|Home | About | Journals | Submit | Contact Us | Français|
Although androgens may play an etiologic role in breast, ovarian, and endometrial cancers, little is known about factors that influence circulating androgen levels. We conducted a cross-sectional analysis among 646 postmenopausal women in the Nurses' Health Study to examine associations between adult risk factors for cancer, including the Rosner/Colditz breast cancer risk score, and plasma levels of testosterone, free testosterone, androstenedione, dehydroepiandrosterone (DHEA), and DHEA sulfate (DHEAS). All analyses were adjusted for age, laboratory batch, and other cancer risk factors. Free testosterone levels were 79% higher among women with a BMI of ≥30 vs. <22 kg/m2 (p-trend<0.01) and 25% higher among women with a waist circumference of >89 vs. ≤ 74 cm (p-trend=0.02). Consuming >30 grams of alcohol a day vs. none was associated with a 31% increase in DHEA and 59% increase in DHEAS levels (p-trend=0.01 and <0.01, respectively). Smokers of ≥25 cigarettes per day had 35% higher androstenedione and 44% higher testosterone levels than never smokers (p-value, F-test=0.03 and 0.01, respectively). No significant associations were observed for height or time since menopause with any androgen. Testosterone and free testosterone levels were approximately 30% lower among women with a hysterectomy vs. without (both p-values<0.01). Overall breast cancer risk was not associated with any of the androgens. Thus, several risk factors, including body size, alcohol intake, smoking, and hysterectomy, were related to androgen levels among postmenopausal women, while others, including height and time since menopause, were not. Future studies are needed to clarify further which lifestyle factors modulate androgen levels.
Sex steroid hormones play a key role in the development of breast, ovarian, and endometrial cancers.1–4 As associations with estrogens have been increasingly well-delineated, more attention has focused on androgens. Prospective studies have linked circulating androgen levels to the risk of postmenopausal breast cancer.5–8 Excessive androgen levels also have been proposed as a causal mechanism in the pathogenesis of ovarian cancer, given the relatively high levels of circulating androgens compared with estrogens, presence of androgen receptors in normal ovarian epithelial cells, and animal data suggesting that testosterone may increase ovarian tumor growth.9 Epidemiologic data have been conflicting but provide some support for an association between androgens and ovarian cancer risk.10–13 Similarly, two prospective studies on endometrial cancer have produced discrepant results regarding associations with androgen levels.14, 15 Androgens have been hypothesized to increase endometrial cancer risk, likely through the aromatization of androgens into estrogens,16 but alternatively might decrease risk by decreasing the proliferative effects of estrogens in the endometrium.17, 18
Although androgens may be important in disease etiology, relatively little is known about non-genetic determinants of circulating levels among postmenopausal women. To assess whether factors associated with breast, ovarian, and endometrial cancers might influence cancer risk by altering androgen levels, we examined associations between several adult risk factors for cancer and plasma levels of testosterone, free testosterone, androstenedione, dehydroepiandrosterone (DHEA), and dehydroepiandrosterone sulfate (DHEAS) in a large cross-sectional study of over 600 participants of the Nurses' Health Study. We further examined associations between circulating androgen levels and overall risk for breast cancer using the Rosner/Colditz risk score for breast cancer.19–21
The Nurses' Health Study (NHS) began in 1976 when 121,701 female registered nurses age 30 to 55 years completed a self-administered, mailed questionnaire. Follow-up questionnaires have been mailed biennially to collect updated exposure and health information. In 1989 and 1990, a subset of NHS participants (n=32,826) provided a blood sample. The nurses received instructions for taking the blood sample and returned it with an ice-pack via overnight courier; 97% of samples were received within 26 hours of being drawn. The nurses also completed a supplementary questionnaire at the time of blood sample collection that included questions on reason for menopause and use of postmenopausal hormones.
The study population consisted of controls from previous nested case-control studies in the NHS on hormone levels and breast cancer risk.6, 22 Over 600 postmenopausal women who had not used postmenopausal hormones in the 3 months prior to blood draw had data available for each androgen. The study was approved by the Institutional Review Board of Brigham and Women's Hospital, and all study participants provided informed consent.
Details on the collection, storage, and assays of the blood data have been published previously.22 Upon receipt, blood samples were separated into plasma, white blood cells, and red blood cells and stored in liquid nitrogen freezers. A validation study compared hormone levels, including testosterone and androstenedione, in blood processed immediately with those processed after mailing (24 to 48 hour delay). Although testosterone levels slightly increased while androstenedione levels slightly decreased, the between-person variability was three times as great as the within-person variability for both hormones.23
All laboratory assays were performed by the Nichols Institute (San Juan Capistrano, CA). Plasma samples were assayed by radioimmunoassay following extraction and celite column chromatography. No separation step was used before the radioimmunoassay of DHEAS. Masked replicate samples were included to assess within-batch (within-plate) variability, which was greatest for DHEA (13.6%).22 Identical quality control samples were included across sets of samples assayed at different time periods (i.e. meta-batch;1 replicate per 10 samples), and indicated some laboratory drift over time.6, 22 We therefore adjusted for laboratory meta-batch in all analyses. The assay detection limit was 5 ng/dl for androstenedione, 2 ng/dl for testosterone, 10 ng/dl for DHEA, and 5 μg/dl for DHEAS. Values below the detection limit were set to half the limit (androstenedione: n=1; testosterone: n=2; DHEA: n=1, DHEAS: n=5). Free testosterone was calculated using the Sodergard formula.24, 25
We examined associations with four anthropometric measures: body mass index (BMI), waist-to-hip ratio (WHR), waist circumference, and height. Waist circumference and height were originally collected in inches but have been converted to centimeters and meters, respectively. Height was collected in 1976. BMI was calculated as kg/m2 using weight from the questionnaire completed at blood collection or, if missing, from the regular 1990 NHS questionnaire (n=11). Waist and hip measurements were obtained from the 1986 NHS questionnaire, which asked women to measure their waist (standing up) and hip (widest) circumference to the nearest quarter of an inch using a tape measure. Women who reported a “0” were set to missing, as was one woman who reported a hip circumference of <51 cm. We also assessed BMI and waist circumference together, comparing women with high BMI and high waist circumference (≥25 kg/m2 and >81 cm, respectively) to women with low BMI and low waist circumference (<25 kg/m2 and ≤81 cm, respectively).
For analyses of alcohol consumption, we used reported alcohol intake in the year prior to the 1990 questionnaire (or, where missing, from the 1986 questionnaire, n=14) Information on cigarette smoking also was collected on the 1990 NHS questionnaire. Family history (mother or sister) of breast cancer was reported in 1976, 1982, 1988, and 1992; family history of ovarian cancer was reported in 1992. Information on hysterectomy and oophorectomy was obtained from the blood collection questionnaire. For women reporting natural menopause or a bilateral oophorectomy, time since menopause was calculated as the age at blood draw minus age at menopause. Tubal ligation was coded as any report of tubal ligation through 1982 or in 1994.
Risk scores for breast cancer using the Rosner/Colditz model were previously developed within the NHS and have been described in detail elsewhere.19–21 Briefly, the Rosner/Colditz model incorporates data on BMI, alcohol intake, height, duration of premenopausal and postmenopausal time, type of menopause, postmenopausal hormone use, parity (including details on age at each birth), history of benign breast disease, and family history of breast cancer. Risk deciles were previously created within 5-year age groups in the larger NHS cohort, and those rankings were used for this analysis. Decile 1 represents the lowest risk group and decile 10 represents the highest risk group. Scores were calculated only for women with non-missing and consistent data on a number of the risk factors (e.g., parity and age at births from different questionnaire years) and with known age at menopause (hence women with simple hysterectomy were excluded).21 Thus, the analysis on the breast cancer risk score included 469 women.
Hormone values were natural log-transformed to improve normality for analyses, and outliers were identified within batch by the extreme studentized deviate many-outlier procedure.26 Using this method, 4 values were dropped for androstenedione, 3 for testosterone, 2 for free testosterone, 1 for DHEA, and 5 for DHEAS.
Linear regression models were used to compute geometric mean hormone levels across exposure categories. All analyses were run in SAS version 9.1.3 and were adjusted for age in years (continuous and centered at 60 years) and laboratory batch. In multivariable analyses, all exposures were mutually adjusted for one another, with the small amount of missing data coded to a missing category as necessary. Because results for analyses adjusting for age and batch only were similar to results from multivariable models, only results from multivariable models are presented. The exposures were defined as follows: BMI (<22, 22 to <25, 25 to <30, ≥ 30 kg/m2), height (≤1.60, >1.60 to 1.63, >1.63 to 1.68, >1.68 m), alcohol intake (none, >0 to 5, >5 to 10, >10 to 15, >15 to 30, >30 grams/day), cigarette smoking (never, past, and current smokers of 1–14,15–24,≥ 25 cigarettes per day), time since natural menopause (≤5, >5 to ≤ 10, >10 to ≤15, >15 years), family history of breast cancer (yes/no), family history of ovarian cancer (yes/no), tubal ligation (yes/no), and hysterectomy (none, uterus only removed, uterus and both ovaries removed). Women who reported hysterectomy with one or unknown ovaries removed were included with the missing category. In addition, WHR (<0.75, 0.75 to <0.80, 0.80 to <0.84, ≥ 0.84) and waist circumference (≤74, >74 to ≤81, >81 to ≤89, >89 cm) were examined in multivariable models (but were not included in all models due to a larger amount of missing data).
Analyses of the Rosner/Colditz breast cancer risk score were adjusted for age and batch, as well as covariates from our multivariable models that were not already included in the risk scores (i.e., we additionally adjusted for smoking, family history of ovarian cancer, and tubal ligation).
When appropriate, continuous values of an exposure were used to calculate tests for trend and partial Spearman correlations. For trend analyses of cigarettes per day, only current smokers were included and were assigned the midpoint of their smoking category (1–14, 15–24, ≥ 25 cigarettes per day). An F-test was used to assess the significance of smoking overall (including never, former, and current smokers).
There were 646 women in our study population: 627 had measures for androstenedione, 631 for testosterone, 613 for free testosterone, 606 for DHEA, and 635 for DHEAS. The average age was 62 years. Among women reporting natural menopause or bilateral oophorectomy, the median age at menopause was 51 years and median time since menopause was 12 years. The median BMI was 25 kg/m2 and median WHR was 0.79 (Table 1). Additional characteristics of the study population are provided in Table 1.
The anthropometric measures (BMI, WHR, waist circumference, and height) were not associated with most circulating androgen levels. However, free testosterone levels were 79% higher among women with a BMI of ≥30 kg/m2 vs. <22 kg/m2 (p-trend<0.01) and 25% higher among women with a waist circumference of >89 cm vs. ≤ 74 cm (p-trend=0.02 adjusting for BMI and the other covariates), and a borderline significant association was observed with WHR (p-trend=0.08). The correlation with free testosterone was strongest for BMI (r=0.33). Results were similar when BMI and waist circumference were cross-classified (high BMI / high waist circumference vs. low BMI / low waist circumference). Combining the two measures did not result in larger differences in mean hormone values and remained significant for free testosterone only (p<0.01).
Alcohol intake was significantly associated with increasing levels of DHEA (p=0.01) and DHEAS (p<0.01) but was not associated with the other androgens (p≥0.07). Compared to non-drinkers, women who consumed 30 or more grams of alcohol per day (about 2 drinks) had 31% higher levels of DHEA and 59% higher levels of DHEAS; hormone levels among women with lower alcohol intakes were similar to or just modestly higher than those among non-drinkers. Cigarette smoking was significantly associated with increased levels of androstenedione (p=0.03) and testosterone (p=0.01). Compared to never smokers, current smokers of 25 or more cigarettes daily had 35% increased androstenedione levels and 44% increased testosterone levels. Never and past smokers had nearly identical levels of androstenedione, testosterone, and free testosterone. For DHEA and DHEAS, the associations with smoking were somewhat U-shaped, with androgen levels highest among never smokers and current smokers of 15–24 or 25 or more cigarettes per day. The trend among current smokers across cigarettes per day was not statistically significant for any androgen (p≥0.12).
No associations were observed between androgen levels and time since natural menopause. Analyses of time since menopause that included women with a bilateral oophorectomy were similarly null (data not shown). Hysterectomy was associated with approximately 30% lower levels of testosterone and free testosterone; the decreases were similar regardless of whether a bilateral oophorectomy also was performed. Borderline significant associations were observed between family history of ovarian cancer and DHEAS levels (p=0.05), as well as between tubal ligation and DHEA (p=0.09).
There were no significant associations between the breast cancer risk score and androgen levels. However, there was a weak positive association between overall risk for breast cancer and levels of free testosterone, which increased by 22% over risk groups (p-trend=0.06; r=0.10, p=0.04).
We examined the association between circulating levels of five androgens and anthropometric, lifestyle, and non-lifestyle factors among postmenopausal women. Increasing BMI was associated with increasing levels of free testosterone but not with androstenedione, total testosterone, DHEA, or DHEAS. Two other large studies that examined associations between free testosterone and BMI among postmenopausal women also reported significant positive associations.27, 28 The observed association is likely due to the known inverse association between sex hormone-binding globulin (SHBG) levels and BMI.27, 28 In contrast to our findings, several,27–31 but not all,32, 33 studies have observed positive associations with testosterone. Most studies did not find an association between BMI and androstenedione,28, 29, 31–34 although significant associations have been reported.27, 30 Consistent with our findings, associations have not been observed between BMI and DHEA29, 30 or DHEAS27–31, 35 in previous studies.
Few studies have examined other anthropometric measures. Our WHR results were similar to our BMI results, with a borderline positive association for free testosterone but no associations with the other androgens. We also observed a weak positive correlation between free testosterone and waist circumference. In a prior study, a positive association was observed between free testosterone and waist circumference, as well as between free testosterone and WHR.28 Another study found a positive association between waist circumference and testosterone and androstenedione, but not DHEAS,31 while a small study of postmenopausal women (n=88) did not observe relationships between WHR and androstenedione, testosterone, or free testosterone.33 We did not observe any association between height and androgen levels, consistent with the few existing studies.29, 31, 34, 35 It is possible that height may primarily reflect hormone levels during youth that do not remain important in adulthood.
Significant, positive associations were observed between daily alcohol intake and levels of DHEA and DHEAS; compared to non-drinkers, women consuming about two or more drinks per day had 31% higher DHEA and 59% higher DHEAS levels. In a randomized controlled feeding study, DHEAS, but not DHEA, was significantly increased among postmenopausal women who drank 15–30 grams of alcohol per day.36 Other epidemiologic studies also support an increase in DHEAS levels with alcohol consumption, although the trend became non-significant after adjusting for covariates in one study.35, 37 In animal studies, alcohol stimulates the adrenal gland, suggesting that DHEAS, which is synthesized exclusively in the adrenal gland, might also increase.37 Consistent with our findings, most prior studies did not find associations between alcohol intake and levels of androstenedione and testosterone,32, 34, 36–38 although it has been suggested that testosterone levels are lower among alcoholics than non-alcoholics.39 Cumulatively, these data suggest that an increase in DHEAS levels (which have been associated with increases in breast cancer risk5, 6, 8, 40) may be another mechanism through which alcohol influences cancer risk. The association with DHEA is less clear and warrants additional assessments.
Smoking was associated with higher levels of androstenedione and testosterone. Other studies also have observed elevated levels of testosterone among current smokers.41, 42 A meta-analysis43 and summary of the literature44 found significantly higher levels of androstenedione in smokers compared with non-smokers, in concordance with our results, but also found higher levels of DHEAS. In our study, women smoking ≥ 25 cigarettes per day had higher mean DHEAS levels than non-smokers and women smoking fewer cigarettes, although smoking overall did not significantly contribute to the model and the trend among current smokers across cigarettes per day was not significant. Smoking might impact androgen levels by increasing their production through stimulation of the adrenal gland; nicotine has been found to increase adrenocorticotropic hormone levels.17, 41, 45 Alternatively, smoking might result in elevated androgen levels by decreasing their metabolic clearance rate; smoking has been found to inhibit both aromatase activity and enzymes that degrade androstenedione and testosterone.17, 41, 44 The general lack of association between smoking and both breast46 and ovarian47, 48 cancer may be due to the offsetting hormonal influences of this complex exposure.
Time since natural menopause was not associated with any of the androgens in our study. Previous studies also have reported null findings,29, 32, 49 although associations have been observed with androstenedione,34, 50 testosterone,50 and DHEAS38 levels in some studies. The reasons for different findings are not clear, although the significant studies tended to control for more covariates than the non-significant studies. Further research is needed to better understand the changes in hormone levels after menopause. We did not observe any substantial associations between family history of breast or ovarian cancer and any of the androgens. When a variety of factors were combined into an overall breast cancer risk score, based on the Rosner/Colditz model, we did not observe any significant associations.
Tubal ligation was associated with marginally significantly lower DHEA levels. In the only prior study of this association, women with tubal ligation had non-significantly lower androgen levels.51 While further study is needed, these data suggest that tubal ligation might decrease risk of ovarian cancer52, 53 and possibly breast cancer54, 55 in part by altering androgen levels. We found significantly lower levels of testosterone and free testosterone among women with a hysterectomy compared to women with an intact uterus. Postmenopausally, the ovaries produce 40% to 50% of circulating testosterone56, 57 so we expected significant declines among women having a hysterectomy with bilateral oophorectomy. However, the lower levels of testosterone and free testosterone were similar for women who had a hysterectomy only and those who had a hysterectomy with bilateral oophorectomy. Other studies support a reduction in testosterone levels with a simple hysterectomy, but have observed greater decreases with a bilateral oophorectomy.49–51 For androgens other than testosterone, both increasing and decreasing levels have been observed.35, 49, 50 It is unclear why we observed similarly reduced testosterone levels among women with their uterus only removed and those who also had both ovaries removed, but it is unlikely to be due to measurement error. Menopausal status and reason for menopause were previously validated in this cohort.58
Our study has several strengths, including its relatively large size and the use of precise androgen assays. In addition, in a reproducibility study conducted in this population59 we found intraclass correlations over a 3-year period of 0.64 for androstenedione, 0.84 for testosterone, 0.53 for DHEA, and 0.81 for DHEAS,22 indicating that a single hormone measurement, as we have here, can reasonably characterize postmenopausal androgen levels.
Limitations of our study also need to be considered. Some exposures (e.g., alcohol) may have acute effects on hormone levels that might be missed in our study, particularly for androgens with shorter half-lives. Information on cancer risk factors was self-reported, and for some factors, such as waist and hip measurements, we were missing data for a substantial number of women. Non-differential misclassification of both cancer risk factor and androgen levels would most likely attenuate associations. Furthermore, in this cross-sectional study, it is not possible to identify the direction of the significant relationships that were observed. For instance, it is possible that cigarette smoking influences cancer risk through an effect on adrostenedione and testosterone, but it also is possible that higher levels of these hormones influence an individual's desire to smoke.
Additionally, our study population was primarily Caucasian and all women were nurses. Although this reduces the likelihood of confounding, results may not be generalizable to all women. However, while the distribution of health factors, such as cigarette smoking, may be different in our population compared with the general U.S. population, it seems likely that the associations observed for specific categories (e.g., never or current smokers) will be relevant to older women in general. Studies in more diverse populations are needed to test this hypothesis and confirm findings. Finally, although our study was large overall, we had limited numbers for specific exposures (e.g., current smokers, particularly heavy smokers).
In conclusion, we found that certain risk factors, such as BMI, alcohol intake, cigarette smoking, and hysterectomy, were associated with circulating androgen levels among postmenopausal women. However, we did not observe significant associations between many of the risk factors we examined and androgen levels. Given the relatively limited data, additional large studies are warranted to elucidate the relationship between androgen levels and risk factors for breast, endometrial, and ovarian cancers.
Supported by grants P01 CA87969 and CA49449 from the National Institutes of Health. Dr. Danforth was supported in part by a National Cancer Institute Training Grant in cancer epidemiology, CA09001. Dr. Eliassen was supported in part by Cancer Education and Career Development Grant R25 CA098566-02 from the National Cancer Institute. Dr. Colditz was supported in part by the American Cancer Society Cissy Hornung Clinical Research Professorship.
Novelty/ Impact of Paper: In a cross-sectional analysis of 646 postmenopausal women in the Nurses' Health Study cohort, we examined associations between adult risk factors for breast, ovarian, and endometrial cancers and circulating androgen levels. Several risk factors, including body size, alcohol intake, cigarette smoking, and hysterectomy, were associated with androgen levels, while others, including height, time since menopause, and overall risk of breast cancer, were not.