A large number of studies have investigated the association between physical activity and all-cause mortality, and the evidence for a beneficial effect is well-established.1 In 2011, a systematic review and meta-analysis of 22 cohort studies concluded that physical activity reduces risk of all-cause mortality.2 The largest benefit was found with moving from no activity to low levels of activity.2 This is particularly interesting because some studies have reported a threshold effect,3–23 suggesting that the beneficial effect of physical activity can be separated into a threshold effect of participation in activity and then a dose–response effect of frequency, duration, or volume. An effect of participation per se would occur if the key aspect of physical activity is to avoid inactivity, that is, a biological mechanism is activated, or inactivated, by the onset of activity.
However, it is also possible that participation/nonparticipation in a specific leisure time activity is associated with obesity, other aspects of lifestyle, general health, or socioeconomic factors. For example, participation in sports requires a certain health status, particularly in the elderly, and may be associated with general health and well-being and awareness of health recommendations; single men are more likely to do housework; and in Denmark, gardening requires a house with a garden and consequently, a certain socioeconomic position. Such factors introduce a risk of confounding. This type of confounding is well-known for alcohol intake,24 where abstainers hold a different and higher risk than would be expected from a downwards extrapolation of the dose–response association to zero. Such considerations may also be relevant to other lifestyle factors like physical activity.
One simple approach to this problem is to separate abstainers (nonactive people) from the analysis of dose response by including an indicator variable for activity. This separation may be useful for three reasons: 1) because of the possibility of a biologically based threshold; 2) because of the risk of confounding if participation in specific activities is directly linked to socioeconomic factors or other lifestyle and health-related factors; and 3) because when summed measures of activity are used (total hours or metabolic equivalents hours of physical activity), abstaining from several of the contributing activities leads to lower values of the summed measure whereas being able to participate in many activities leads to higher levels of the summed measure of activity, thereby creating a “false” dose response, if the summed measure is not adjusted properly. The use of indicator variables in the regression model (rather than excluding those abstaining from one or more activities) maintains statistical power.
In this prospective cohort study, we evaluated the associations between all-cause and cause-specific mortality and physical activity as measured by the leisure time activities, sports, riding a bicycle, gardening, walking, “do-it-yourself” activities, and housework. The dose–response effect of physical activity was separated from the possible difference in mortality according to participation/nonparticipation in each type of leisure time activity by including indicator variables for participation.
The Diet, Cancer and Health Study Population
The cohort has been described previously.25 Briefly, all men and women aged 50–64 years and resident in the greater Copenhagen and Aarhus areas were selected from the Danish Central Population Registry and invited to participate in the Diet, Cancer and Health Study between December 1993 and April 1997. All participants were born in Denmark and had no previous diagnosis of cancer. A total of 29,861 women and 27,146 men were enrolled into the study (33% and 37% of these invited). All participants completed a questionnaire on health, social, and lifestyle factors, including physical activity, smoking, education, and selected previous diseases. Anthropometric measures were obtained at the study clinic. Informed consent was obtained from all study participants during the visit to the study clinic, and the study protocol was approved by the Copenhagen Ethical Committee on Human Studies.
Assessment of Leisure Time Physical Activity
Physical activity was assessed by a self-administered, interviewer-checked questionnaire where occupational physical activity was reported according to five predefined categories (sedentary work, standing work, manual work, heavy manual work, and no occupation) and leisure time physical activity as hours per week spent on sports, cycling, walking, gardening, housework (cleaning and shopping), and “do-it-yourself” activities (eg, house repair). For each type of activity, the participants were asked to report the number of hours spent per week during summer and winter in the last year. The two values for each activity were averaged; therefore, the lowest possible positive value of this average was 0.5 hour of physical activity per week. Three hundred thirty-three participants were excluded because of missing values in one or more of the physical activity variables, and 15 participants were excluded because of implausibly long times spent on leisure physical activity (>105 hours per week). The physical activity questions have been validated and were found to rank individuals satisfactorily with regard to physical activity level.26–28
Assessment of Other Variables
Smoking habits were assessed as past and current smoking, time since smoking cessation, and amount smoked daily. Current tobacco consumption (cigarettes, cigars, cheroots, or pipe) was calculated in grams per day using conversion factors of 1, 4.5, 3, and 3 grams, respectively. Education was described as years of education after secondary school. Blood pressure, total serum cholesterol, weight, and height were measured at the study clinic, and body mass index (BMI) was calculated as weight (kg) per squared height (m2). Previous acute myocardial infarction, angina pectoris, stroke, hypertension, and hypercholesterolemia were self-reported.
Six hundred sixty-seven participants were excluded because of missing values, leaving a final study cohort of 29,129 women and 26,576 men.
Ascertainment of Mortality
All-cause mortality was the primary endpoint of the study. Complete follow-up of emigration and vital status was possible through information from the Civil Registration System using unique personal identification numbers. Follow-up time was calculated from date of inclusion until the date of emigration, death, or 31 March 2010, whichever came first. Data on cause-specific mortality were retrieved from the Danish National Cause of Death Registry.
The associations between physical activity and sex-specific all-cause and cause-specific mortality were estimated by Cox proportional hazard models with follow-up time as the underlying time axis and stratification by the integer of age. The six leisure time physical activities (sports, cycling, walking, gardening, housework, and do-it-yourself activity) were included simultaneously in the three models with different sets of adjusting variables. The basic model (model A in tables) included all variables related to physical activity (leisure time activities and occupational activity) in addition to age and follow-up time, except for the analyses presenting the effect of participation in leisure time activities, for which the time spent on each activity was not included. The second model (model B in tables) extended the first model by including the potential confounders smoking, education, and self-reported cardiovascular diseases measured at baseline. In the third model (model C in tables), the potentially mediating variables, BMI, blood pressure, and serum cholesterol were added. Cause-specific mortality was investigated as competing risks.
Leisure time activity was investigated in three ways: 1) a categorical analysis comparing active to nonactive for each activity; 2) a linear effect of time spent on each activity while allowing a different mortality for nonparticipants; and 3) modeling the dose–response effect using a linear spline, as suggested by Greenland,29,30 while allowing a different mortality for nonactive people for each activity. Linear splines use the within-category variation in physical activity and mortality to estimate the slopes of fitted lines within each interval of physical activity, allowing the risk to vary both within and between categories in a continuous way. The piecewise linear function meet in join points (knots) placed at boundaries for selected exposure categories; at these points, the curves are allowed to change direction.29 In this study, we placed boundaries at exposure quartile cutpoints, except for housework for men and do-it-yourself activity for women, for which tertiles were used because of limited variability. In the dose–response analyses, nonparticipants were included by assigning indicator variables of participation (value 1 or 0 for the specific activity). The rationale behind this strategy is that mortality among people who do not participate in a specific type of leisure time physical activity (the abstainers) may deviate from the extrapolation-to-zero activity based on a dose–response relationship between mortality and physical activity among the active, that is, there may be a threshold of effect caused by a true biological mechanism, confounding, or both. Because exclusion of the abstainers from the analyses would decrease the statistical power, the potentially different mortality was captured by the indicator variable.
Homogeneity among the effects of the various types of leisure time activities was tested to investigate whether the indicators of participation could be summed to a total number of activities and whether hours of the types of leisure time activity could be summed to total hours of activity. We evaluated this by testing whether the effects of the separate activities were unequal in a mutually adjusted analysis. We found that neither number of activities nor total hours of activities were appropriate as risk indicators.
Participants were categorized as smokers, never-smokers, and ex-smokers supplemented with a linear effect of daily tobacco consumption in grams and, for ex-smokers, a linear effect of years since smoking cessation. Education after secondary school was categorized as none, short (1–3 years), medium (3–4 years), and high (>4 years). Self-reported cardiovascular health data concerning hypertension, hypercholesterolemia, angina pectoris, acute myocardial infarction, and stroke were categorized according to severity: none, hypertension and/or hypercholesterolemia, angina pectoris, acute myocardial infarction, and/or stroke. Participants were classified according to the highest degree of severity if several conditions were present.
For BMI, systolic blood pressure, and serum cholesterol, the use of linear splines were necessary to obtain adequate fits. BMI was modeled as a linear spline with boundaries at 18.5, 25, and 30 kg/m2, corresponding to the World Health Organization categories of obesity.31 Similarly, baseline systolic blood pressure was log-transformed and modeled as a linear spline with boundaries at 120, 140, and 180 mmHg corresponding to the categories of systolic hypertension used in clinical practice.31 Baseline serum cholesterol was modeled as a linear spline with a single boundary at 5 mmol/L, corresponding to 200 mg/dL, the cutpoint for elevated cholesterol in clinical practice.31
We evaluated the proportional hazards assumption by performing separate analyses of deaths within three intervals of follow-up time (boundaries at 2 and 4 years after baseline). These analyses gave similar estimates for all three follow-up periods, indicating no important deviations from proportionality (results not shown). We also tried age as the underlying time scale with adjustment for follow-up time; these analyses were not different from the presented analyses with follow-up time as time scale and stratification according to age at baseline in whole years. Adjustment for some simple dietary variables (whole grain, red and processed meat, fruit, and vegetables [grams/day]) did not affect the results for model C, and we decided not to include dietary variables in our models.
Tests were based on the likelihood ratio test statistic calculated from Cox partial likelihood. Confidence intervals (CIs) were based on Wald’s test of the corresponding regression parameters, that is, on the log-scale for the rate ratios.
Study participants were followed from recruitment through 31 March 2010 for a median of 15.6 years (range among survivors: 13.1–16.3 years). A total of 2696 women and 4044 men died during that period. The basic characteristics of the final study population are shown in Table 1, and levels of leisure time physical activities among cohort members are shown in Table 2.
In the first survival analysis, we investigated the all-cause mortality associated with participation in six types of leisure time physical activity. The estimated associations for participation (yes/no) in leisure time physical activity and mortality are shown in Table 3. Engaging in sports, cycling, and gardening was associated with lower mortality among both women and men. Among men, engaging in “do-it-yourself” activities was also associated with lower mortality, and among women, walking was associated with lower mortality. Adjustment for baseline values of the potential confounders generally weakened the associations slightly (model B). The potentially mediating variables, BMI, systolic blood pressure, and serum cholesterol are predictors of mortality, but including them did not change the estimates substantially (model C).
Table 4 presents the mortality rate ratios for the dose–response associations between the average number of hours per week spent on each specific type of leisure time physical activity and all-cause mortality, adjusted for the potentially different mortality rates among persons abstaining from a specific activity. We found slightly higher mortality with more hours spent on the leisure time activities, walking, and gardening among both women and men. More time spent on housework was associated with a slightly higher mortality among men, but lower mortality among women. The associations with mortality were small, in the order of 1–2% increased risk per additional hour of activity per week beyond an average of half an hour per week (the minimum value of time spent on a single type of activity in this study). Again, adjustment for potential confounders did not change the associations.
Because time spent on physical activity in general was not associated with any beneficial effects, we used linear splines to investigate whether linear modeling of the effect of physical activity was too simple a description of the biology, that is, whether there were signs of nonmonotonic associations. Figure 1 shows the fitted linear splines for the log-scaled rate ratios for each sex for each of the six physical activities, using the nonactive (abstainers) of the same sex as the reference level (indicated by a black dot in the plots). In general, the start of the curves (corresponding to an average activity of half an hour per week) showed low rate ratios, indicating a large difference in mortality between abstainers and those with very low activity. Furthermore, the curves were in general at or below the horizontal reference line showing that active people in general had lower mortality rates than abstainers. Most of the physical activities showed some fluctuations in risk by hours per week, but no major departure from linearity. None of the curves showed a U-form for both sexes. Consequently, the association between physical activity and mortality was considered adequately described by the simple model presented in Table 3, where the association was confined to a difference between active and nonactive people in each specific activity.
We also investigated cause-specific mortality, that is, mortality from cancer, cardiovascular disease, respiratory disease, diabetes, and other causes, in relation to participation in and time spent on each of the six types of leisure time activity (Table 5). These results were generally in line with the results on overall mortality. In particular, lower mortality with participation in sports and gardening was consistent across the various causes of mortality.
Among women and men in the Diet, Cancer and Health cohort, we found that participation in some leisure time activities (sports, cycling, gardening, and do-it-yourself activities for men) was associated with lower mortality. However, there were no signs of a lower mortality associated with more time spent on each activity, except for a less than 1% decrease in mortality per additional hour of housework for women. This result was not simply because of limited power because the CIs were quite narrow. Fitted linear splines did not reveal any inflexion or nonmonotonic tendencies. The inclusion of lifestyle and health-related confounding variables weakened the associations between leisure time physical activities and mortality, but did not change the pattern substantially.
Overall, our study does not support the hypothesis of lower mortality with more time spent on physical activity, but it indicates that participation in activity may be particularly important in the prevention of early death.
A biological effect of physical activity on all-cause mortality would presumably show up as a dose–response relationship of some kind, perhaps with an inflexion or even a U-shape. We thoroughly investigated the shape of the relation between the number of hours per week of leisure time activity and all-cause mortality using linear splines, which provide more statistical power than categorical analysis.29,30 We found no evidence of any biologically meaningful, beneficial effect with more physical activity. The beneficial effect—if any—of the types of physical activity considered in this study was confined to a difference between the active and the nonactive. However, if we had conducted the inherently confounded analyses of time spent on leisure time activity without including the indicators of participation, that is, without separating the effect of participation from the dose–response effect of more activity, we would have found (like most other studies on this topic) beneficial effects of more time spent on leisure time activities (example sports: women 0.96, 0.93–0.98; men 0.96, 0.94–0.98, per each extra hour of sports per week). These results highlight the problem with ignoring the threshold effect of participation in specific activities. In analyses on cause-specific mortality, statistical power was reduced, but the main results were similar to the results on all-cause mortality and not confined to a specific cause of death.
The major strengths of our study are the large number of deaths and the standardized data collection methods for physical activity and potential confounders. The study also has some limitations in that the information on leisure time physical activity was self-reported and subject to some misclassification. However, it seems unlikely that the misclassification is severe enough to reverse a truly beneficial effect of increasing hours of activity, given the observed adverse associations. However, a validation study showed higher correlation of sports and cycling with energy expenditure,28 suggesting that the strength of association between the remaining types of physical activity and mortality could be underestimated. The cohort may represent only a selected subgroup of the Danish population as only 35% of the invited subjects chose to participate, but it is not obvious why the relation between physical activity in hours per week and mortality should be affected by this self-selection. A possible source of bias is prevalent disease; instead of excluding the first years of follow-up, the mortality rates were allowed to depend on time in study. Furthermore, no signs of time-dependent effects of the variables were seen when investigated through separate analyses performed for events within three intervals of follow-up time with boundaries at 2 and 4 years after baseline. Finally, to diminish the potential bias caused by prevalent cardiovascular disease, adjustments were made for previous cardiovascular diseases and current cardiovascular risk factors. None of these actions affected the results markedly.
For most of the leisure time physical activities measured, active people had lower all-cause mortality compared with nonactive. However, except for women and housework, none of the six leisure time activities showed any risk-reducing trends with longer time spent on the activity. In fact, similar or even higher mortality rates were observed. Lee and Paffenbarger32 reported similar observations, and there may be an upper threshold for the beneficial effect of physical activity. Such a threshold may in fact be quite low. In the Norwegian Women and Cancer Study, the beneficial effect of physical activity was most prominent in the lower range of the activity,9 and in a recent study on Taiwanese men and women participating in a medical screening program, only a minimum amount of physical activity was needed to reduce all-cause mortality.23 It is possible that the important component is to minimize total inactivity like sitting or lying down. In fact, Weller and Corey22 showed a clear dose–response pattern for an adverse effect of hours of sitting, in contrast to the effect of walking, where the association was confined to a higher mortality in the lowest quartile that included the totally inactive. Similar results have been found by a range of other studies, all showing markedly higher mortality at the lowest physical activity level.3,6,9,11–13,23 In particular, watching TV and reading more than 6 hours per day has been associated with higher mortality in a study of Swedish, postmenopausal women,11 and in a study of US adults, watching TV 5 or more hours per day was associated with higher mortality.33 Unfortunately, in the only study on doubly labeled water as an objective measure of daily energy expenditure, the authors presented tertiles of energy spent on physical activity.34 Consequently, it is not possible to evaluate whether the association between physical activity and mortality is because of a steady improvement or is confined to a threshold effect in the extremes of the physical activity—these two different biological scenarios may lead to identical statistical results regardless both when assuming a linear effect (as opposed to linear splines) and when dividing according to the tertiles (as opposed to quartiles or quintiles). However, the National Health and Nutrition Examination Survey used accelerometry to evaluate time spent on moderate to vigorous activity in relation to all-cause mortality and found markedly higher mortality in the two lowest quartiles of activity.35 Similar findings have been reported in other studies using accelerometry.8,14,20,36
The findings of increased mortality with more time spent on walking and gardening beyond an average of half an hour per week is remarkable. However, the size of the effects was very small (less than 3% higher mortality per each additional hour of activity) and the importance of these findings seems negligible.
In conclusion, for most of the leisure time activities in this study, a lower mortality was observed among those who participated in the activity compared with those who did not. No additional benefit was found with more time spent on the physical activity, after separating the effect of participation from the dose–response effect. Consequently, our results support the hypothesis that exceeding a rather low threshold of activity may account for most of the beneficial effect of leisure time activity—a pattern that could be caused by biology, or confounding, or both. In relation to longevity, avoidance of inactivity seems to be an important aspect of physical activity, and the lack of a dose–response effect of time spent on activity does not support a beneficial effect of increasing amounts of physical activity.
We acknowledge the contributions of Katja Boll (programmer, Danish Cancer Society), Connie Stripp (dietician, Danish Cancer Society), and Jytte Fogh Larsen (secretary, Danish Cancer Society) in the collection and management of data. Mette Suntum is thanked for excellent statistical inspiration.
1. World Health Organization. Global Health Risks: Mortality and Burden of Disease Attributable to Selected Major Risks. 2009 Geneva World Health Organization
2. Woodcock J, Franco OH, Orsini N, Roberts I. Non-vigorous physical activity and all-cause mortality: systematic review and meta-analysis of cohort studies. Int J Epidemiol. 2011;40:121–138
3. Andersen LB, Schnohr P, Schroll M, Hein HO. All-cause mortality associated with physical activity during leisure time, work, sports, and cycling to work. Arch Intern Med. 2000;160:1621–1628
4. Andersen LB. Relative risk of mortality in the physically inactive is underestimated because of real changes in exposure level during follow-up. Am J Epidemiol. 2004;160:189–195
5. Barengo NC, Hu G, Lakka TA, Pekkarinen H, Nissinen A, Tuomilehto J. Low physical activity as a predictor for total and cardiovascular disease mortality in middle-aged men and women in Finland. Eur Heart J. 2004;25:2204–2211
6. Bijnen FC, Caspersen CJ, Feskens EJ, Saris WH, Mosterd WL, Kromhout D. Physical activity and 10-year mortality from cardiovascular diseases and all causes: The Zutphen Elderly Study. Arch Intern Med. 1998;158:1499–1505
7. Blair SN, Kohl HW 3rd, Paffenbarger RS Jr, Clark DG, Cooper KH, Gibbons LW. Physical fitness and all-cause mortality. A prospective study of healthy men and women. JAMA. 1989;262:2395–2401
8. Blair SN, Kampert JB, Kohl HW 3rd, et al. Influences of cardiorespiratory fitness and other precursors on cardiovascular disease and all-cause mortality in men and women. JAMA. 1996;276:205–210
9. Borch KB, Braaten T, Lund E, Weiderpass E. Physical activity and mortality among Norwegian women - the Norwegian Women and Cancer Study. Clin Epidemiol. 2011;3:229–235
10. Bucksch J. Physical activity of moderate intensity in leisure time and the risk of all cause mortality. Br J Sports Med. 2005;39:632–638
11. Carlsson S, Andersson T, Wolk A, Ahlbom A. Low physical activity and mortality in women: baseline lifestyle and health as alternative explanations. Scand J Public Health. 2006;34:480–487
12. Hillsdon M, Thorogood M, Murphy M, Jones L. Can a simple measure of vigorous physical activity predict future mortality? Results from the OXCHECK study. Public Health Nutr. 2004;7:557–562
13. Hu G, Tuomilehto J, Silventoinen K, Barengo NC, Peltonen M, Jousilahti P. The effects of physical activity and body mass index on cardiovascular, cancer and all-cause mortality among 47 212 middle-aged Finnish men and women. Int J Obes (Lond). 2005;29:894–902
14. Kampert JB, Blair SN, Barlow CE, Kohl HW 3rd. Physical activity, physical fitness, and all-cause and cancer mortality: a prospective study of men and women. Ann Epidemiol. 1996;6:452–457
15. Khaw KT, Jakes R, Bingham S, et al. Work and leisure time physical activity assessed using a simple, pragmatic, validated questionnaire and incident cardiovascular disease and all-cause mortality in men and women: The European Prospective Investigation into Cancer in Norfolk prospective population study. Int J Epidemiol. 2006;35:1034–1043
16. Kushi LH, Fee RM, Folsom AR, Mink PJ, Anderson KE, Sellers TA. Physical activity and mortality in postmenopausal women. JAMA. 1997;277:1287–1292
17. Rosengren A, Wilhelmsen L. Physical activity protects against coronary death and deaths from all causes in middle-aged men. Evidence from a 20-year follow-up of the primary prevention study in Göteborg. Ann Epidemiol. 1997;7:69–75
18. Schnohr P, Lange P, Scharling H, Jensen JS. Long-term physical activity in leisure time and mortality from coronary heart disease, stroke, respiratory diseases, and cancer. The Copenhagen City Heart Study. Eur J Cardiovasc Prev Rehabil. 2006;13:173–179
19. Schooling CM, Lam TH, Li ZB, et al. Obesity, physical activity, and mortality in a prospective chinese elderly cohort. Arch Intern Med. 2006;166:1498–1504
20. Stevens J, Evenson KR, Thomas O, Cai J, Thomas R. Associations of fitness and fatness with mortality in Russian and American men in the lipids research clinics study. Int J Obes Relat Metab Disord. 2004;28:1463–1470
21. Sundquist K, Qvist J, Sundquist J, Johansson SE. Frequent and occasional physical activity in the elderly: a 12-year follow-up study of mortality. Am J Prev Med. 2004;27:22–27
22. Weller I, Corey P. The impact of excluding non-leisure energy expenditure on the relation between physical activity and mortality in women. Epidemiology. 1998;9:632–635
23. Wen CP, Wai JP, Tsai MK, et al. Minimum amount of physical activity for reduced mortality and extended life expectancy: a prospective cohort study. Lancet. 2011;378:1244–1253
24. Tjønneland A, Christensen J, Thomsen BL, et al. Lifetime alcohol consumption and postmenopausal breast cancer rate in Denmark: a prospective cohort study. J Nutr. 2004;134:173–178
25. Tjønneland A, Olsen A, Boll K, et al. Study design, exposure variables, and socioeconomic determinants of participation in Diet, Cancer and Health: a population-based prospective cohort study of 57,053 men and women in Denmark. Scand J Public Health. 2007;35:432–441
26. Cust AE, Smith BJ, Chau J, et al. Validity and repeatability of the EPIC physical activity questionnaire: a validation study using accelerometers as an objective measure. Int J Behav Nutr Phys Act. 2008;5:33
27. Pols MA, Peeters PH, Ocké MC, Slimani N, Bueno-de-Mesquita HB, Collette HJ. Estimation of reproducibility and relative validity of the questions included in the EPIC Physical Activity Questionnaire. Int J Epidemiol. 1997;26(suppl 1):S181–S189
28. Wareham NJ, Jakes RW, Rennie KL, et al. Validity and repeatability of a simple index derived from the short physical activity questionnaire used in the European Prospective Investigation into Cancer and Nutrition (EPIC) study. Public Health Nutr. 2003;6:407–413
29. Greenland S. Dose-response and trend analysis in epidemiology: alternatives to categorical analysis. Epidemiology. 1995;6:356–365
30. Greenland S. Avoiding power loss associated with categorization and ordinal scores in dose-response and trend analysis. Epidemiology. 1995;6:450–454
31. World Health Organization. Obesity: Preventing and Managing the Global Epidemic. Report of a WHO consultation. WHO Technical Report Series no. 894. 2002 Geneva World Health Organization
32. Lee IM, Paffenbarger RS Jr. Do physical activity and physical fitness avert premature mortality? Exerc Sport Sci Rev. 1996;24:135–171
33. Matthews CE, George SM, Moore SC, et al. Amount of time spent in sedentary behaviors and cause-specific mortality in US adults. Am J Clin Nutr. 2012;95:437–445
34. Manini TM, Everhart JE, Patel KV, et al. Daily activity energy expenditure and mortality among older adults. JAMA. 2006;296:171–179
35. Koster A, Caserotti P, Patel KV, et al. Association of sedentary time with mortality independent of moderate to vigorous physical activity. PLoS One. 2012;7:e37696
36. Arraiz GA, Wigle DT, Mao Y. Risk assessment of physical activity and physical fitness in the Canada Health Survey mortality follow-up study. J Clin Epidemiol. 1992;45:419–428