Dose-Response Relationship of Physical Activity to Premature and Total All-Cause and Cardiovascular Disease Mortality in Walkers

Purpose To assess the dose-response relationships between cause-specific mortality and exercise energy expenditure in a prospective epidemiological cohort of walkers. Methods The sample consisted of the 8,436 male and 33,586 female participants of the National Walkers' Health Study. Walking energy expenditure was calculated in metabolic equivalents (METs, 1 MET = 3.5 ml O2/kg/min), which were used to divide the cohort into four exercise categories: category 1 (≤1.07 MET-hours/d), category 2 (1.07 to 1.8 MET-hours/d), category 3 (1.8 to 3.6 MET-hours/d), and category 4 (≥3.6 MET-hours/d). Competing risk regression analyses were use to calculate the risk of mortality for categories 2, 3 and 4 relative to category 1. Results 22.9% of the subjects were in category 1, 16.1% in category 2, 33.3% in category 3, and 27.7% in category 4. There were 2,448 deaths during the 9.6 average years of follow-up. Total mortality was 11.2% lower in category 2 (P = 0.04), 32.4% lower in category 3 (P<10−12) and 32.9% lower in category 4 (P = 10−11) than in category 1. For underlying causes of death, the respective risk reductions for categories 2, 3 and 4 were 23.6% (P = 0.008), 35.2% (P<10−5), and 34.9% (P = 0.0001) for cardiovascular disease mortality; 27.8% (P = 0.18), 20.6% (P = 0.07), and 31.4% (P = 0.009) for ischemic heart disease mortality; and 39.4% (P = 0.18), 63.8% (P = 0.005), and 90.6% (P = 0.002) for diabetes mortality when compared to category 1. For all related mortality (i.e., underlying and contributing causes of death combined), the respective risk reductions for categories 2, 3 and 4 were 18.7% (P = 0.22), 42.5% (P = 0.001), and 57.5% (P = 0.0001) for heart failure; 9.4% (P = 0.56), 44.3% (P = 0.0004), and 33.5% (P = 0.02) for hypertensive diseases; 11.5% (P = 0.38), 41.0% (P<10−4), and 35.5% (P = 0.001) for dysrhythmias: and 23.2% (P = 0.13), 45.8% (P = 0.0002), and 41.1% (P = 0.005) for cerebrovascular diseases when compared to category 1. Conclusions There are substantial health benefits to exceeding the current exercise guidelines.


Introduction
The health benefits produced by physical activity are welldocumented [1][2][3][4]. Less is known about the dose-response relationships between specific exercises and mortality. In part, this is because most epidemiological cohorts lack the statistical power to draw inferences for specific activities, and most do not include very many subjects who exercise at high doses. In addition, most studies use exercise duration (i.e., time spent exercising) to calculate the energy expenditure [1,5]. Our studies of walkers and runners suggest that exercise energy expenditure that is calculated from duration exaggerates the amount exercise performed [6][7][8].
The current report examines the dose-response relationships between exercise energy expenditure and mortality in the National Walkers' Health Study [28][29][30]. Whereas other epidemiological cohorts were designed for general purpose, the National Walkers' Health Study was designed specifically to test the health benefits of walking. Results are presented for all reported exercise, and for walking in particular. In addition, the analyses are not restricted to the underlying cause of death, i.e., ''the disease or injury that initiated the chain of morbid events that led directly and inevitably to death'' [31]. Contributing causes of death are also included for two reasons: first, a single underlying cause of death may be difficult to identify and may depend upon the physician's training, specialty, and familiarity with the patient [32][33][34]; and second, exercise may affect survival via its effect on one or more of the contributing causes of death.

Study design
The National Walkers' Health Study is a prospective epidemiological cohort that has been described in detail in several publications [28][29][30]. The cohort was initially recruited between 1998 and 2001 among subscribers of a walking magazine and among participants of walking events. At baseline, participants completed a two-page questionnaire on demographics (age, race, education), walking history (age when began walking at least 12 miles per week, current average weekly mileage), weight history (greatest and current weight; weight when started walking for exercise; least weight as a walker; body circumferences of the chest, waist, and hips; bra cup size), diet (vegetarianism and the current weekly intakes of alcohol, red meat, fish, fruit, vitamin C, vitamin E, and aspirin), current and past cigarette use, history of heart attacks and cancer, and medications for blood pressure, thyroid, cholesterol, or diabetes. The study protocol was approved by the University of California Berkeley Committee for the Protection of Human Subjects, and all subjects provided a signed statement of informed consent.

Questionnaire items
Intakes of meat, fish and fruit were based on the questions ''During an average week, how many servings of beef, lamb, or pork do you eat'', and ''…pieces of fruit do you eat''. Alcohol intake was estimated from the corresponding questions for 4-oz. (112 ml) glasses of wine, 12-oz. (336 ml) bottles of beer, and mixed drinks and liqueurs. Alcohol was computed as 10.8 g per 4-oz glass of wine, 13.2 g per 12 oz. bottle of beer and 15.1 g per mixed drink [35]. Correlations between these responses and values obtained from 4-day diet records in 110 men were r = 0.46 and r = 0.38 for consumptions of meat and fruit, respectively. These values agree favorably with published correlations between food records and more extensive food frequency questionnaires for red meat (r = 0.50), and somewhat less favorably for fruit intake (r = 0.50) [36]. Self-reported height and weight from the questionnaire have been found previously to correlate strongly with their clinic measurements (r = 0.96 for both) [24]. Walking energy expenditure was computed by converting the reported distance walked into duration (i.e., reported distance divided by the reported mph speed), which was used in association with reported intensity to calculate MET-hours/d [5]. Energy expenditures from other (non-running and non-walking) activities were calculated from the reported time spent participating in the activities and their published MET values [5].

Mortality surveillance
Mortality surveillance was completed through December 31, 2008 using the National Death Index (International Classification of Disease codes version 10 [37]). Both underlying cause of death and the entity axis diagnoses of contributing causes were analyzed.

Statistical analyses
Hazard ratios (HR) from Cox proportional hazard analyses were used to compare all-cause mortality to exercise energy expenditure when adjusted for baseline age (age and age 2 ), race, sex, years of education, baseline smoking status (yes/no), ever smoked (yes/no), prior history of a heart attack, prior history of a cancer, aspirin use, and intakes of meat, fruit, and alcohol. Semihazard ratios (SHR) from competing risks regression were used to test whether cause-specific deaths were significantly related to exercise energy expenditure when adjusted for these covariates [38]. In these analyses, the competing risk was death due to all other causes. All analyses were performed using the statistical software package Stata (version 11, Stata Corp, College Station, TX). Ninety-five percent confidence intervals (95%CI) are reported for both HR and SHR.
A quadratic expression of MET-hours/d (i.e., MET 2 ) was included to test whether the dose-response relationship between mortality and MET-hours/d of exercise was significantly nonlinear. The two-degree of freedom Chi square test was used to test whether men and women showed the same dose-response relationship between mortality and exercise. Specifically, the Chi square test was applied to the difference in the log likelihoods between a model that included separate linear and quadratic regression coefficients for men and women with one that had the same linear and quadratic regression coefficients for both sexes. Results are presented as relative risk (HR, SHR) and their percent reductions in risk (calculated as 100*(HR-1) or 100*(SHR-1)) per MET-hours/d, and for four categories of exercise energy expenditure: category 1 is falling short of the 450 MET minutes per week recommended for health (#1.07 MET-hours/d), category 2 is meeting the 450 to 750 MET min/wk guideline exercise recommendations (1.07 to 1.8 MET-hours/d [4]), category 3 is exceeding the guidelines by one-to two-fold (1.8 to 3.6 MET-hours/d), and category 4 is exceeding the guidelines by greater than two-fold ($3.6 MET-hours/d). Access to the data required human use approval. Table 1 presents the cohort's baseline characteristics by exercise level. The substantially greater number of females reflects their greater participations in walking events, and their greater proportion among subscribers to walking publications. BMI at baseline (BMI baseline ) was available for 96.8% of the sample. Greater exercise was associated with younger age, less smoking, fewer prior heart attacks or cancers (females only), less meat and greater fruit and alcohol consumption, and lower BMI baseline . BMI when first began walking 12 or more miles per week (BMI starting walking ) was available for 72.4% of the sample. Most of those with missing BMI starting walking were subjects who simply never achieved that level of activity, i.e. BMI starting walking was unreported for 54.7% of the sample who exercised ,1.07 MET-hours/d at Figure 1. Relative risk (hazard ratio) for all-cause mortality vs. exercise energy expenditure adjusted for age (age plus age 2 ), race, education, smoking status (current and prior history), prior history of heart attack and cancer, daily intakes of alcohol, meat and fruit, aspirin use. Brackets designate 95% confidence intervals. Significance levels for the risk reduction relative not achieving the minimum recommended exercise level (i.e., ,1.07 MET-hours/d) are coded: * P,0.05, { P,0.01,`P,0.001, and 1 P,0.0001. doi:10.1371/journal.pone.0078777.g001 baseline, 45.0% of the sample who exercised 1.07 to 1.8 METhours/d, 13.2% of the sample who exercised 1.8 to 3.6 METhours/d, and 12.6% of the sample who exercised $3.6 METhours/d.

All-cause mortality
One-thousand two hundred thirteen of the 8,436 men (14.4%) and 1235 of the 33,586 women surveyed at baseline (3.7%) died during the average 9.6-year follow-up. Figure 1 displays the sexspecific reductions in risk associated with greater exercise. Both sexes exhibited significantly greater reductions in risk for exceeding the exercise recommendations compared to simply meeting the recommendations (25% greater risk reduction in males, P = 0.0004; 22.8% greater risk reduction in females, P = 0.001 for $1.8 MET-hours/d vs. 1.07 to 1.8 MET-hours/ d). Adjustment for medication use and BMI baseline somewhat attenuated the risk reductions. There was no significant difference in the dose-response relationship between males and females (P = 0.93, analyses not displayed), so their data were combined and analyzed adjusted for sex.
When the sexes were combined, Table 2 shows that exercise energy expenditure was inversely related to all cause mortality (significant linear effect), and that the decrease in mortality leveled off (i.e., plateaued) at higher energy expenditures (significant quadratic effect). To depict the degree of nonlinearity the analyses were therefore repeated in Table 3 using the four categories of energy expenditure (see methods). The categorical analysis shows that most of the risk reduction was achieved by 1.8 to 3.6 METhours/d. The risk reduction was moderately attenuated by adjustment for baseline medication use and BMI baseline (Table 4) or BMI starting walking (Table 5). Exceeding the exercise recommendations was associated with a 24.1% greater risk reduction compared to merely meeting the recommendations (HR: 0.759, 95%CI: 0.680 to 0.847, P = 10 26 for $1.8 MET-hours/d vs. 1.07 to 1.8 MET-hours/d).

Cause-specific mortalities
Sex also did not significantly affect the dose-response relationships between MET-hours/d of exercise and the risk for CVD (P = 0.88 for sex difference), ischemic heart disease (IHD, *Hazard ratios (95% confidence intervals) from Cox proportional hazard analyses. { Semi-hazard ratios (95% confidence intervals) from competing risk regression analyses. All analyses adjusted for age (age, age 2 ), race, sex, education, prior heart attack, aspirin use, and intakes of red meat, fruit, and alcohol. Total mortality also adjusted for cancer history. doi:10.1371/journal.pone.0078777.t002 P = 0.85), dysrhythmias (P = 0.99), heart failure (P = 0.81), hypertensive disease (P = 0.86), cerebrovascular disease (P = 0.89), or diabetes mortality (P = 0.66). Therefore, the male and female data for these deaths were also combined and adjusted for sex when analyzed. Table 2 shows that except for heart failure, the doseresponse relationships were all significantly nonlinear. Their analyses by categories of exercise energy expenditure are presented in Tables 3, 4, 5 for underlying causes of death, and in tables 6, 7, 8 for all disease-related mortality (contributing plus underlying cause). Twenty-five percent of IHD-related deaths had a non-CVD underlying cause of death, as did 22% cerebrovascular disease, 31% heart failures, 38% hypertensive disease, and 43% dysrhythmia-related deaths.

Cardiovascular disease
The risk of CVD as an underlying cause of death decreased 23.6% for meeting the recommended exercise levels, 35.2% for exceeding the recommendations by 1 to 2 fold, and 34.9% for exceeding the recommendations by $2-fold (Table 3). Similar results were obtained for all CVD-related deaths (Table 6), which were only moderately attenuated when adjusted for medications and BMI starting walking (Tables 5 and 8). The reduction CVDrelated mortality was significantly greater for exceeding the recommendations compared to merely meeting them (SHR: 0.738, 95%CI: 0.634 to 0.859, P = 0.0001). When adjusted for BMI baseline (Tables 4 and 7), the hazard ratios for both 1.8 to 3.6 MET-hours/d and $3.6 MET-hours/d remained significant and showed $25% risk reductions relative to inadequate exercise.

Ischemic heart disease
IHD represented about one-half of all CVD deaths. The reduction in risk for all IHD-related mortality was similar to that of all CVD-related mortality ( Table 6). Adjustment for baseline medication use and BMI starting walking did not weaken the reductions in risk (Table 8), whereas adjustment for BMI baseline did (Table 7). Nevertheless, exceeding the recommendations by $1.8 MET-hours/d was associated with a 21.1% risk reduction relative to inadequate exercise for all IHD-related deaths, even when adjusted for BMI baseline (SHR: 0.789, 95%CI: 0.647 to 0.963, P = 0.02).

Cerebrovascular disease
The risk for death due to cerebrovascular disease as an underlying cause was significantly less in those who exercised inadequate exercise, with or without adjustment for BMI (BMI baseline or BMI starting walking ). Two-thirds of the diagnoses were ''stroke not specified as hemorrhage or infarction'' or ''unspecified cerebrovascular disease'' that prevented a more refined endpoint analyses.

Dysrhythmias
Despite only 48 deaths ascribed to dysrhythmias as the underlying cause, their risk was significantly less in those who exercised $1.8 MET-hours/d vs. ,1.07 MET-hours/d (Table 3), which generally persisted when adjusted for baseline medication

Heart failure
There were only 53 deaths due to heart failure as an underlying cause, but five times as many heart failure-related deaths, and their risk was significantly reduced by exercising 1.8 to 3.6 METhours/d and $3.6 MET-hours/d vis-à-vis ,1.07 MET-hours/d, with or without BMI adjustment (Tables 6-8). Exceeding the exercise recommendations reduced the risk of heart failure by 36.6% (SHR: 0.634, 95%CI: 0.444 to 0.906, P = 0.01) relative to simply meeting the recommendations.

Other circulatory diseases
Deaths due to other circulatory diseases (i.e., exclusive of IHD, cerebrovascular disease, heart failure, dysrhythmia, or hypertensive disease) were unrelated to physical activity.

Walking
Walking represented more than 96% of the exercise energy expenditure ,1.8 MET-hours/d, 90% to 92% of the expenditure Table 6. Survival analyses of all related mortality, i.e., underlying or contributing causes of death, vs. categories of physical activity. between 1.8 and 3.6 MET-hours/d, but only 67% to 68% of energy expenditure above 3.6 MET-hours/d. The proportions were very similar for both males and females (Table 1). Not surprisingly, then, the results for walking are very similar to those for total exercise, particularly at the lower range of energy expenditure. Figure 2 show that the decrease in risk was greatest between 1.8 and 3.6 MET-hours/d, and although the risk reduction diminished at higher levels of walking energy expenditure, there was no significant difference in the risk reductions above 1.8 MET-hours/d. Only 5% of the sample expended $5.4 MET-hours/d. Table 9 shows that generally, the reductions in risk associated with walking energy expenditure were very similar, albeit slightly weaker in magnitude, than those observed for all exercise.

Discussion
The National Walkers' Health Study was specifically recruited to characterize the dose-response relationships between walking and health. Its highest walking category, i.e. $5.4 MET-hours/d, corresponds to $1.4 hours or $5.0 miles (8 km) per day of brisk walking. This exercise category is substantially greater than those included in other studies. It is greater than all but one of the studies included in Hamer and Chida's meta-analyses of CVD [16], all of those included in Jeon et al.'s meta-analyses of diabetes Table 7. Survival analyses of all related mortality, i.e., underlying or contributing causes of death, vs. categories of physical activity adjusted for medication use and BMI baseline .  [27], and all of those included in Moose et al meta-analyses of over 650,000 subjects [39]. It is over twice as great as the top walking categories reported by Manson et al. for the Nurses' Health Study [9] and for the Women's Health Initiative Observational Study [10]. The inclusion of highly active walkers in our study allows us to conclude that most of the reductions in total, CVD, and diabetes mortality appears to be achieved by exercising between 1.8 and 3.6 MET-hour/d.

All cause mortality
Consistent with prior studies [17,18,[21][22][23]39] our analyses showed that greater exercise was associated with lower all-cause mortality. However, they also showed significantly greater risk reductions for patients who exceeded the exercise recommendations relative to simply achieving them, with additional benefits accruing through at least 2.7 MET-hours/d (the midpoint and average of the 1.8 to 3.6 MET-hour/d category). Moreover, reductions in all-cause mortality were demonstrated for both total and premature deaths, in both men and women, and excluding deaths during the first three years of follow-up. The 2008 physical activity guidelines suggested a target of 2 to 2.5 h/wk of moderate intensity activity for reducing total mortality [1]. Our results suggest that this would provide only about one-half of the potential risk reduction, and that 2.7 MET-hours/d (5 h or 28 km/wk) is a better target. The 32.9% risk reduction we observed is not inconsistent with the 40% risk reduction for 7 hr/wk presented in the 2008 recommendations [1], and the approximately 35% risk reduction suggested by the analyses of Moore et al. [39]. Table 3 shows that significant risk reductions for all CVD and IHD were achieved by meeting the recommendations, with significant further reductions in all CVD mortality by exceeding them. Our data showed no additional risk reduction beyond twice the recommended level. For both diagnoses, the nonlinearity of the dose-response relationship was statistically significant. In addition, our data suggested that the reductions in the risks for hypertensive disease, heart failure, dysrhythmias, and other CVD as underlying causes of death were as great, or greater, than that of IHD. Others also report that the reductions in CVD risk with walking are not limited to IHD [23]. As with IHD, the doseresponse relationships for these other conditions were all significantly nonlinear, with the majority of their risk reductions occurring by 1.8 to 3.6 MET-hours/d ( Table 2). The 30% to 40% reductions in cerebrovascular disease mortality we observed were somewhat greater than the 25% to 30% reductions reported for physical activity by others [1], perhaps reflecting our use of the distance-based metric. Remarkably, there were five mutually exclusive CVD endpoints (i.e., underlying causes) that showed consistently lower mortality with greater MET-hours/d walked. Unlike other studies, the current analyses were not restricted to the underlying cause of death. There is no evidence from Tables 2 to 8 that restricting the analyses to the underlying cause of death was more informative than including all related deaths. In other words, there is no evidence that the reductions in risk were diluted by including contributing causes of death for some other underlying cause. In fact, exercise was associated with lower CVD risk even when CVD was not the underlying cause. This means that the mechanism by which walking reduces mortality may not simply involve the underlying cause, but also conditions linking the initial underlying cause to death. We also caution that the assignment of an underlying cause of death by physicians may be problematic [32][33][34] and that our analyses of all listed causes may circumvent the difficulty of choosing a single underlying cause among several viable alternatives.

Underestimating the health benefits of walking
Our cohort did not include a totally sedentary group. Therefore, the risk reductions reported here probably underestimate the potential for risk reduction relative to completely sedentary populations. Specifically, others report substantial risk differences that fall within our lowest exercise category (0 to 1.07 MET-hours/d). For example, the Women's Health Study reported that there was a 63% difference in coronary heart disease risk and 44% difference in diabetes risk between non-walkers and those walking between 1 and 1.5 h/wk [11,26]. In addition, the Nurses' Health Study [9] and the Women's Health Initiative Observational [10] reported that most of reduction in cardiovascular disease and ischemic heart disease risk had occurred by the 70 th percentile of their walking distribution. Similarly, the Health Professionals' Follow-up Study reported that there was a 40% reduction in total mortality by the 60 th percentile of their walking distribution [17].

Comparison with runners
The current findings for walkers differ somewhat from our prior analyses of runners, in which we showed prospectively that the risk for nonfatal coronary heart disease decreased 30% (P = 0.006) by running $6 MET-hours/d when compared to running 3-6 METhours/d [40]. In addition, men and women who ran $4.1 METhours/d were at significantly lower risk for stroke than those running 2 to 4 MET-hours/d, and those that ran $8.2 METhours/d were at 60% lower risk than those who ran ,2 METhours/d [41]. In contrast, Table 3 suggests that the majority of the risk reduction for IHD and cerebrovascular disease occurred by 1.8 to 3.6 MET-hours/d of walking. Several factors may explain the difference: 1) vigorous exercise (e.g. running) may produce different health benefits than moderate intensity exercise (e.g., walking); 2) runners may differ genetically or behaviorally from walkers; and 3) exercise may affect nonfatal and fatal coronary heart disease differently.
Adjustment for BMI and diabetes, blood pressure, and cholesterol medication use Tables 4 and 7 showed that the risk reduction for most causes of death with exercise (CVD-, IHD-, cerebrovascular disease-, heart failure-, dysrhythmia-, and diabetes-related deaths) remained significant when adjusted for BMI baseline and medication use. The reductions in risk were somewhat diminished by the adjustment, but nevertheless remained clinically and statistically very significant. These analyses likely represent over-adjustment because running and walking are known to reduce hypertension, high cholesterol and type 2 diabetes risk [42][43][44][45], and to attenuate age-related weight gain [46,47]. In this regard, the adjustments for BMI starting walking in Tables 5 and 8 are probably more appropriate than the adjustment for BMI baseline , given that weight control may mediate some of the health benefits of walking.

Limitations
There are important limitations to these analyses. Vital status in known only from the National Death Index and therefore some subjects who have died are likely to be misclassified as alive. The set of covariates used in the analyses are somewhat restricted and all are measured with error, which will cause their effect to be underestimated by the adjustment. The ability to exercise harder may be an innate quality of health and not directly related to habitual physical activity. The greater risk for heart failure mortality with less exercise could be a consequence of cardiac insufficiency at an early, preclinical, stage of the disease.

Conclusions
These analyses suggest there are important health benefits to exceeding the current exercise recommendations for health ($750 MET minutes per week or $1.8 MET-hours/d [9]) rather than just satisfying them (450 to 750 MET minutes per week [9]). These health benefits include reductions in disease mortality that are not traditionally associated with walking, including both heart failure and dysrhythmias.