts of physical activity generally decline with age (31), while the risks of diabetes, total mortality, and CVD mortality all increase with age (15,25). Consequently, a crude association between low physical activity and increased risk of disease may be traceable, in part, to the influence of age.
Although the extent to which age was confounding the association between walking and incident diabetes was not explicitly studied by Weinstein et al. (34) or by Hu et al. (16) in their studies reviewed above, Folsom et al. (12) found that age seemed to exert an essentially constant effect on the association of physical activity with incident diabetes across age groups. In a cohort of postmenopausal Iowa women, Folsom et al. (12) showed that the relative risk of developing diabetes comparing low, medium, and high levels of physical activity remained graded, inverse, and of similar magnitude when stratified by three 5-yr age groups (55-59, 60-64, and 65-69 yr). The range in relative risk in the three age groups for the medium (0.73-0.76) and high (0.54-0.62) categories of physical activity was relatively narrow (these risks were calculated using low activity as the referent), suggesting a constant beneficial effect across the age of 15 yr. These findings are particularly interesting when they are compared with those of the Diabetes Prevention Program (19), which indicated that lifestyle intervention (physical activity and diet intended to produce modest weight loss) among high-risk adults having IGT yielded significantly better results than were obtained from the drug metformin, with seemingly greater benefit with increasing age (the age groups were 25-44, 45-59, and >60 yr) (19). Hence, it may be worth examining whether the effect of walking alone on incident diabetes remains constant across ages for women (and for men), especially when a broader age range is examined.
Walking is part of a total physical activity pattern. Hence, it may be difficult to examine the completely unique contribution of walking to disease outcomes such as diabetes. This is especially the case, if walkers are more physically active in other modes, such as gardening, or in other, more vigorous pursuits such as jogging. As noted in this review, however, most of the studies made some form of adjustment for the confounding influence of other modes of physical activity. For example, Weinstein et al. (34) adjusted for total energy expenditure in kilocalories spent per week in all activities other than walking. Hu et al. (16) restricted their analyses to women who did not perform vigorous physical activity. In addition, Hu et al. (16) also adjusted for total time spent in walking when examining the association of intensity of walking on diabetes incidence. Tanasescu et al. (29) adjusted for vigorous activity in examining the association of the volume of walking energy expenditure on all-cause mortality among persons with diabetes. Finally, Gregg et al. (14) controlled for physical activity participation other than walking and also excluded those persons who had participated in moderate or vigorous activity or who had performed >2.0 h·wk−1 of nonwalking in exploring the influence of walking intensity on CVD mortality among persons with diabetes. Future research should continue to explore the unique contribution of walking on diabetes outcomes (or its complications) by incorporating adjustments for other nonwalking activity and vigorous physical activity participation, as appropriate.
Adjustment for a confounding variable presumed to be in the causal path will, in most instances, diminish the association between the exposure and outcome. For walking and diabetes, BMI provides an excellent illustration. In the Women's Health Study (33,34), the authors found that when BMI was added to a multivariate-adjusted model, the observed protective effect (measured as hazard ratios) for weekly walking time on incident diabetes diminished for each of four walking categories. Using no walking as the referent, these hazard ratios were (h·wk−1) <1.0 (0.82 to 0.95), 1.0-1.5 (0.68 to 0.87), 2.0-3.0 (0.49 to 0.66), and >4.0 (0.64 to 0.89) (see also Fig. 1). This finding suggests that total adiposity may be presumed to be in the causal path between walking and incident diabetes. We should note, however, that other adiposity measures more proximally aligned to the etiology and development of diabetes (e.g., central adiposity, such as waist circumference) may be expected to reduce the observed relative risks even more when they are added to adjustments already including BMI. Evidence supporting this phenomenon comes from the recent intervention study by Di Loreto et al. (10), which assigned men and women to a physical activity group intended to reach 10 MET·h·wk−1 (or approximately 3.3 h·wk−1), primarily via brisk walking. The levels of activity that the participants actually achieved produced progressively increasing declines in measures of adiposity as the dose increased toward the highest observed level of 40 MET·h·wk−1. Weight loss (kg) decreased up to approximately 3.75% and waist circumference (cm) decreased even more-up to approximately 7% of baseline levels. Because total mass reflects gains in fat-free mass, the total amount of fat loss may have been greater. Waist circumference, however, may be better than total mass in indicating the risk of metabolic problems, because it may better reflect lipid accumulation in central storage areas. In essence, these measures of adiposity may help to depict more accurately the causal chain and, therefore, help to focus attention on important strategies for reducing the incidence of disease. The role of central adiposity, in particular, and of other variables in the causal path of walking and diabetes remains to be explored further.
Other potential confounding variables may also be considered, as when Wannamethee et al. (32) examined the effect of adjusting for certain variables on the association of total amounts of physical activity with incident diabetes. With age and BMI in their model, they noted that a significant but attenuated association prevailed for their two highest activity categories (moderate, moderately vigorous/vigorous) when adjustment was made for insulin used as a marker of overall insulin resistance. The attenuation continued to the point of nonsignificance, however, with additional adjustment for γ-glutamyltransferase-an enzyme that serves as a potential marker of hepatic insulin resistance. Hence, this analysis serves as a nice example of using epidemiologic approaches to amplify the evidence on the potential physiological effects of physical activity (in this case, they apparently occur via insulin resistance) at the population level. It would be informative to see how walking would fare in a similar analysis.
ISSUE OF BIAS IN EPIDEMIOLOGIC STUDIES
Biases distort the true nature of epidemiologic findings. Biases vary with the type of epidemiologic design, the type of disease or health outcome, the exposure to physical activity, other risk factors of interest, the population being investigated, or combinations of these variables (6). One form of bias thought to influence the observational cohort studies covered in this review is differential misclassification of exposure (e.g., walking) by disease status. In such studies, persons are excluded from the study or in its subsequent analyses when they already have the disease to be examined. This helps to remove the bias that may occur when including persons with the disease could influence the validity of the exposure. Even after removing persons with known disease, however, Gregg et al. (14) observed that adults who self-reported a "large" increase in heart rate or breathing during walking (when they were compared with the referent, i.e., nonwalkers) had an increased risk of CVD mortality (relative risk = 1.12, 95% CI = 0.45-2.78), albeit this was not statistically significant. The authors surmised that adults reporting what might seem to be perceptions of increased effort may have actually had exertional coronary ischemia or other undetected cardiovascular or pulmonary disease that was unmasked to some degree by being active. When the authors eliminated from the analyses those persons dying during the first 2 yr to remove the influence of previously undetected disease, the amount of risk reduction for persons perceiving large increases in heart rate or breathing rate was similar to those with small or moderate increases in perceived effort.
Self-reported walking speed may also influence estimates of risk in studies of walking and diabetes. If self-selected walking speed differs by race or ethnicity, as suggested by a recent study (22), self-reports of how fast participants in a study are walking (especially if these persons are asked to compare their speed to that of others) might be biased by racial or ethnic category. Manghera (22) made timed measures of 60-ft (18.3-m) distances from which self-selected walking speeds could be estimated for 450 California men, 150 each in three "presumed" racial/ethnic groups-African American, Hispanic, and white (as observed and designated by the author). She noted that white men walked faster than African American men or Hispanic men in a business area and in a recreational area (e.g., shopping mall). In a timed crosswalk, Hispanic men had a somewhat intermediate observed speed than the two other groups, with whites still walking at the fastest speed. No tests of statistical significance were reported. At present, we do not know whether racial/ethnic status affects the association of walking with incident diabetes or with mortality outcomes in persons who already have the disease.
There are a variety of areas worth pursuing as part of future research initiatives in walking and diabetes. For one, methodological research should be used to identify and quantify the error from walking classifications that differ systematically by racial or ethnic group. We have seen, for example, that the assessment of walking, in particular self-reported speed, might vary with racial/ethnic status, at least for certain situations (22). In addition, if misclassification occurs for amounts and intensity of walking during these selected instances, we might conclude that it could also occur for walking as part of work, household chores, transportation, or caregiving activities. These differences, in turn, may lead to false conclusions about the link between walking and diabetes (and its outcomes). A second line of inquiry would be to better characterize the dose of walking (e.g., frequency, intensity, and duration, and their interactions) to clarify links to physiological mechanisms by which walking may reduce the risk of type 2 diabetes. Such research should make special note of 1) low levels of walking intensity, 2) the accumulation of bouts of walking, and 3) the ways in which walking could combine with other activities (e.g., cycling) to achieve diabetes-related health effects. A third area is that future epidemiologic studies must explore the dose-response relationships between walking and either incident diabetes or disease outcomes among those who already have diabetes; the role of confounding variables, such as central adiposity, which influence the association of walking and diabetes and the physiological mechanisms that might operate at the population level. Finally, because interventions cost money and time, research that incorporates cost-effectiveness, as was the case in at least one intervention involving physical activity where walking was quite common (10), is needed for interventions depending exclusively on walking to help allocate limited economic resources.
Diabetes profoundly raises important concerns for leaders in public health and for clinicians, as in much of the developed world, we now have an epidemic of this frequently devastating disease. The limited epidemiologic data we have, which relates walking to the prevention of incident diabetes or to mortality outcomes in persons who already have the disease, seem to be congruent with earlier reviews and reports on physical activity in general, such as the first Surgeon General's Report on Physical Activity and Health (31). The strength of the reductions in risk that we have seen fits in well with the results seen for physical activity in general. Similarly, the limited studies available, suggesting a dose-response relationship between walking and the outcomes discussed, suggest that public health recommendations for physical activity might apply to walking as a subpart of physical activity. Given the scant data on walking and diabetes, numerous important research topics remain to be pursued.
The authors thank the editorial contributions of Peter L. Taylor, M.B.A. (Senior Editor, Palladian Partners, Silver Spring, MD).