Secondary Logo

Journal Logo

Review Article

Physical Activity and Breast Cancer: A Systematic Review

Monninkhof, Evelyn M.*; Elias, Sjoerd G.*; Vlems, Femke A.; van der Tweel, Ingeborg; Schuit, A Jantine§; Voskuil, Dorien W.; van Leeuwen, Flora E.on behalf of TFPAC

Author Information
doi: 10.1097/01.ede.0000251167.75581.98

ArticlePlus

Click on the links below to access all the ArticlePlus for this article.

Please note that ArticlePlus files may launch a viewer application outside of your web browser.

Most of the established risk factors for breast cancer such as family history of the disease, early age at menarche, late age at menopause, late age at first childbirth, and nulliparity are not easily amenable to intervention. In contrast, physical inactivity is a modifiable risk factor that has been associated with increased risk of breast cancer.1–3 Physical activity may protect against breast cancer through reduced lifetime exposure to sex steroid hormones, reduced exposure to insulin and insulin-like growth factors, and prevention of overweight and obesity.4 Increase of physical activity is therefore a potentially promising preventive measure against breast cancer.5–10

Numerous observational studies have assessed the association between physical activity and breast cancer risk. Although most studies reported that high physical activity is associated with decreased risk, some reported the contrary, whereas others reported no relation. The lack of consistency may be attributable to differences in methods for assessing physical activity such as the types of physical activity covered (eg, occupational, recreational, and household). Studies also vary greatly as to the ages at which the physical activity level was assessed. So far, the critical time periods in life with respect to physical activity and breast carcinogenesis are unknown as are the optimal frequency, duration, and intensity of physical activity needed to reduce risk of breast cancer. Inconsistent results have been reported regarding these aspects of activity.

Previous reviews2,3,11 did not perform a meta-analysis because of the large heterogeneity in exposure measurements and study populations, which complicates the pooling of results. Moreover, these reviews did not extensively consider the impact of study quality.

The purpose of this review is to provide an update of the epidemiologic evidence for an association between physical activity and breast cancer risk. Additionally, we quantify the methodologic quality of studies and explore whether study quality contributes to the discrepancies in outcome among the studies. Furthermore, we aimed to identify critical time periods in life during which activity might have the strongest effects on breast cancer risk. Finally, we evaluated whether the association between physical activity and breast cancer risk is modified by menopausal status or body mass index (BMI) as has been suggested by several authors.

METHODS

Search and Selection of Literature

We identified studies through a systematic review of literature available on PubMed through February 2006. The databases were searched using the following terms: “physical activity” OR “exercise, physical” OR “exercises, physical” OR “physical exercise” OR “physical exercises” OR “sedentary lifestyle” AND “breast neoplasms” OR “breast cancer”. To identify papers describing several site-specific cancers (including breast cancer), we combined the physical activity search terms with “neoplasms” OR “cancer”. From these publications, the bibliographic lists were hand-searched for additional papers.

Two reviewers (E.M. and S.E.) independently selected studies for inclusion in the review. The criteria for inclusion were case–control or cohort studies investigating the association between physical activity and breast cancer in males or females with incidence, prevalence, or mortality as the end point, more than 10 cancer cases included in the analysis, published in English, and providing sufficient information to calculate a confidence interval for the risk estimate. When multiple reports on the same source population had been published, we included the paper with the most extensive follow-up or most relevant measure of physical activity.

We chose to include only studies assessing leisure time activity or total activity (occupational as well as nonoccupational). Insight into the association between total physical activity and breast cancer is of etiologic interest, whereas the association with leisure time activity may give insight into possibilities for preventive strategies. Studies based only on occupational activity were excluded from this analysis for 2 reasons. First, most of these studies used crude methods for exposure assessment (eg, job title). Second, in earlier years, a large number of women were not engaged in paid employment or were employed for only a limited time; consequently, occupational activity comprised only a small proportion of total activity in these women. Both issues would have inevitably led to substantial errors in measuring physical activity in these studies and subsequently to an underestimate of the strength of the associations with breast cancer risk.

Data Extraction

The data extraction and study quality assessment were independently performed by 2 reviewers (E.M. and S.E.). Any disagreement was resolved by consensus or by consultation with a third reviewer (F.v.L.). Interrater agreement of the data extraction of the 2 reviewers was 92% for the cohort studies and 94% for the case–control studies. Data were extracted using standardized data extraction forms specifically created for this review and were subsequently entered into a database. All data entry was double-checked. We documented study size, characteristics of the study population, components of physical activity, and methodologic characteristics. Components of physical activity included the source of activity (ie, total activity or leisure time activity), the dose of the activity (ie, frequency, intensity, duration), and the time period in life during which it was performed. The results (fully adjusted risk estimates and 95% confidence intervals [CIs] for the highest vs lowest level of activity) for total and leisure time activity on breast cancer risk were documented separately. We also looked for evidence of a dose–response relationship by a subjective evaluation of the ordering of the risk estimates irrespective of the P value of a trend test. If studies reported various exposure measures, we used the following hierarchical documentation order: duration of physical activity (hours/week), metabolic intensity (MET) (duration * intensity), energy expenditure (KJ or Kcal), frequency, and lastly a subjective composite score (eg, simple ordinal ranking of physical activity levels). We chose duration of physical activity (hours/week) as our primary exposure measure because it was reported in the majority of the studies and it is most easily translated into preventive measures. If risk estimates were reported stratified by menopausal status and no overall estimate was presented, we pooled the pre- and postmenopausal risk estimates by a precision weighted mean procedure.12 For studies assessing activity in several time periods during adult life, the following hierarchical order was used: lifetime physical activity, recent activity, and activity during adulthood. Additionally, because of our special interest in whether physical activity in certain periods of life would be most effective, we also considered separately the results of studies assessing more life periods, change of activity patterns, or activity during adolescence.

To be able to investigate whether breast cancer risk is modified by menopausal status or BMI, we extracted the relevant data of all papers addressing these issues. The role of BMI is also important with respect to confounding. Physical activity and BMI tend to be inversely correlated, because regular physical activity is an important method of controlling weight. Furthermore, BMI is associated with postmenopausal breast cancer risk. The results of adjustment for BMI, however, should be interpreted cautiously because BMI could be in the causal pathway between physical activity and breast cancer risk. If this is the case, adding BMI as a confounding factor in the statistical analysis would attenuate the effect of physical activity. To assess the potential confounding or mediating effect of BMI, we evaluated the effect of adjustment for BMI in all of the studies for which it was reported.

Quality Assessment

Because no suitable quality assessment instrument was available for this research topic, we developed a quality scoring system that captured both generic methodologic issues and issues specific to our subject (available with the electronic version of this article). The scoring system can be applied to cohort and case–control studies, which enables grouping of the results of both study designs. The items of the scoring system were categorized according to 3 important sources of error in observational studies (the major headings), ie, selection, misclassification, and confounding bias. Criteria for physical activity assessment methods were partly adopted from the review by Powell and colleagues.13 The quality scoring system contained 19 items (selection = 5, misclassification = 11, and confounding = 3). In this research area, the potential for confounding bias is judged less important. Studies that adequately adjusted for potential confounders show that the magnitude of the risk estimates do not materially change after adjustment, whereas bias due to selection or misclassification is expected to lead to larger effects on risk estimates. Thus, the major headings (selection, misclassification, and confounding bias) are weighted 2:2:1. The maximum attainable score is 105 (see Appendix 1). In the tables, the quality score is presented as percentage of the maximum attainable score.

Analysis

Cohort studies and case–control studies were categorized according to study quality irrespective of whether total or leisure time activity was reported. Studies were classified into 2 groups (higher and lower quality) based on the median study quality of all studies (cohort as well as case–control studies).

We assessed heterogeneity statistically using a formal test14 and found statistical evidence for heterogeneity for total and leisure time activities in cohort and case–control studies. The effect estimates for pre- and postmenopausal breast cancer were also statistically heterogeneous. In light of this statistical evidence, as well as because of the subjective indications of heterogeneity, we decided not to provide pooled estimates. A qualitative summary was undertaken according to a best-evidence synthesis. Until now, there have been no standard criteria for such a synthesis in cancer epidemiology. Therefore, we developed the criteria for the best-evidence synthesis ourselves. In the best-evidence synthesis, the available evidence for an inverse association between physical activity and breast cancer risk was summarized by taking into account the methodologic quality of the studies and the consistency of the available evidence (Table 1). We also performed the best-evidence synthesis on all studies irrespective of methodological quality. An inverse association between physical activity and breast cancer risk within a study was defined as a risk estimate for the highest versus the lowest level of activity of below 0.8 irrespective of statistical significance. No association was considered as a risk estimate between 0.8 and 1.25, and an increased risk was a risk estimate greater than 1.25. The consistency of study findings was defined on the basis of the proportion of studies that reported a decreased risk, no association, and an increased risk (see Table 1). In a sensitivity analysis, we evaluated the relative impact of changing our definition of a decreased risk in the best-evidence synthesis from 0.8 to 0.85 and 0.75. We also evaluated the impact of focusing on only statistically significant results. Furthermore, we evaluated whether exclusion of male breast cancer studies changed the conclusions.

TABLE 1
TABLE 1:
Levels of Evidence for an Inverse Association Between Physical Activity and Breast Cancer Risk

A trend analysis based on summarized dose–response data was performed to obtain insight into the change in breast cancer risk by increasing activity levels.15 For this analysis, we selected studies with a similar exposure measure, ie, studies measuring physical activity in hours per week or frequency of physical activity in times per week. For frequency in times per week, we assumed that the duration of one episode of activity equals 1 hour. Because the data of the 4 cohort studies that assessed duration or frequency of physical activity were insufficient for analysis, the trend analysis was restricted to the case–control studies. In the trend analysis, the lowest activity level was chosen as the reference category. For each study, a slope and its associated variance were estimated using the adjusted odds ratios at various dose levels.16,17 A correction was applied for the dependence of each dose-specific odds ratio with respect to the same reference category. Subsequently, the study-specific slopes were combined into a weighted average slope using a random effects meta-analysis method.17

Graphic displays were used to investigate whether the risk estimates differed between studies with lower and higher quality. To examine whether a priori selected design factors explained heterogeneity between the results from different studies, we performed metaregression analyses.18 The design factors studied included the methodologic quality score (selection, misclassification, and confounding score) and the number of cases. All factors were included in a multivariate model (full model). Subsequently, we evaluated the explained variance (R2) and the contribution of the different design factors. Possible publication bias was investigated by funnel plots. Risk estimates were plotted against the number of breast cancer cases as a marker for study size. Funnel plots show the variation in study results according to study size, and asymmetric funnel plots may indicate publication bias.19

RESULTS

We identified 76 potentially relevant articles concerning physical activity and breast cancer risk. Twenty-eight articles were excluded because occupational activity alone was assessed (n = 14), a confidence interval could not be calculated (n = 1), or the article was part of multiple publications on the same source population (n = 13). The main characteristics and results of the remaining 19 cohort studies20–38 and 29 case–control studies39–67 are summarized in Tables 2 and 3, respectively. The studies were published between 1994 and 2006. The number of cases ranged from 46 to 3424 in cohort studies and from 81 to 6888 in case–control studies. Three case–control studies were based on male breast cancer cases only.52,55,60 Three cohort and 6 case–control studies assessed total activity. There was considerable variation among the studies with respect to the age of the participants, the physical activity measures used, the period of life for which they obtained activity measurement, and the length of follow-up (cohort design). The quality score as percentage of the maximum score ranged from 51% to 82% (median 71%) for the cohort studies and from 32% to 80% (median 67%) for the case–control studies.

TABLE 2
TABLE 2:
Characteristics and Results of the Cohort Studies Stratified by Source of Activity
TABLE 2
TABLE 2:
(Continued)
TABLE 3
TABLE 3:
Characteristics and Results of Case–Control Studies Stratified by Source of Activity
TABLE 3
TABLE 3:
(Continued)
TABLE 3
TABLE 3:
(Continued)
TABLE 3
TABLE 3:
(Continued)
TABLE 3
TABLE 3:
(Continued)
TABLE 3
TABLE 3:
(Continued)

Cohort Studies

Figure 1A shows the results of the cohort studies stratified by quality score. The 3 studies assessing total activity show inconsistent results: one study37 found a decrease in risk (RR <0.8; see “Methods”), one study observed no association,29 and one study38 found an increased breast cancer risk (RR >1.25) with evidence for a dose–response effect.

FIGURE 1.
FIGURE 1.:
Results of studies of physical activity and breast cancer risk according to study design (A, cohort; B, case–control), source of activity (◆ leisure time, ⋄ total), and study quality. Quality score is expressed as a percentage of the maximal attainable score. Depicted are risk estimates for the highest versus the lowest category of activity with corresponding 95% confidence intervals. A decreased risk is defined as a risk estimate of <0.80, no association as a risk estimate between 0.80 and 1.25 (shaded area of figure), and an increased risk as a risk estimate higher than 1.25. Confidence intervals without a bar at the end extend beyond the depicted range.

Eight21–28 of the 17 cohort studies on leisure time activities showed a decreased breast cancer risk (RR <0.8) and the other 920,29–36 reported no association. Figure 1A shows that in general, the lower quality studies (ie, a quality score below the median) show greater risk reductions than the higher quality studies. Except for the Iowa 65+ Health Study,21 the risk reductions of the studies with a higher quality score ranged from 21% to 39%. The risk reduction of 80% in the Iowa 65+ Health Study should be considered with caution because this estimate was based on only 2 cases in the highest activity level. Additionally, the study population was considerably older than in the other studies. Evidence for an inverse dose–response relationship was seen in 6 of 7 studies that reported a decreased breast cancer risk for leisure time activity in more than 2 categories.21–23,25,26,28

Case–Control Studies

Figure 1B shows the results of the case–control studies on male and female breast cancer risk stratified by quality score. Four of the 6 studies40,43,45,52,53,59 assessing total activity found a decrease in breast cancer risk (RR <0.8) ranging from 21% to 52%. A dose–response relationship was indicated by the studies of Verloop et al40 and John et al53; however, no P values for trend were presented. Although Mittendorf et al45 and Johnson et al52 presented a significant test for trend, the graphic presentation does not suggest a dose–response effect.

Among 28 case–control studies39–44,46–67 that reported data on leisure time activity and breast cancer risk, 1439,41,44,46,48–50,55,57,58,60,63,64,66 observed an inverse association with an odds ratio <0.8. In these 14 studies, risk reduction ranged from 23% to 65%. Evidence for a dose–response relationship was found in 9 of 14 case–control studies that reported a decreased breast cancer risk for leisure time activity.39,41,46,48,50,57,58,64,66

Specific Life Periods

The association between physical activity before age 20 and female breast cancer risk was analyzed in 324,27,33 cohort studies and 1839,40,42–46,48–51,54,59,61,62,64,66,67 case–control studies (see Fig. 2). Approximately half of these studies (10 of 21) observed a reduction in breast cancer risk. Risk reductions ranged from 27% to 58%.

FIGURE 2.
FIGURE 2.:
Results of studies of leisure time physical activity before age 20 and breast cancer risk according to study design and study quality, as in Figure 1.

One cohort study and 13 case–control studies evaluated physical activity in more than one life period. Only 5 of these studies adjusted for activity in the other life periods or evaluated changes of activity patterns over time. The cohort study of Margolis et al33 found that only a change from being physically inactive to active from age 30 to enrollment was beneficial, whereas a consistently high activity level and activity at all other time points (age 14, age 30, and at enrollment) were not associated with a decreased breast cancer risk. The study of Steindorf et al43 did not find reduced risks of premenopausal breast cancer for women who were active only in adolescence or young adulthood nor for women who were active in both age periods. Verloop et al40 found that lifetime recreational activity was associated with a reduced risk, but could not identify a specific time period in which physical activity was associated with a stronger decrease in risk. They did not observe a beneficial effect of recent activity. Moradi et al44 found that recent leisure time activity was associated with a reduced risk of postmenopausal breast cancer in contrast to activity from age 18 to 30 and during childhood. Furthermore, they reported that women who switched from no activity when they were age 18 to 30 years to more than 2 hours per week in recent years also had a reduced risk. The study of Matthews et al48 observed a decreased risk for women who were active both in adolescence and adulthood compared with inactive women. Initiating exercise in adulthood was also associated with a reduction in risk, whereas no association was found for exercise in adolescence only. Three46,47,66 of the 926,46,47,50,54,61,62,66,67 studies that examined activity at older ages without adjusting for activity levels at younger ages (and vice versa) reported lack of an inverse association for physical activity during adolescence, whereas they did find an inverse association for physical activity at other ages.

The Effect of Body Mass Index

Eleven cohort and 8 case–control studies reported the association between physical activity and breast cancer risk according to strata of BMI (Tables 4 and 5). These studies did not consistently show differential effects of physical activity on breast cancer risk either in women with a low or high BMI or when the results for effect modification by BMI were analyzed according to menopausal status. One cohort study33 and 9 case–control studies43,47,50,51,53,57,62,64,65 evaluated effect modification of BMI based on statistical testing and did not find a difference.

TABLE 4
TABLE 4:
Results of Cohort Studies Assessing the Effect of Physical Activity on Pre- and Postmenopausal Breast Cancer in Strata of High and Low BMI (kg/m2)
TABLE 5
TABLE 5:
Results of Case–Control Studies Assessing the Effect of Physical Activity on Pre- and Postmenopausal Breast Cancer in Strata of High and Low BMI

Most studies in this review examined the potential confounding or mediating effect of BMI. The change in magnitude of the association between physical activity and breast cancer risk after adjustment for BMI was relatively small in these studies.

Effect Modification by Menopausal Status

In Figure 3, we present the results of the studies according to menopausal status. Seven cohort studies22–24,28,33,34,36 and 15 case–control studies39,41–43,48,53,54,56–59,61,64,66,67 assessed leisure time activity among premenopausal women. Two23,24 of the cohort studies observed a decreased breast cancer risk (RR <0.8). Eight39,41,42,48,57,58,64,66 of the 15 case–control studies observed a decrease in risk (RR <0.8). Among the 3 case–control studies that also assessed the effect of total activity, one53 reported a decreased risk and 243,59 reported no association.

FIGURE 3.
FIGURE 3.:
Results of studies on physical activity and breast cancer risk according to study design, source of activity, and study quality as in Figure 1, as well as menopausal status.

The majority of the studies that considered risk among postmenopausal women showed a reduced risk (Figs. 3C, D). Ten20,21,22,23,25–28,31,33 of the 1321,22,23,25–28,31–35,51 cohort studies examining the effect of leisure time activity observed a decrease in risk. One cohort study that assessed total activity found an increase in postmenopausal breast cancer risk.38 Ten39,41,44,48,54,57,58,61,63,66 of the 1639,41,42,44,47,48,53,54,56–59,61,63,66,67 case–control studies that assessed leisure time activities found a decrease in postmenopausal breast cancer risk. Total activity in relation to postmenopausal breast cancer risk was studied in 253,59 case–control studies. One59 of these studies found a risk reduction and one53 found no association.

Best-Evidence Synthesis

Table 6 gives a qualitative summary of the strength of evidence with regard to the association between physical activity and breast cancer risk. The higher-quality studies (quality score above the median) provided inconclusive evidence for the association with total activity and no evidence for leisure time activity. When we considered all studies without taking study quality into consideration, the evidence was indecisive for both sources of activity. Evidence for a dose–response relationship was observed in 55% and 50% of the higher quality studies that reported a decreased breast cancer risk for leisure or total activity, respectively.

TABLE 6
TABLE 6:
Best-Evidence Synthesis: A Summary of the Strength of Evidence for the Relation Between Physical Activity and Breast Cancer Risk

The summary of the strength of evidence regarding pre- and postmenopausal breast cancer risk concerns only leisure time activity, because the number of studies assessing the association between total activity and pre- or postmenopausal breast cancer risk was too small (and by definition, the evidence is inconclusive). The higher quality studies provided indecisive evidence for the association between leisure time activity and premenopausal breast cancer risk. For postmenopausal breast cancer risk, the higher quality cohort studies provided strong evidence. This was also the case when all studies (regardless of study quality) were considered.

Table 7 also shows the sensitivity analysis of the best-evidence synthesis. When we defined a decreased risk as a RR or odds ratio (OR) of <0.85 instead of <0.8, the evidence for an inverse association between leisure time activity and (overall) breast cancer risk changed from nil to strong, whereas the evidence for total activity changed from indecisive to moderate. The evidence for pre- and postmenopausal breast cancer risk did not change. A decreased risk, defined as a RR or OR of <0.75 instead of <0.8, changed the evidence for leisure time activity from indecisive to no evidence (based on higher quality studies only). Furthermore, the evidence for an association with postmenopausal breast cancer risk changed from strong to indecisive. When we focused on statistically significant results only, the evidence for an inverse association with total activity changed from indecisive to moderate. However, the evidence for pre- and postmenopausal breast cancer changed from indecisive to nil and from strong to nil, respectively. Exclusion of male breast cancer studies did not alter the conclusions.

TABLE 7
TABLE 7:
Best-Evidence Synthesis: Sensitivity Analysis of the Strength of Evidence for the Relation Between Physical Activity and Breast Cancer Risk

Trend or Dose–Response Analysis

Seventeen case–control studies39,40,42,44,46,47,49,50,52,53,56,59,61,62,64–66 that assessed physical activity (15 leisure time activity, 2 total activity) in hours per week were eligible for the trend analysis in which regression slopes (trends) across different physical activity levels for each individual study were estimated and finally pooled to one combined slope.

The quality score as percentage of the maximum score of these studies ranged from 32% to 80% (median 68%). The random effects analysis resulted in a combined slope (bactivity) of −0.056 (95% CI = −0.084 to −0.029), indicating a 6% decrease of breast cancer risk (all breast cancer) for each additional hour of physical activity per week. Compared with inactive persons, the OR for being physically active for 1 hour/week was 0.94 (95% CI = 0.92 to 0.97) and decreased, for example, to 0.76 for individuals being active 5 hours/week. Additional analysis revealed a nearly identical trend estimate when we focused exclusively on the higher quality case–control studies.

Exploring Heterogeneity in Study Results

We first used graphic analysis to examine whether methodologic study quality can explain heterogeneity in study results. Figures 4A and 4B show that there is no clear association between the quality score and the magnitude of the risk estimates in cohort or in case–control studies. Among the cohort studies, we observed a trend for studies with a favorable confounding score to show larger risk reductions (Fig. 4C). For case–control studies, there was a trend for publications with a favorable misclassification score to have odds ratios closer to unity (Fig. 4D).

FIGURE 4.
FIGURE 4.:
Relation between quality scores and risk estimates in studies of physical activity and breast cancer risk.

We next used metaregression analyses to explore the impact of study quality and study size on study outcome. The components of study quality (selection bias, misclassification bias, and confounding) and number of cases explained 58% of the heterogeneity in study outcome in cohort studies and 20% in case–control studies. In cohort studies, most of the heterogeneity was explained by the confounding score and the number of cases. A more favorable confounding score was associated with larger risk reductions. In case–control studies, heterogeneity was largely explained by the misclassification score. A more favorable misclassification score was associated with odds ratios closer to unity.

Funnel Plots

Figures 5A and 5B show funnel plots for cohort and case–control studies, respectively. The funnel plots show no clear evidence of publication bias or other forms of bias. Risk estimates of the larger studies cluster around 0.8 in both cohort and case–control studies.

FIGURE 5.
FIGURE 5.:
Funnel plots of studies.

DISCUSSION

The results of this systematic review are based on a rigorous and standardized analysis of the available literature on physical activity and breast cancer risk.

Our overall conclusions are based on the integration of cohort and case–control studies in the best-evidence synthesis, considering only results from the higher quality studies (quality score above the median). For postmenopausal breast cancer, we found strong evidence for risk reductions of at least 20% with leisure time activity (range, 20–80%). For pre- and postmenopausal breast cancer combined, the evidence for an association with leisure time or total activity was much weaker (nil and indecisive, respectively) using our a priori set criteria for defining levels of evidence. However, when we changed our criteria for a decreased risk to a risk estimate below 0.85 (instead of 0.80), the evidence was strong and moderate, respectively, indicating a 15% to 20% overall breast cancer risk reduction. For premenopausal breast cancer, the evidence for an association with physical activity was indecisive. Evidence for a dose–response relationship was observed in approximately half of the higher quality studies that reported a decreased risk.

In this review, we found no major differences between cohort and case–control studies in the proportion of studies that reported a decreased risk, which strengthens the evidence that physical activity reduces breast cancer risk. However, the magnitude of the risk reduction differed between the cohort and case–control studies of higher quality. The higher quality case–control studies showed a wide distribution in risk reductions ranging from 20% to 60%. The risk reductions in the higher quality cohort studies are smaller and show less variation (20–40%). A possible explanation for the larger risk reductions in case–control studies might be that physical activity is often more thoroughly assessed in case–control studies (less random misclassification) compared with cohort studies. Alternatively, the larger risk reductions in case–control studies may be explained by recall bias, which is supported by our metaregression analysis exploring the impact of the quality score. We found that case–control studies with a favorable misclassification score have odds ratios closer to unity.

Although our criteria for defining levels of evidence in our best-evidence synthesis are arbitrary, we believe they are reasonable to judge consistency of results. Because our criteria, defined a priori, have not been used before, we performed a sensitivity analysis (Table 7). The sensitivity analysis revealed that changing the definition of a relevant risk reduction led to different conclusions, especially for pre- and postmenopausal breast cancer combined. When we defined a decreased risk as a RR or OR of <0.85 instead of <0.8, the evidence for an inverse association between leisure time activity and overall breast cancer risk changed from nil to strong. Additionally, the evidence for total activity changed from indecisive to moderate (Table 7). These changes indicate that quite a large number of studies reported modest risk reductions of 15% to 20%. Furthermore, the best-evidence synthesis did not take study size into account. Therefore, we also evaluated the impact of counting only statistically significant risk reductions in the best-evidence synthesis. This analysis showed that the evidence for an inverse association became stronger for breast cancer overall but weaker for postmenopausal breast cancer. The disappearance of the evidence with postmenopausal breast cancer, when focusing on statistical significance results only, might be explained by the fact that these estimates were based on smaller numbers.

The trend analysis showed a statistically significant association between the amount of physical activity and breast cancer risk and estimated that breast cancer risk decreases by 6% (95% CI = 3% to 8%) for each additional hour physical activity per week. Although this analysis makes full use of the information in all exposure categories, it also has some limitations. The trend analysis was based only on case–control studies, which might be a problem if the case–control studies overestimate the risk reduction from physical activity. However, the trend estimate of 6% persists when focusing exclusively on the high-quality case–control studies, ie, studies that tend to show slightly weaker associations between physical activity and breast cancer risk. The trend analysis assumes a linear trend between breast cancer risk and physical activity when the adjusted ORs were regressed on the amount of physical activity.16 The true shape of the dose–response curve, however, might be nonlinear. The potential of a nonlinear relation, eg, a threshold effect or a J-curve, is an important issue in devising any public health recommendations. However, the assessment of the precise course of the association between the amount of physical activity and breast cancer risk is difficult to assess because the categories and types of measures used have shown large variations across the studies conducted so far.

We compared the highest versus lowest level of activity. This approach will not always provide the most pronounced risk estimate, because the association between physical activity and breast cancer risk might not be linear and the number of cases in the highest activity level in many studies was limited. Evaluating the results of the 3 studies with less than 10 cases in the highest activity level, we found one study60 that showed a risk estimate deviant from the risk estimate in the adjacent category. Exclusion of this study, however, did not change our conclusions. We also examined whether study outcome is related to the amount of contrast between the highest and lowest activity level, because insufficient contrast might explain why some studies fail to find an association. We found no clear association among contrast in actual activity levels, reference level of activity, and magnitude of the risk estimate.

We excluded studies with only occupational physical activities for several methodologic reasons (see “Methods”). A previous review by Gammon et al3 presented the results of 9 studies on occupational activity. Four of these 9 studies reported a decrease in breast cancer risk according to our definition (risk estimate <0.8). That review also highlighted several (expected) methodologic flaws of these studies, eg, potential misclassification of exposure status due to the use of job title as measure of physical activity and lack of control for other factors. In future studies, job activity should be measured with more detailed and comprehensive questionnaires. Moreover, it should be a component of total activity, because this parameter is etiologically most relevant.

Explicit attention should also be paid to household activities as part of total activity because this may be an important source of activity. In the studies of this review, household activity was not always part of total activity, which may have led to measurement error.

Because we included only published studies, the potential influence of publication bias on the conclusions should be discussed. Small sample studies that found no effect or a risk increase for physical activity might not have been published. In a quantitative meta-analysis, this might lead to biased results. In this review, the conclusions are not likely to be influenced by publication bias because the funnel plots were not asymmetric. Furthermore, our conclusions are mainly based on the higher quality studies. We assume that higher quality studies, in which much attention is paid to the design and measurement of physical activity, are likely to be published irrespective of the outcome.

We developed a quality scoring system specifically for this review, because we could not find a suitable scoring method for the subject at hand. Our quality scoring system comprised the main aspects of study design and assessment of physical activity. However, it is not a direct measure of validity or precision and may not have captured all methodologic aspects adequately. Furthermore, criteria for cutoff levels are arbitrary. Despite these drawbacks, we believe we have created a reasonable quality scoring system for this research area. Our scoring system empirically supports previous opinion68 that case–control studies may overestimate the true effect of physical activity due to recall bias.

Because it is not possible to establish a direct causal link between physical activity and breast cancer risk in observational studies, a next important question to address is whether physical activity actually protects against breast cancer. The mechanisms by which physical activity may protect against breast cancer are still unclear. Endogenous sex hormones play an important role in the development of breast cancer69 as is evident from the association with menstrual cycle and reproductive characteristics such as age at menarche, menopause, and first birth.70,71 Physical activity has been shown to influence certain menstrual characteristics (eg, later age at menarche, amenorrhea, anovulatory, and irregular cycles),72 body size (which affects estrogen exposure in postmenopausal women)10,73 and serum hormone levels.9 It is therefore plausible that physical activity may decrease breast cancer risk through hormone-related pathways, although other pathways such as effects on metabolic hormone levels (eg, insulin) and growth factors (eg, IGF) may also contribute.

CONCLUSION AND RECOMMENDATIONS

The conclusions and recommendations should be read in the light of the previously mentioned methodologic choices and concerns. This qualitative summary of observational studies shows strong evidence for an inverse association between physical activity and postmenopausal breast cancer with risk reductions ranging from 20% to 80%. For premenopausal breast cancer, however, the evidence is much weaker and judged by us as indecisive. For overall breast cancer, physical activity is associated with a modest (15–20%) decreased risk. Evidence for a dose–response relationship is observed in approximately half of the higher quality studies that reported a decrease in risk. The results of the trend analysis of 17 case–control studies suggest a 6% reduction in risk per hour physical activity per week assuming that the activity would be sustained over a longer period of time.

Being physically active as an adult appears to reduce breast cancer risk irrespective of the amounts of activity before age 20. The association between physical activity and breast cancer risk does not seem to be attributable to confounding factors such as BMI. Also, physical activity does not appear to exert its effect through a reduction of BMI, and the effect of physical activity does not seem to be modified by BMI.

Future studies are needed to assess which aspects of physical activity are most strongly related to the risk of breast cancer, how much physical activity is necessary in which period of life, and whether there is interaction with BMI (additive or multiplicative). Intervention studies should help to elucidate the causal pathway through which these effects occur.

ACKNOWLEDGMENTS

The authors thank Petra Peeters and Matti Rookus for their contribution to the development of the quality scoring system.

REFERENCES

1. Vainio H, Kaaks R, Bianchini F. Weight control and physical activity in cancer prevention: international evaluation of the evidence. Eur J Cancer Prev. 2002;11(suppl 2):S94–100.
2. Friedenreich CM. Physical activity and cancer prevention: from observational to intervention research. Cancer Epidemiol Biomarkers Prev. 2001;10:287–301.
3. Gammon MD, John EM, Britton JA. Recreational and occupational physical activities and risk of breast cancer. J Natl Cancer Inst. 1998;90:100–117.
4. Weight Control and Physical Activity. Lyon: IARC Press; 2002.
5. Muti P, Quattrin T, Grant BJ, et al. Fasting glucose is a risk factor for breast cancer: a prospective study. Cancer Epidemiol Biomarkers Prev. 2002;11:1361–1368.
6. Key TJ, Appleby PN, Reeves GK, et al. Body mass index, serum sex hormones, and breast cancer risk in postmenopausal women. J Natl Cancer Inst. 2003;95:1218–1226.
7. Shi R, Yu H, McLarty J, et al. IGF-I and breast cancer: a meta-analysis. Int J Cancer. 2004;111:418–423.
8. Bergstrom A, Pisani P, Tenet V, et al. Overweight as an avoidable cause of cancer in Europe. Int J Cancer. 2001;91:421–430.
9. McTiernan A, Tworoger SS, Ulrich CM, et al. Effect of exercise on serum estrogens in postmenopausal women: a 12-month randomized clinical trial. Cancer Res. 2004;64:2923–2928.
10. Irwin ML, Yasui Y, Ulrich CM, et al. Effect of exercise on total and intra-abdominal body fat in postmenopausal women: a randomized controlled trial. JAMA. 2003;289:323–330.
11. Friedenreich CM, Rohan TE. A review of physical activity and breast cancer. Epidemiology. 1995;6:311–317.
12. Greenland S. Meta-analysis. In: Rothman KJ, Greenland S, eds. Modern Epidemiology. Philadelphia: Lippincott-Raven Publishers; 1998:643–673.
13. Powell KE, Thompson PD, Caspersen CJ, et al. Physical activity and the incidence of coronary heart disease. Annu Rev Public Health. 1987;8:253–287.
14. Petitti DB. Statistical methods in meta-analysis. In: Petitti DB, ed. Meta-Analysis, Decision Analysis and Cost-Effectiveness Analysis. Methods for Quantitative Synthesis in Medicine. New York: Oxford University Press; 2000:94–118.
15. Petitti DB. Trend or ‘dose–response’ analysis. In: Petitti DB, ed. Meta-Analysis, Decision Analysis and Cost-Effectiveness Analysis. Methods for Quantitative Synthesis in Medicine. New York: Oxford University Press; 2000:127–132.
16. Greenland S, Longnecker MP. Methods for trend estimation from summarized dose–response data, with applications to meta-analysis. Am J Epidemiol. 1992;135:1301–1309.
17. Berlin JA, Longnecker MP, Greenland S. Meta-analysis of epidemiologic dose–response data. Epidemiology. 1993;4:218–228.
18. Greenland S. Quantitative methods in the review of epidemiologic literature. Epidemiol Rev. 1987;9:1–30.
19. Sterne JAC, Egger M, Davey Smith G. Investigating and dealing with publication and other biases. In: Egger M, Davey Smith G, Altman DG, eds. Systematic Reviews in Health Care. Meta-Analysis in Context. London: BMJ Publishing Group; 2001:189–208.
20. Lee IM, Rexrode KM, Cook NR, et al. Physical activity and breast cancer risk: the Women's Health Study (United States). Cancer Causes Control. 2001;12:137–145.
21. Cerhan JR. Physical activity, physical function, and the risk of breast cancer in a prospective study among elderly women. J Gerontol A Biol Sci Med Sci. 1998;53:M251–M256.
22. Sesso HD, Paffenbarger RS Jr, Lee IM. Physical activity and breast cancer risk in the College Alumni Health Study (United States). Cancer Causes Control. 1998;9:433–439.
23. Thune I, Brenn T, Lund E, et al. Physical activity and the risk of breast cancer. N Engl J Med. 1997;336:1269–1275.
24. Wyshak G, Frisch RE. Breast cancer among former college athletes compared to non-athletes: a 15-year follow-up. Br J Cancer. 2000;82:726–730.
25. Wyrwich KW, Wolinsky FD. Physical activity, disability, and the risk of hospitalization for breast cancer among older women. J Gerontol A Biol Sci Med Sci. 2000;55:M418–M421.
26. Patel AV, Callel EE, Bernstein L, et al. Recreational physical activity and risk of postmenopausal breast cancer in a large cohort of US women. Cancer Causes Control. 2003;14:519–529.
27. McTiernan A, Kooperberg C, White E, et al. Recreational physical activity and the risk of breast cancer in postmenopausal women: the Women's Health Initiative Cohort Study. JAMA. 2003;290:1331–1336.
28. Breslow RA, Ballard-Barbash R, Munoz K, et al. Long-term recreational physical activity and breast cancer in the National Health and Nutrition Examination Survey I epidemiologic follow-up study. Cancer Epidemiol Biomarkers Prev. 2001;10:805–808.
29. Tehard B, Friedenreich CM, Oppert JM, et al. Effect of physical activity on women at increased risk of breast cancer: results from the E3N Cohort Study. Cancer Epidemiol Biomarkers Prev. 2006;15:57–64.
30. Rockhill B, Willett WC, Hunter DJ, et al. A prospective study of recreational physical activity and breast cancer risk. Arch Intern Med. 1999;159:2290–2296.
31. Moradi T, Adami HO, Ekbom A, et al. Physical activity and risk for breast cancer a prospective cohort study among Swedish twins. Int J Cancer. 2002;100:76–81.
32. Moore DB, Folsom AR, Mink PJ, et al. Physical activity and incidence of postmenopausal breast cancer. Epidemiology. 2000;11:292–296.
33. Margolis KL, Mucci L, Braaten T, et al. Physical activity in different periods of life and the risk of breast cancer: the Norwegian–Swedish Women's Lifestyle and Health cohort study. Cancer Epidemiol Biomarkers Prev. 2005;14:27–32.
34. Luoto R, Latikka P, Pukkala E, et al. The effect of physical activity on breast cancer risk: a cohort study of 30,548 women. Eur J Epidemiol. 2000;16:973–980.
35. Dirx MJ, Voorrips LE, Goldbohm RA, et al. Baseline recreational physical activity, history of sports participation, and postmenopausal breast carcinoma risk in the Netherlands Cohort Study. Cancer. 2001;92:1638–1649.
36. Colditz GA, Feskanich D, Chen WY, et al. Physical activity and risk of breast cancer in premenopausal women. Br J Cancer. 2003;89:847–851.
37. Fraser GE, Shavlik D. Risk factors, lifetime risk, and age at onset of breast cancer. Ann Epidemiol. 1997;7:375–382.
38. Dorgan JF, Brown C, Barrett M, et al. Physical activity and risk of breast cancer in the Framingham Heart Study. Am J Epidemiol. 1994;139:662–669.
39. Yang D, Bernstein L, Wu AH. Physical activity and breast cancer risk among Asian-American women in Los Angeles: a case–control study. Cancer. 2003;97:2565–2575.
40. Verloop J, Rookus MA, van der Kooy K, et al. Physical activity and breast cancer risk in women aged 20–54 years. J Natl Cancer Inst. 2000;92:128–135.
41. Ueji M, Ueno E, Osei-Hyiaman D, et al. Physical activity and the risk of breast cancer: a case–control study of Japanese women. J Epidemiol. 1998;8:116–122.
42. Taioli E, Barone J, Wynder EL. A case–control study on breast cancer and body mass. The American Health Foundation—Division of Epidemiology. Eur J Cancer. 1995;31A:723–728.
43. Steindorf K, Schmidt M, Kropp S, et al. Case–control study of physical activity and breast cancer risk among premenopausal women in Germany. Am J Epidemiol. 2003;157:121–130.
44. Moradi T, Nyren O, Zack M, et al. Breast cancer risk and lifetime leisure-time and occupational physical activity (Sweden). Cancer Causes Control. 2000;11:523–531.
45. Mittendorf R, Longnecker MP, Newcomb PA, et al. Strenuous physical activity in young adulthood and risk of breast cancer (United States). Cancer Causes Control. 1995;6:347–353.
46. D'Avanzo B, Nanni O, La Vecchia C, et al. Physical activity and breast cancer risk. Cancer Epidemiol Biomarkers Prev. 1996;5:155–160.
47. McTiernan A, Stanford JL, Weiss NS, et al. Occurrence of breast cancer in relation to recreational exercise in women age 50–64 years. Epidemiology. 1996;7:598–604.
48. Matthews CE, Shu XO, Jin F, et al. Lifetime physical activity and breast cancer risk in the Shanghai Breast Cancer Study. Br J Cancer. 2001;84:994–1001.
49. Marcus PM, Newman B, Moorman PG, et al. Physical activity at age 12 and adult breast cancer risk (United States). Cancer Causes Control. 1999;10:293–302.
50. Levi F, Pasche C, Lucchini F, et al. Occupational and leisure time physical activity and the risk of breast cancer. Eur J Cancer. 1999;35:775–778.
51. Lee IM, Cook NR, Rexrode KM, et al. Lifetime physical activity and risk of breast cancer. Br J Cancer. 2001;85:962–965.
52. Johnson KC, Pan S, Mao Y. Risk factors for male breast cancer in Canada, 1994–1998. Eur J Cancer Prev. 2002;11:253–263.
53. John EM, Horn-Ross PL, Koo J. Lifetime physical activity and breast cancer risk in a multiethnic population: the San Francisco Bay area breast cancer study. Cancer Epidemiol Biomarkers Prev. 2003;12:1143–1152.
54. Hu YH, Nagata C, Shimizu H, et al. Association of body mass index, physical activity, and reproductive histories with breast cancer: a case–control study in Gifu, Japan. Breast Cancer Res Treat. 1997;43:65–72.
55. Hsing AW, McLaughlin JK, Cocco P, et al. Risk factors for male breast cancer (United States). Cancer Causes Control. 1998;9:269–275.
56. Hirose K, Hamajima N, Takezaki T, et al. Physical exercise reduces risk of breast cancer in Japanese women. Cancer Sci. 2003;94:193–199.
57. Gilliland FD, Li YF, Baumgartner K, et al. Physical activity and breast cancer risk in Hispanic and non-Hispanic white women. Am J Epidemiol. 2001;154:442–450.
58. Friedenreich CM, Rohan TE. Physical activity and risk of breast cancer. Eur J Cancer Prev. 1995;4:145–151.
59. Friedenreich CM, Courneya KS, Bryant HE. Influence of physical activity in different age and life periods on the risk of breast cancer. Epidemiology. 2001;12:604–612.
60. Ewertz M, Holmberg L, Tretli S, et al. Risk factors for male breast cancer—a case–control study from Scandinavia. Acta Oncol. 2001;40:467–471.
61. Dorn J, Vena J, Brasure J, et al. Lifetime physical activity and breast cancer risk in pre- and postmenopausal women. Med Sci Sports Exerc. 2003;35:278–285.
62. Chen CL, White E, Malone KE, et al. Leisure-time physical activity in relation to breast cancer among young women (Washington, United States). Cancer Causes Control. 1997;8:77–84.
63. Carpenter CL, Ross RK, Paganini-Hill A, et al. Effect of family history, obesity and exercise on breast cancer risk among postmenopausal women. Int J Cancer. 2003;106:96–102.
64. Bernstein L, Henderson BE, Hanisch R, et al. Physical exercise and reduced risk of breast cancer in young women. J Natl Cancer Inst. 1994;86:1403–1408.
65. Bernstein L, Patel AV, Ursin G, et al. Lifetime recreational exercise activity and breast cancer risk among black women and white women. J Natl Cancer Inst. 2005;97:1671–1679.
66. Adams-Campbell LL, Rosenberg L, Rao RS, et al. Strenuous physical activity and breast cancer risk in African-American women. J Natl Med Assoc. 2001;93:267–275.
67. Gammon MD, Schoenberg JB, Britton JA, et al. Recreational physical activity and breast cancer risk among women under age 45 years. Am J Epidemiol. 1998;147:273–280.
68. Rothman KJ. Epidemiology. An Introduction, 1st ed. Oxford: Oxford University Press; 2002.
69. Key T, Appleby P, Barnes I, et al. Endogenous sex hormones and breast cancer in postmenopausal women: reanalysis of nine prospective studies. J Natl Cancer Inst. 2002;94:606–616.
70. Hsieh CC, Trichopoulos D, Katsouyanni K, et al. Age at menarche, age at menopause, height and obesity as risk factors for breast cancer: associations and interactions in an international case–control study. Int J Cancer. 1990;46:796–800.
71. Ewertz M, Duffy SW, Adami HO, et al. Age at first birth, parity and risk of breast cancer: a meta-analysis of 8 studies from the Nordic countries. Int J Cancer. 1990;46:597–603.
72. Bernstein L, Ross RK, Lobo RA, et al. The effects of moderate physical activity on menstrual cycle patterns in adolescence: implications for breast cancer prevention. Br J Cancer. 1987;55:681–685.
73. Astrup A. Physical activity and weight gain and fat distribution changes with menopause: current evidence and research issues. Med Sci Sports Exerc. 1999;31:S564–S567.
74. Friedenreich CM, Courneya KS, Bryant HE. Relation between intensity of physical activity and breast cancer risk reduction. Med Sci Sport Exerc. 2001;33:1538–1545.
75. Friedenreich CM, Bryant HE, Courneya KS. Case–control study of lifetime physical activity and breast cancer risk. Am J Epidemiol. 2001;154:336–347.
© 2007 Lippincott Williams & Wilkins, Inc.