|Home | About | Journals | Submit | Contact Us | Français|
Recreational physical activity has been both positively and inversely associated with cancer risk for postmenopausal women, acting presumably through hormonal mechanisms. Relatively little is known about the effects of exercise on postmenopausal steroid hormone levels. The authors evaluated the association between recreational activity and plasma steroid hormones among 623 US healthy, postmenopausal women in the Nurses’ Health Study not using exogenous hormones at the time of blood draw (1989–1990). Participants self-reported recreational physical activity by questionnaire in 1986, 1988, and 1992. Plasma samples were assayed for estrogens, androgens, and sex hormone-binding globulin. Geometric mean hormone levels adjusted and not adjusted for body mass index were calculated. In general, estrogen and androgen levels were lower in the most- and the least-active women compared with those reporting moderate activity, suggesting a U-shaped relation. For example, estrone sulfate levels in quintiles 1–5 of metabolic equivalent task-hours were 197, 209, 222, 214, and 195 pg/mL, respectively. Tests for nonlinearity using polynomial regression were significant for several estrogens, androgens, and sex hormone-binding globulin (2-sided P ≤ 0.01). These results suggest the possibility of a nonlinear relation between recreational physical activity and hormone levels in postmenopausal women.
Recreational physical activity has been associated with reduced risk of postmenopausal breast (1–6) and endometrial (7–9) cancers. For ovarian cancer, results have been inconsistent (10), with some investigations observing a modest protective association (11–17) and others suggesting that physically active women may have a higher risk than less-active women (18–20). For example, a recent, large prospective study observed a significant 2-fold increase in ovarian cancer incidence in women reporting more than 4 hours per week of vigorous activity compared with those rarely engaging in vigorous activity (relative risk=2.38, 95% confidence interval: 1.29, 4.38; P for trend < 0.01) (18).
Physical activity may influence risk of various cancers in part by altering sex steroid hormone levels (21), which have been positively associated with risk in some studies (22–25). However, relatively few studies have directly assessed the relation between physical activity and postmenopausal hormone levels (26–35). After menopause, circulating estrogens are almost exclusively produced through extraglandular aromatization of androgens, primarily androstenedione. Exercise may reduce estrogen levels by decreasing body fat and thereby reducing the conversion of androgens to estrogens. Levels of androgens, which are still produced by the ovaries in postmenopausal women, also may be affected (29). The extent to which body mass modifies this relation after menopause is unclear, although results from an exercise intervention trial suggest that the effect of exercise on sex steroid levels may be influenced by loss of body fat (28, 29). To address these questions, we assessed the association between participation in recreational physical activity and steroid hormone levels in a cross-sectional study of postmenopausal women, with special attention paid to the potential modifying effect of body mass.
The Nurses’ Health Study is a cohort of 121,700 US female registered nurses who responded to a mailed questionnaire in 1976. Participants were 30–55 years of age at the time of the initial mailing and provided information on their medical history and health-related behaviors, such as use of oral contraceptives and smoking status. Cohort members have completed questionnaires every 2 years thereafter to update information on various risk factors and identify new diagnoses of cancer and other diseases. The protocol for this study was approved by the Human Research Committees at Brigham and Women's Hospital, Boston, Massachusetts.
In 1989–1990, we asked participants to provide a blood sample; samples were ultimately received from 32,826 women aged 43–69 years. Collection has been described in detail previously (36). Briefly, we provided participants with a kit containing blood-collection supplies and asked them to have their blood drawn and returned to us via overnight courier. Ice packs were supplied to keep samples cool during mailing. Participants completed a short questionnaire to record the day and time of collection, current body weight, and use of medications including hormone replacement therapy. Ninety-seven percent of samples were received by our laboratory within 26 hours of draw. Upon receipt, samples were centrifuged; were separated into plasma, red cell, and white cell components; and were archived at −130°C or colder in continuously monitored liquid nitrogen freezers.
Participants included in the present analysis were women chosen as controls for a nested case-control study of steroid hormone levels and breast cancer (22, 37). One to 2 controls were matched to each breast cancer case on the basis of age; menopausal status; use of postmenopausal hormones; and month, time of day, and fasting status at blood collection. Women eligible for the present analysis included all controls who were postmenopausal at blood collection (i.e., had reported no menstrual periods for at least 1 year), had not used postmenopausal hormones for at least 3 months prior to collection, and had not received a diagnosis of cancer other than nonmelanoma skin cancer as of May 31, 1998.
On Nurses’ Health Study questionnaires in 1986 and 1988, participants were asked to report the average amount of time each week they engaged in specific activities during the past year. These activities included jogging (i.e., slower than 10 minutes per mile (1 mile=1.6 km)), running (i.e., 10 minutes per mile or faster), bicycling (including on a stationary machine), lap swimming, tennis, squash or racquet ball, aerobic activities (including calisthenics, aerobic dance, and rowing on a machine), and walking or hiking outdoors (including walking while playing golf). In addition, participants were asked about their usual walking pace and the number of flights of stairs they climbed daily. Similar questions were included on the 1992 Nurses’ Health Study questionnaire, along with new questions on low-intensity activities (yoga, stretching, toning) and other vigorous activities (lawn mowing). These questions have been validated in a similar population and are described in detail elsewhere (38). Briefly, in a sample of 153 women, the correlation was 0.79 (95% confidence interval: 0.64, 0.88) between physical activity scores calculated by physical activity questionnaire and a series of 4 past-week activity recalls completed over the previous year.
We evaluated the total amount of time per week participants reported engaging in physical activity. In addition, we calculated the total number of metabolic equivalent (MET)-hours of activity for each participant to account for the frequency, duration, and intensity of physical activity (39). To accomplish this goal, we multiplied the amount of time spent on each activity by its typical energy expenditure requirement in METs to calculate MET-hours for that activity. We then summed the contributions of each activity and divided participants into quintiles and deciles.
We evaluated other factors that could potentially be associated with physical activity and/or steroid hormone levels. We assessed fasting status and time of day of blood collection, as well as body mass index (BMI, calculated as weight in kilograms/height in meters2), using the questionnaire completed at blood draw. We assessed other factors at approximately the time of blood draw using information provided on previous Nurses’ Health Study questionnaires, including age, age at menarche, parity and age at first birth, weight at age 18 years, age at menopause, history of tubal ligation and oral contraceptive use, and smoking status. We used a semiquantitative food frequency questionnaire administered as part of the main study questionnaire in 1990 to assess total energy and alcohol intake.
All plasma steroid hormones were analyzed in 4 batches. All batches of estrone, estradiol, androstenedione, testosterone, dehydroepiandosterone (DHEA), and dehydroepiandosterone sulfate (DHEAS), along with the last 3 batches of estrone sulfate, were assayed at Quest Diagnostic's Nichols Institute (San Juan Capistrano, California). The first batch of estrone sulfate and first 2 batches of sex hormone-binding globulin (SHBG) were assayed at the Longcope Steroid Radioimmunoassay Laboratory at the University of Massachusetts Medical Center (Worcester, Massachusetts). The remaining batches of SHBG were assayed by the Massachusetts General Hospital's Reproductive Endocrinology Unit Laboratory (Boston, Massachusetts).
Methods used for hormone assays have been described in detail previously (22, 37). Briefly, samples were extracted with a mixture of hexane and ethyl acetate (4:1) and were applied to a celite column. Steroids were then eluted from the column (celite in ethylene glycol), and the fractions were analyzed by radioimmunoassay (40–44). DHEAS was assayed by radioimmunoassay without a prior separation step (45). Estrone sulfate was first extracted from plasma with an organic solvent and was then analyzed by chromatography and radioimmunoassay (46). Levels of free estradiol and bioavailable estradiol (i.e., estradiol that is either free or loosely bound to albumin) were directly measured in a subset of samples. SHBG was measured with an immunoradiometric kit from FARMOS Diagnostica (Orion Corp., Turku, Finland) in the Longcope laboratory and with the AxSYM Immunoassay system (Abbott Diagnostics, Chicago, Illinois) in the Massachusetts General Hospital Reproductive Endocrinology Unit laboratory.
To allow for assessment of batch-to-batch variation and laboratory precision, we randomly distributed blinded quality control samples (equivalent to 10% of replicates) along with all analytic samples. Mean within-batch coefficients of variation ranged from 6% for DHEAS to 13.6% for DHEA (22, 37).
All statistical analyses were conducted by using SAS software (SAS Institute, Inc., Cary, North Carolina). We evaluated the distribution of each steroid hormone and identified outliers, defined as women whose levels were greater than the absolute value of the 75th percentile plus 3 times the interquartile range; between 2 (SHBG and androstenedione) and 14 (estrone sulfate) women were excluded from analyses of each hormone.
For each type of physical activity we assessed, for our primary analysis we averaged each woman's level of participation in 1986 and 1988 to provide a more stable assessment of activity participation in the 1–4 years prior to blood collection. We estimated geometric least squares mean hormone levels across categories of average of physical activity level by regressing the natural log of hormone levels on potential confounders, then adding the mean log hormone levels to the average of the residuals and exponentiating this value. The final results were the predicted hormonal levels at the reference value for the confounders. All P values presented are 2-sided.
To assess the influence of body mass on the activity-hormone relation, we calculated geometric mean hormone levels both adjusting and not adjusting for BMI at the time of blood collection. We evaluated linear trends across categories using multivariable linear regression and Wald tests. Because preliminary results suggested the possibility of a U-shaped relation between physical activity and hormone levels, we conducted a posteriori evaluation of nonlinearity using polynomial regression. We first regressed continuous natural log (ln) MET-hours per week of physical activity on ln hormone levels. We then added a second-order term (i.e., the square of ln MET-hours of activity) to the model and performed an F test to ascertain whether inclusion of this term explained more of the variation in ln hormone levels than expected because of chance alone.
In addition, we conducted subanalyses stratified by BMI at the time of blood collection (<23, 23–<27, ≥27 kg/m2). Our decision to use nonstandard categories of BMI was based largely on sample size and power considerations; very few participants had a BMI ≥30 kg/m2, and parameter estimates using this category cutpoint were unstable. We also stratified by weight change between 1988 and date of blood draw (weight loss, no change/weight gain), with weight change measured by subtracting weight reported in 1988 from weight reported on the questionnaire administered at blood collection. Analyses were repeated by using physical activity as assessed in 1988 only and by the average of activity in 1988 and 1992. Furthermore, we assessed the effect of average activity in 1988–1992 after stratifying by change in physical activity between 1988 and 1992 (increase of ≥1 quintile of METs in 1992, no change, decreased by ≥1 quintile). Finally, we evaluated these associations among women who had not reported an oophorectomy prior to blood draw.
We limited our main analysis to the 623 women who had both complete physical activity assessments in 1986 and 1988 and available plasma samples. Characteristics of study members in quintiles 1, 3, and 5 of MET-hours per week of activity are presented in Table 1. Distribution of most characteristics varied by level of physical activity. The mean age of study participants was approximately 62 years (range, 46–69). Mean BMI at the time of blood draw ranged from 24.8 kg/m2 for women reporting the highest level of physical activity to 27.4 kg/m2 for those reporting the lowest. Current smoking was reported by 8.6% of participants in quintile 5 compared with 20.2% of women in quintile 1.
The number of plasma samples available for analysis varied between hormones, ranging from 417 for bioavailable estradiol to 607 for DHEAS (Table 2). Measured hormone levels were consistent with those expected for postmenopausal women not using hormone therapy.
Geometric mean hormone and SHBG levels by hours per week of physical activity are presented in Table 3. Before adjustment for BMI at blood collection, estradiol, bioavailable estradiol, and free estradiol appeared to be inversely related to physical activity levels (P=0.002–0.02). In contrast, SHBG was positively related to activity level (P= 0.002). Of the androgens evaluated, we did not observe substantial evidence of a linear relation with activity.
We observed strong correlations between BMI and estrogen and SHBG levels. Pearson correlation coefficients ranged from 0.32 (P < 0.0001) for estrone sulfate to 0.59 (P < 0.0001) for bioavailable estradiol. SHBG was inversely related to BMI (r =−0.38; P < 0.0001). In contrast, androgen levels were not significantly correlated with BMI (range of r, −0.006 to 0.03). Physical activity was significantly inversely correlated with BMI, with the strongest correlations observed for activity measured as MET-hours per week (r=−0.17; P < 0.0001). After further adjustment for BMI, estradiol and SHBG levels were no longer linearly associated with hours per week of activity (Table 3). The relation between androgen levels and physical activity was largely unchanged after adjustment for BMI.
In analyses not adjusting for BMI, we observed some evidence of a linear relation between hormone levels and MET-hours per week of activity (Table 4), although adjustment for BMI attenuated these associations. For all of the hormones evaluated, levels tended to be lowest in women reporting the highest level of activity (i.e., quintile 5) compared with those reporting moderate levels (i.e., quintiles 2–4). For example, for estradiol, mean levels over increasing quintiles were 6.4, 6.9, 6.4, 6.8, and 6.1 pg/mL, respectively. To assess nonlinearity, we fit linear and polynomial regression models and assessed model fit (Table 5). F-test results indicated that model fit significantly improved with the addition of a second-order term to the base linear regression model for estradiol, free estradiol, bioavailable estradiol, estrone, DHEA, androstenedione, and SHBG (range of P=0.0002 for estradiol to P=0.007 for free estradiol).
We stratified our population by BMI (<23, 23−<27, ≥27 kg/m2) at the time of blood collection. Median MET-hours per week in quintile 1 and quintile 5 were relatively similar across the 3 levels of BMI (BMI <23: median of quintile 1=1.7 MET-hours per week, quintile 5=35.2; BMI 23–<27: quintile 1=1.5, quintile 5=34.1; and BMI ≥27: quintile 1=1.7, quintile 5=29.9). In leaner women (BMI <23 kg/m2), estradiol, estrone, and estrone sulfate levels were lowest in those in quintile 1 and quintile 5 of MET-hours per week compared with those in the middle quintiles (results not shown). For example, estradiol levels across the 5 quintiles were 4.7, 6.1, 5.2, 5.6, and 4.5 pg/mL, respectively. Androgen levels also demonstrated a similar pattern in leaner women; for example, androstenedione levels across quintiles 1–5 were 66, 100, 75, 70, and 56 ng/dL, respectively. We did not observe consistent associations between physical activity and steroid hormone levels in moderate-weight and heavier women. Results also tended to be stronger for women reporting any weight loss between 1988 and date of blood draw than for those reporting no change in weight or weight gain. We did not find change in physical activity level between 1988 and 1992 to be related to hormone levels (results not shown).
Results from secondary analyses using physical activity levels reported in 1988 only, average of 1988 and 1992, average activity in 1988–1992 stratified by change in activity, and analyses restricted to women without prior oophorectomy were similar to those from the main analysis (results not shown). Results evaluating hormone levels over deciles of physical activity were also similar (results not shown).
Although we observed some evidence of a linear association of estrogen and SHBG levels with recreational physical activity in postmenopausal women, for several hormones, the relation with activity appeared nonlinear. Levels of several hormones were 10%–15% lower in the least and most active women than in those reporting moderate activity. This pattern was reversed for SHBG: levels were higher in women in the extreme quintiles of activity compared with those in the middle categories. For estradiol, estrone, DHEA, androstenedione, and SHBG, results from regression analyses indicated that adding a second-order term significantly improved model fit.
A limited number of studies have assessed these associations, with inconsistent results. In several studies, levels of estrone, estradiol, testosterone, and/or androstenedione have been inversely associated with physical activity (26, 27, 32, 33, 35). However, in one study, estradiol levels were significantly higher in women regularly engaging in vigorous activity compared with less-active women (30). Wu et al. (31) observed a significant inverse correlation between physical activity and SHBG levels (r=−0.24; P < 0.05), while others have found either no relation (30) or a positive association (26, 35).
In a randomized trial, McTiernan et al. (28) observed greater decreases in estrone and free estradiol levels in women assigned to a physical activity intervention than in controls and modest increases in SHBG levels. In the intervention group, estrone, estradiol, and free estradiol decreased by a greater proportion in women losing body fat than in those not changing or gaining body fat during the intervention. Androgen levels did not differ significantly between groups (29), but they decreased by a greater proportion in exercisers losing body fat. These findings underscore the importance of evaluating these associations in the context of BMI and/or weight change.
Most recently, van Gils et al. (35) evaluated sex steroid hormone levels adjusted and nonadjusted for BMI. As in our study, unadjusted results suggested a linear inverse relation between activity and estradiol and positive associations with DHEAS and SHBG. However, after adjustment for BMI, linear associations were attenuated, and, for several hormones—including estrone, estradiol, and testosterone—mean levels were highest in women reporting moderate activity compared with both active and inactive women. Potential U-shaped associations between hormones and activity have been suggested in other studies (32, 33), but, to our knowledge, no other studies have formally tested for nonlinearity.
Our results suggest that many postmenopausal women who engage in light-moderate recreational activity may in fact have higher estrogen and perhaps androgen levels than sedentary women. It is also possible that relatively few older women exercise frequently or intensely enough to lower hormone levels below those of sedentary women. If higher estrogen and androgen levels do indeed increase the risk of ovarian cancer, this factor may help explain results from recent analyses suggesting that frequent vigorous activity in postmenopausal women may increase risk (18–20).
For example, in our population, ovarian cancer risk was somewhat higher for women reporting moderate-vigorous activity compared with sedentary women (19). Similar results were observed in the Iowa Women's Health Study (18, 20). In other studies (11–13, 15, 16), a lower risk of ovarian cancer was associated with only the highest levels of activity. In studies with small numbers of cases, such as many ovarian cancer studies, the range of activity that can be assessed may be small and the level of activity in the top category may be relatively low; protective effects of activity on cancer risk may be observable at higher levels only. This may not be a problem in studies of breast and endometrial cancer, which tend to include more cases and thus have sufficient statistical power to evaluate the effects of higher levels of activity. However, the associations of sex steroid hormones with ovarian cancer remain unclear (47). Whereas one study found circulating androgen and estrogen levels to be positively associated with ovarian cancer risk (25), others have not observed associations between risk and androgen levels (48–50). Additional studies exploring the role of physical activity and hormones in ovarian cancer etiology are needed.
One limitation of our study concerns the timing of activity assessment versus hormone measurement. Blood collection took place 1–2 years after activity assessment in 1988. Changes in activity level between 1988 and blood collection could have resulted in misclassification of activity and limited our ability to assess this relation, especially if changes in activity have an immediate effect on hormone levels, as suggested by McTiernan et al. (28, 29). To help limit misclassification, we averaged physical activity participation in 1986 and 1988 instead of assessing one time period only. MET-hours of activity reported in 1986 and 1988 were highly correlated (r=0.67; P < 0.0001), as were levels in 1988 and 1992 (r=0.50; P < 0.0001), suggesting that activity levels in our population remained fairly consistent over time. Furthermore, results based on activity in 1988 and the average of 1988 and 1992 were similar to those from the main analysis. We did not find change in activity between 1988 and 1992 related to hormone levels. It should be recognized that our assessment of physical activity was based on self-report of the most common types of recreational activity practiced by our participants and did not consider occupational and household activities. Finally, results may not be generalizable to women of color because our study population was nearly all white, although it is unlikely that ethnicity alters the physiologic relation between activity and steroid hormone levels.
In conclusion, results from our study suggest that recreational physical activity may not be linearly related to postmenopausal steroid hormone levels. Estrogen and androgen levels may be lowest in women reporting high activity levels, and perhaps also in sedentary women, and highest in women reporting moderate activity. Additional randomized trials in which the effects of intense physical activity on hormone levels in lean, moderate, and heavier women are assessed are needed to further clarify these associations.
Author affiliations: Department of Public Health, University of Massachusetts, Amherst, Massachusetts (Elizabeth R. Bertone-Johnson); Channing Laboratory, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts (Shelley S. Tworoger, Susan E. Hankinson); and Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts (Shelley S. Tworoger, Susan E. Hankinson).
This work was supported by Public Health Services grants CA49449 and CA87969 from the National Cancer Institute, National Institutes of Health, Department of Health and Human Services.
Conflict of interest: none declared.