METHODS

To identify papers for this review, we conducted a literature search using Medline (US National Library of Medicine) to look for papers published between January 1966 and December 2000. The keywords used were related to physical activity (physical activity, physical fitness, cardiorespiratory fitness, exercise, or walking) and mortality (mortality, death, or fatal outcome).

For inclusion in this review, the papers had to present results specifically for women, use the end point of all cause mortality (rather than cause specific mortality, such as cardiovascular disease mortality), and examine the association of physical activity or fitness with all cause mortality. We excluded review papers and papers not published in the English language because we did not have the facility to translate papers written in other languages. In addition, we checked the reference lists of papers included in this review, as well as review papers not included, to search for other eligible papers not identified by our computerised literature search. By these means, we identified 38 papers.

Two authors independently reviewed each of the papers and summarised the data (table 1). In two circumstances,^5–8 the same data were used to examine the association of physical fitness and all cause mortality in women. We present these data only once in table 1, choosing the publication in which the data were analysed in more detail.^6,8

RESULTS

A total of 37 prospective cohort studies and one retrospective cohort study that examined the association between physical activity or fitness and all cause mortality in women were identified (table 1). Most of the studies were conducted in western, industrialised countries, with half (19 studies) being based in the United States. The Aerobics Center Longitudinal Study^5–8 and the Framingham Heart Study^9–12 each provided the data for four publications, comprising 21% of the studies reviewed. Sixteen (42%) studies were based in Europe, including five from Finland^13–17 and four from Germany.^18–21 The remaining three studies were based in Canada^22,23 and Israel.²⁴

Most (32 (86%)) of the studies focused exclusively on physical activity, while four assessed only physical fitness^6,8,16,25 and another two investigated both physical activity and physical fitness.^14,22 Of the 34 studies investigating physical activity, 13 captured total physical activity and 23 measured leisure time physical activity (the sum exceeds 34 because some studies had both measures). Only six of the 23 studies that measured leisure time physical activity also measured occupational physical activity^17,26–28 or non-recreational physical activity.^23,29 Of the studies examining physical fitness, three used a maximal exercise test on a treadmill.^6,8,25

Are higher levels of physical activity or fitness associated with lower death rates in women?

The first question we sought to answer was whether women who are physically active or physically fit postpone mortality compared with their less active or less fit peers. Of the 38 studies included in this review, 23 reported significant inverse associations between physical activity (or physical fitness) and death rate in at least one subgroup.^{6,8–11,14–17,21,23,25–36} Another nine studies observed at least a 20% difference in death rates between the most and least active women, which was not statistically significant during the period of study.^{18–20,22,24,37–40} The lack of statistical significance may have been due, in part, to the small number of deaths in these studies. In the remaining six studies, investigators reported less than a 20% difference in death rates during follow up between the most and least active (or fit) women.^{5,7,12,13,41,42}

We calculated the median relative risk, comparing the most with the least active (or fit) women, from all 38 studies to estimate an average magnitude of effect of physical activity or fitness on mortality. For all 38 studies, the median relative risk was 0.66 (table 2)—that is, the most active or fit women in these studies were 34% less likely, on average, to die during the period of study. We then estimated the median relative risk separately for studies of total physical activity, leisure time physical activity, occupational or non-recreational physical activity, and physical fitness. Based on 11 studies, the median relative risk for women who were most active with regard to total physical activity, compared with those who were least active, was 0.75.^{9–11,18,21,23,30,34,37,39,41} For leisure time physical activity (23 studies), the corresponding median relative risk was 0.66.^{5,7,13–15,17–20,22–24,26–29,31–33,35,36,38,42} This figure was 0.54 for occupational or non-recreational physical activity (six studies),^{17,23,26–29} and 0.55 for physical fitness (six studies).^{6,8,14,16,22,25} Because the latter two estimates were based on only a small number of studies, there is less confidence in these values.

Is the magnitude of association similar to that seen in men?

To examine whether the association of physical activity or fitness with all cause mortality in women is of similar magnitude to that seen in men, we conducted a similar exercise for studies conducted in men (reviewed in references 4, 43, 44; and references 6, 9, 13, 14, 16, 17, 20, 21, 29–31, 36, 37, 39, 45, 46). The median relative risk for all cause mortality in men, comparing most active with least active, was 0.65 based on all studies of men that we identified (n = 67). For studies of total physical activity, this was 0.66 (17 studies), for leisure time physical activity, it was 0.70 (30 studies), for occupational or non-recreational physical activity, it was 0.65 (10 studies), and for physical fitness, it was 0.55 (10 studies). These data indicate similar associations in both sexes.

Is there a dose-response relation?

To assess whether a dose-response relation exists, we examined papers in which investigators categorised subjects into at least three levels of physical activity or fitness. We identified 28 such studies. Sixteen of these showed evidence (not always statistically evaluated by the authors) of an inverse dose-response relation between physical activity (or fitness) and all cause mortality in at least one subgroup of subjects (table 1).^{8,9,15,16,19,20,22,25,27,29–33,37,38} Of the remaining 12 studies, eight reported data compatible with a J shaped or reverse J shaped curve,^{5,7,10,11,14,23,26,42} and four observed non-significant associations of physical activity with all cause mortality.^18,24,35,41

For a more formal examination of the dose-response relation, we assessed studies that performed statistical tests for a trend of declining all cause death rates with increasing volume of physical activity (or fitness). Of 13 studies identified, eight observed significant p values for trend.^{8–11,15,23,25,32} The remaining five studies reported non-significant trends.^5,7,24,41,42

How much physical activity is required?

To address this question, we first investigated the volume (amount) of energy expenditure associated with significantly lower death rates in women. Because physical activity was assessed very differently in the various studies, it was difficult to compare volume of energy expenditure across studies. We attempted to make comparisons across studies by estimating the volume of energy expenditure as kJ (1 kcal ≈ 4.2 kJ) or MET hours (1 MET is the energy expended while resting quietly; it is equivalent to 4.2 kJ/kg body weight per hour) in studies in which sufficient data were provided.

Significantly lower death rates were observed at these approximate volumes of energy expenditure: 1050–6720 kJ/week (250–1600 kcal/week) or 0.5–3.5 MET hours/day³²; 2100 kJ/week (500 kcal/week) or 1 MET hour/day^9,27; 2100–2940 kJ/week (500–700 kcal/week) or 1.0–1.5 MET hours/day²⁶; 4200 kJ/week (1000 kcal/week) or 2.2 MET hours/day¹⁵; 5460 kJ/week (1300 kcal/week) or 2.8 MET hours/day²³; 5880 kJ/week (1400 kcal/week) or 3 MET hours/day¹¹; 9660 kJ/week (2300 kcal/week) or 5 MET hours/day.¹⁰

We next examined what intensity of physical activity is required for lower death rates. Few studies collected or analysed data on intensity of physical activity, precluding firm conclusions. In the Iowa Women's Health Study,³² participation in moderate intensity physical activity was associated with lower death rates. However, in the Canada Fitness Survey,²² non-vigorous physical activity was not associated with lower death rates, but an inverse association, of borderline significance, was observed with vigorous physical activity. In the Copenhagen City Heart Study,²⁶ participation in sports was associated with additional risk reduction among women who were already participating in other activities of light, moderate, or vigorous intensity.

Several studies examined frequency^20,32,35 or duration^5,18,42 of physical activity in relation to all cause mortality. However, because these studies did not control for the volume of energy expenditure, the findings on frequency or duration essentially reflect the findings for volume of energy expended. Therefore, the independent associations with frequency or duration could not be evaluated in this review.

Is there effect modification by age?

Finally, we were interested in whether physical activity was beneficial for all cause mortality in both younger and older women. We identified 11 studies^{9,14–17,26–29,32,41} in which data were presented separately for women aged <65 years at baseline and 13 studies^{10,13,24,26,29–35,39,42} in which data were provided for those aged ≥65 years at baseline. In women aged <65 and ≥65 years, the median relative risk comparing most active with least active women was 0.67 and 0.58, respectively. The use of other age cut off points—for example, 55 or 60 years—produced similar relative risks for both younger and older women. In six studies, investigators provided findings separately for women of different ages, ranging from 20 to ≥75 years.^{20,26,29,32,41,42} In all but one study²⁹ the magnitude of risk reduction was lower among older women than among younger women.

DISCUSSION

There is clear evidence from 38 studies conducted primarily in the United States and Europe that physically active women postpone mortality compared with inactive women. Formal meta-analysis using raw data from investigators is beyond the scope of this review; instead we conducted a systematic qualitative review using published data. This review indicates that the median risk reduction over the period of study is 34%, a magnitude of association that closely parallels that for men.

Most studies reviewed reported at least a 20% difference in death rate between the most active and least active women, and most results were statistically significant. In the six studies that did not find an association, physical assessments may not have been sufficiently precise or valid. Although many other studies also suffered from similar limitations in physical activity measurement, they still reported significant results. Therefore the reported findings are likely to underestimate the magnitude of the true association.

Although 28 studies classified women into at least three levels of physical activity, allowing for assessment of dose-response, only 13 formally tested for a trend of declining all cause mortality with increasing levels of physical activity. Eight observed significant inverse trends. Future studies of physical activity and mortality in women need to use measurements that allow a more detailed analysis of dose-response effects. Whether the apparent inverse association follows a linear or non-linear trend also merits consideration, as few studies have examined this.

Because the prevalence of physical inactivity is high,^3,47 an issue of public health importance is the minimum amount of physical activity required to postpone early mortality in women. Few published studies have provided specific estimates of energy expenditure, again highlighting the need for better quantitative assessments of physical activity in epidemiological research. We converted various measures of physical activity to kJ/week or kcal/week, assuming the average body weight of a woman to be 65 kg. On the basis of studies from which volume of energy expenditure could be estimated, an energy expenditure of about 4200 kJ/week (1000 kcal/week) seemed to be sufficient to avert premature mortality. This level of physical activity is in line with recommendations from various US organisations that had based these targets primarily on data in men.^1–3 In fact, a few studies suggest that even lower levels of energy expenditure may be associated with significant reduction in all cause mortality in women.^9,26,27,32

We also attempted to assess the role of physical activity intensity in postponing mortality in women. Studies of physical activity intensity and mortality in women are limited and inconsistent in their findings. For example, the Iowa Women's Study separately examined moderate and vigorous activity, noting an inverse association between moderate intensity physical activity and mortality.³² The inverse association with vigorous intensity physical activity was only of borderline significance, presumably because few women undertook vigorous activities. However, Villeneuve et al²² reported from the Canada Fitness Survey that non-vigorous leisure time physical activity in subjects with no vigorous leisure time physical activity was not associated with death rate. Further studies are needed to clarify the role of physical activity intensity in postponing mortality in women. Studies of physical activity intensity and risk of coronary heart disease in women have yielded more consistent findings. For example, the Nurses' Health Study⁴⁸ and Women's Health Study⁴⁹ both indicate that moderate intensity activities are associated with lower risk of coronary heart disease. No papers have examined whether, with the same level of energy expenditure, activity bouts of shorter duration (and greater frequency) or longer duration (and lesser frequency) have different effects on all cause mortality in women. Several intervention studies have shown that total physical activity may be more important than duration or frequency for coronary risk factors,^50,51 and large scale observational studies are necessary to confirm or refute these findings for clinical end points such as premature mortality.

When we examined the association of physical activity with death rates in women of different ages, we observed that higher levels of physical activity reduce death rates in both younger and older women. These benefits across all ages of women extend beyond just a reduction in premature mortality. Ferrucci et al⁵² notes that higher levels of physical activity are also associated with fewer years of disability before death, helping women to live more enjoyable lives.

Several limitations of the studies reviewed need to be considered. The assessment of physical activity was often based on instruments developed for men. For example, care of children and household chores were often not accounted for. In the Canada Fitness Survey, Weller et al^23,53 observed that, for women, non-leisure time physical activity comprised most (82%) of their daily energy expenditure. In this study, investigators did not find an association between physical activity and death rates when examining only leisure time energy expenditure. However, after taking into account non-leisure activities, an inverse relation was present. This underscores the importance of assessing non-leisure activities, such as those mentioned above. The challenge is to develop reliable and valid instruments for measuring such activities.

Another limitation is that residual confounding by other lifestyle and dietary habits may be a plausible explanation for an inverse association between physical activity and mortality. Physically active women are likely to be healthy in other ways such as eating behaviour, body weight, and smoking. However, many studies did adjust for potential confounders, and continued to observe significant inverse associations. As all the studies reviewed are observational epidemiological studies, a causal relation cannot be definitely established. However, randomised clinical trials, which provide more rigorous data for cause and effect inferences, are unlikely to be conducted because of the lack of feasibility in maintaining compliance with physical activity over the long term needed in studies of mortality.

We also had difficulty assessing minimum levels of physical activity required for averting premature mortality. Different cut off points for physical activity were used, not all of which allowed interpretation of the amount of physical activity required.

We were interested in the association of physical activity and death rates in women before and after the menopause, but no studies provided this information. When we examined older women separately using the cut off points of 65, 60, or 55 years, as proxies for identifying postmenopausal women, we found that postmenopausal women also showed inverse associations between physical activity and death rates.

Finally, changes in physical activity were typically not considered in the studies reviewed. Lissner et al²⁷ reported that those who decreased their physical activity level were at increased mortality risk relative to those who maintained activity levels. Recent results from the Nurses' Health Study (not included in this review because the study was published outside the time window for inclusion) also support these findings.⁵⁴

Physical inactivity is pervasive: more than 60% of American women do not engage in the recommended amount of physical activity, and 25% are not active at all.³ Similar levels of inactivity are observed in England: according to the Health Survey for England in 1994,⁴⁷ 60% of women were underactive and 20% were inactive. This is unfortunate, as the present review adds to the list of benefits associated with physical activity in postponing premature mortality in women. This review also supports current physical activity guidelines of at least 30 minutes of physical activity of moderate intensity on most, preferably all, days of the week.^1–3,55 Although these guidelines were developed primarily from data in men, this review supports its applicability for women as well.