We observed stronger effects for PA among postmenopausal women compared to premenopausal women, which is consistent with previously published data2–5
. The effects obtained using average MET-hours as a composite measure of intensity and duration were similar to estimates observed for duration alone. Our results did indicate some variations in risk based on timing of RPA occurrence. We found that RPA over the life-course, particularly during the reproductive (among parous women) and postmenopausal years, decreases breast cancer risk. The observed inverse associations are consistent with most other studies that have examined the effect of PA on breast cancer risk reduction4,5,19,20
where an average 25% risk reduction is reported5,19
Results from the current study show that RPA during the reproductive years and after menopause are most critical for risk reduction. Our observations likely reflect the role of PA in energy balance and obesity mediated mechanisms (e.g. insulin resistance and inflammation) which most commonly manifest after adolescence. We found little evidence of benefit from early-life activity which is consistent with most 21–24
, but not all25,26
, investigations. While few studies are able to assess activity across the life-course, those that have a comprehensive lifetime assessment of PA report inverse associations2,27
. Using a similar PA assessment to the current study, Bernstein and colleagues report a 17% risk reduction among women in the highest quartile of lifetime activity compared inactive women27
. We also found a 17% risk reduction for lifetime RPA among the most active women despite our wide age distribution (age 20–98 vs. age 35–64). Given the noticeable difference in age distribution for this study, we re-assed our main effects restricted to women age <80, <70, and <60 years, respectively. We found no difference between these restricted analyses and the analyses among women of all ages (data not shown).
Contrary to a recent review which reports an inverse dose-response association between PA and breast cancer risk5
, the observed effect for RPA did not decrease in a dose-response manner. A lack of linear dose-response could be interpreted as weak evidence of an association between RPA and breast cancer risk. However, it is possible that the RPA breast cancer association may follow a J- or U-shape. In the LIBCSP, a substantial proportion of women reported high levels of RPA permitting us to consider a wide range of effects. Studies indicate that sustained PA is a strong inducer of lipid peroxidation and reactive oxygen species28,29
. These changes may cause DNA damage, mutations in proto-oncogenes or tumor-suppressor genes and, if unrepaired, transformation of normal epithelium to a malignant phenotype30,31
. Studies also show that vigorous physical exercise may depress immune function32
. Inconsistencies in dose-response may therefore reflect the underlying distribution of RPA among study participants. For example, the median lifetime RPA in the Women’s Contraceptive and Reproductive Experiences study was 1.2 hrs/week27
, while the median lifetime RPA was 6.35 hrs/week in the current study. Our results, in combination with animal and clinical data, suggest that sustained involvement in vigorous activity may mitigate the known protective effect of RPA resulting in a J- or U-shaped association.
Although our results do not show statistically significant associations among HR-positive or HR-negative cases, they do suggest that RPA during the reproductive period may preferentially decrease HR-positive tumors. This further supports the hypothesis that breast cancer is a heterogeneous disease with varying etiologic pathways33
. When we assessed ER-negative cases compared to ER-positive cases the patterns of association were not markedly different between groups (OR=0.80 and 0.85, respectively), suggesting that the role of PA in risk reduction is not entirely mediated through an estrogen pathway. It should be noted that in these analyses the cell sizes were particularly small for ER−/PR− cases in comparison with any positive cases. Our findings are comparable to other reports that find no difference in the PA-breast cancer association by hormone receptor status27,34,35
, but in contrast to some studies which report greater decreases among ER- cases compared to ER+ cases35,36
, and still others which show stronger associations for ER+ tumors37,38
In this large case-control study we found that breast cancer risk was generally the greatest among women jointly classified as having high levels of adiposity and little RPA. While the results were consistent with our hypothesis, we found only one multiplicative interaction and no evidence of additive interaction among our indicators of energy balance. It is noteworthy that postmenopausal RPA reduced the adverse effects of obesity on breast cancer risk to approximately null, but did not completely obliterate the effect of postmenopausal weight gain. These observations likely reflect differences in the effect of weight maintenance vs. weight gain39
with the latter being potentially more deleterious during the postmenopausal years11
. The timing of weight gain may therefore be an important factor in understanding the weight-PA interaction among postmenopausal populations. While stratification would help us better uncover these associations, even with a study sample of 3000 women we did not have adequate statistical power to evaluate the three-way joint effects of weight gain, BMI and RPA. Few studies have examined modification by weight change40–44
. In the only other study42
to assess the joint effects of weight gain and PA using the common referent analysis, investigators also reported that high PA did not eliminate the excess breast cancer risk caused by weight gain.
Studies of PA and breast cancer risk have mixed results on the modifying effects of BMI. Several investigations have found risk reductions only among physically active lean women43–48
, while others report risk reductions in all BMI categories42,49
. A 2008 review of 16 studies estimated that risk reductions were approximately 25% among women with BMI between 22 and 25 kg/m2 and 20% among women with BMI ≥25 kg/m2. There were near null effects of PA on breast cancer risk among women with BMI ≥30 kg/m2, although few studies reported effects in this stratum5
. Our results are consistent with this review.
The strengths of our study are numerous and include its population-based design and large sample size which increased our power to detect small associations, assess subgroup analyses, and evaluate joint effects of weight indicators and RPA. Our RPA assessment provided a wide range of activities that contribute to energy expenditure in this population of women. Multiple time periods throughout the lifespan were evaluated as well as several parameters of RPA. While in this analysis we were unable to assess all potential sources of PA, a comprehensive 2008 review of physical activity parameters and breast cancer risk showed that the greatest risk reductions were for RPA (20% risk reduction). Activity related to occupation, transportation and living each resulted in ~14% risk reduction5
. Few studies have considered PA from all sources. Given the high socioeconomic status and of Long Island women8
we expected little variation by alternative sources of activity and anticipate that any additional variation would result in a more pronounced risk reduction. Although our RPA measurement has not been validated, this instrument was useful in revealing important relationships between exercise and breast cancer risk in other epidemiologic studies9,50
Despite the large overall sample size a limitation of this study was its relatively homogenous population. Study participants were on average more affluent and educated than the US population. Results are therefore not readily applicable to all women. Our study population included few women who were nulliparous8
. We were therefore unable to perform a stratified analysis by nulliparity when assessing the effect of timing on breast cancer risk. Errors in reporting or differential reporting by cases and controls have the potential to bias the study results. A spurious inverse association could have occurred if PA was systematically underreported by cases or over reported by controls. While regular PA, like other healthy behaviors, may result in some social desirability, we suspect that these biases would persist in both case and control groups resulting in non-differential misclassification of RPA. Given the exposure variable was not simply dichotomous however, the direction of such a bias would be unpredictable51
. To reduce the probability of recall bias and misclassification, Long Island investigators used a comprehensive questionnaire to obtain detailed information on most study variables enhancing our ability to assess RPA and control for confounding by important breast cancer risk factors.