Sex Hormone Concentrations and the Risk of Breast Cancer Recurrence in Postmenopausal Women without Hot Flashes

In 2007, the Women's Healthy Eating and Living (WHEL) randomized trial reported that an intervention designed to increase fruit, vegetable, and fiber intake and decrease fat intake did not influence breast cancer prognosis (1). However, a subgroup analysis found that the WHEL dietary intervention did significantly reduce the risk of additional breast cancer events among women who did not report hot flash symptoms at baseline (2).

Hot flash symptoms, such as marked increases of body temperature and sweating, are a common disturbance for perimenopausal and postmenopausal women (3). Although the etiology of hot flashes is complex, evidence supports a relationship between hot flash symptoms and sex hormone concentrations (4–6). Specifically, among healthy postmenopausal women, low concentrations of estrogen and estrogen withdrawal can instigate symptoms (5). Additionally, there is evidence that androgens, particularly testosterone, may maintain the thermoregulatory setpoint via endorphins and norepinephrine (5).

Although the data are conflicting, randomized trials suggest that a diet low in fat and high in fruits and vegetables can reduce circulating levels of estradiol and estrone among postmenopausal women (7–9). If true, this might translate to a decreased risk of breast cancer recurrence (10). Therefore, this analysis was conducted to determine if the reported protective effect of the WHEL dietary intervention for this subgroup was the result of an intervention-associated change in sex hormones. We focus our analysis on postmenopausal women to eliminate those who would not typically report hot flash symptoms (i.e., premenopausal women) and those whose fluctuating diurnal hormonal concentrations would confound results (i.e., perimenopausal women). Additionally, we examined whether the impact of the dietary intervention varied by baseline concentrations of sex hormones in an effort to understand why differences were seen by reported hot flash status.

Details regarding the WHEL study have been published (1). Briefly, women were randomized during 1995–2000. Inclusion criteria included ages 18–70 years; successful treatment for invasive, operable, early stage breast cancer, and no more than 4 years past primary diagnosis. The study randomized 3,088 women who were followed for an average of 7.3 years; the trial concluded in mid-2006.

Postmenopausal was defined as self-reported date of the last menstrual cycle at least 1 year prior to randomization. We further limited to women with baseline or year one total estradiol levels ≤100 pg/mL to exclude women who may have been biologically premenopausal or perimenopausal. Self-reported hot flash symptoms were categorized as none versus mild, moderate, or severe within the 4 weeks prior to baseline.

We report on serum samples that were analyzed for women without hot flash symptoms; data from a representative sample of women who reported hot flash symptoms was not available. Matched baseline and year one samples were analyzed in the same batch at the Reproductive Endocrine Research Laboratory at the University of Southern California by analysts blinded to study group using previously reported methods (11). Briefly, estrogen and testosterone concentrations were determined by radioimmunoassay after organic solvent extraction and celite chromatography; intraassay and interassay coefficients of variation (CV) ranged from 6% to 9% and 12% to 14%, respectively. Sex hormone-binding globulin (SHBG) concentrations were determined “using the Immunlite 2000 analyzer and a two-site chemiluminometric sandwich assay (Siemens Medical Solutions Diagnostics)” with intraassay and interassay CVs of 6.5% and 7.8%, respectively. Assay sensitivities were 3 ng/dL for estradiol, 20 pg/mL for testosterone, and 0.2 nmol/L for SHBG. Values below the assay sensitivities (n = 1) were set to 1 unit below detection limit. Concentrations of bioavailable estradiol and testosterone were computed using law of mass action equations (12). Total estradiol and testosterone reflect hormone that is free or bound to albumin or SHBG; bioavailable fractions of estradiol and testosterone represent the fraction of total hormone that is free or bound to albumin. Nine women in our final subset did not have their blood samples analyzed at the laboratory sited above, yet they did have their baseline and year one samples analyzed as part of an earlier study (13). We imputed their missing values with these observations and found no meaningful differences in the results when these women were excluded.

Additional breast cancer event status was collected as previously described (1). Events were defined as local, regional, or distant recurrence, or new primary; the WHEL study did not include ductal carcinoma in situ or lobular carcinoma in situ as events. Time to outcome was date of randomization to date of additional event. Women who did not have an additional event were censored at the date of last contact or study end (June 1, 2006). Women who died due to conditions unrelated to breast cancer were censored at the date of death. We further excluded women who had a breast cancer recurrence within the first year of study enrollment.

Univariate comparisons of baseline characteristics and time to additional event were completed with Cox proportional hazard modeling; the log-rank test was used to assess significance. Baseline measures significantly different by arm (P < 0.10) or significantly related to outcome (P < 0.10) were included as covariates in the adjusted models.

Baseline, year one, and year one change in sex hormone concentrations were compared by arm. Baseline and year one values were natural log (ln)-transformed for statistical comparisons; year one change data were not transformed. A series of Cox proportional hazard models were used to assess the impact of each sex hormone concentration on event status. Time to additional event was fit on intervention arm and baseline hormone concentration (ln-transformed), adjusted for antiestrogen use, site, and any necessary covariates as described above. Likelihood ratio tests were used to test for the inclusion of an interaction term between baseline concentration and the intervention arm; P values ≤0.05 were considered significant. Because body mass index (BMI) and oophorectomy status impacts sex hormone concentrations, we conducted sensitivity analyses by including baseline BMI in all models, and we repeated all analyses excluding women with any oophorectomies. All analyses were conducted with R, version 2.9.2.

Of the 3,088 women randomized in the WHEL study, 577 postmenopausal women did not experience hot flash symptoms at baseline. Of these, we excluded 130 women; 22 experienced an additional event within 1 year, 93 did not have a year one blood sample, and 15 had a total estradiol level >100 pg/mL (5 in the intervention arm, 10 in the comparison arm). The final sample size was 447 postmenopausal women, equally distributed by arm.

Women in the intervention arm were more likely to have ever used hormone replacement therapy (HRT; 62.2% vs. 52.0%; P = 0.038) and to have had an oophorectomy (20.0% vs. 9.9%; P = 0.004) versus the comparison arm. There was no difference in antiestrogen therapy use by arm (63.4%). Tamoxifen accounted for the majority (98%) of all antiestrogen therapies. Baseline measures significantly related to an additional event were number of positive nodes (3, 1–3, vs. 0; P < 0.001) and tumor size (>2 cm vs. ≤2 cm; P = 0.029). Therefore, all multivariate models were adjusted for site and antiestrogen use (stated a priori); any previous HRT use and oophorectomy status (differed by arm); and number of positive nodes and tumor size (related to outcome). Overall, the risk of additional events remained lower in the intervention arm; there were n = 22 (9.8%) events in the intervention arm and 42 (18.9%; P = 0.009) in the comparison arm for an adjusted hazard ratio of 0.5 (95% CI: 0.3–0.9; P = 0.011).

Concentrations of sex hormones did not differ by study arm at baseline nor did they change over 1 year (Table 1). Adjusting for antiestrogen use, oophorectomy status, and HRT use did not alter these results. Additionally, we checked for any differential changes in sex hormones by antiestrogen use by including an interaction term between the intervention arm and antiestrogen use; no interactions were significant (all P > 0.630, data not shown).

There was a significant, positive, main effect of baseline bioavailable testosterone on the risk of recurrence (HR for one unit increase in ln-transformed values 1.69, 95% CI: 1.00–2.84; P = 0.049). Significant interactions between the intervention and baseline concentrations of total testosterone (likelihood ratio test, P = 0.015) and bioavailable testosterone (P = 0.050) were found (Table 2). Specifically, the intervention dietary pattern appeared more protective for those with higher baseline concentrations of total or bioavailable testosterone. For example, the event rate increased over increasing quartiles of baseline bioavailable testosterone for those in the comparison arm: Q1 n events = 6 (10.3%), Q2 n = 11 (19.0%), Q3 n = 11 (24.4%), and Q4 n = 14 (24.6%), respectively; whereas decreasing over increasing quartiles for those in the intervention arm: Q1, n events = 7 (13%), Q2 n = 6 (11.5%), Q3 n = 5 (8.3%), and Q4 n = 4 (7.7%), respectively. Figure 1 demonstrates this interaction, inserting values of testosterone that correspond to the 25th, 50th, and the 75th percentiles of baseline concentrations into models 4 and 5 from Table 2, and computing the hazard ratios for the intervention arm versus the comparison arm. The intervention was more protective for women at the 75th percentile of baseline total (HR 0.3, 95% CI: 0.2–0.7) and bioavailable (HR 0.4, 95% CI: 0.2–0.7) testosterone than for women at the 25th percentile of baseline total (HR 0.8, 95% CI: 0.4–1.5) and bioavailable (HR 0.8, 95% CI: 0.4–1.5) testosterone. No significant interactions were seen with baseline total estradiol, bioavailable estradiol, or SHBG. Sensitivity analyses showed that adjusting all models for baseline BMI did not impact the results. All results were similar when limiting to women without oophorectomies (n = 357), with the interaction between study arm and baseline bioavailable testosterone (as in Table 2, model 5), becoming more significant; the likelihood ratio test for the interaction term was P = 0.015, and the interaction term HR was 0.2, 95% CI: 0.1–0.7.

These results report on a worse prognosis for women with increased bioavailable testosterone concentrations among postmenopausal women who do not report hot flash symptoms at baseline. However, these data also show that the WHEL dietary intervention was protective for this subset of women and was significantly more protective for women with higher baseline testosterone concentrations. These results agree with previous studies that have reported significant, positive associations between testosterone and breast cancer recurrence among postmenopausal women (14, 15), and add additional information related to a protective effect induced by a dietary intervention. As an example of this differential effect, the dietary intervention nearly halved the risk of an additional event for women with baseline concentrations near the 75th percentile of total (37.5 ng/dL) or bioavailable (15.5 ng/dL) testosterone. However, the dietary intervention did not change sex hormone concentrations over 1 year. One possible explanation for why we did not observe a change in concentrations might be that women in the WHEL trial did not lose weight (1), whereas other dietary studies that analyzed estrogen concentrations are confounded by weight loss (7, 8). Regardless, these results imply that the protective effect was not due to a change in circulating sex hormones, and suggest the intervention affected other pathways related to breast tumor formation such as those related to glycemic control or inflammation (16–19).

For example, bioavailable testosterone has been positively correlated to markers of insulin resistance (20–23), which may then relate to tumor formation (16–18). In the DIANA randomized dietary intervention trial among postmenopausal women (24), significant reductions in testosterone, insulin area, and fasting glucose and significant increases in SHBG were seen over 4.5 months among participants on the intervention diet that was “low in animal fat and refined carbohydrates and rich in low-glycemic index foods.” However, once adjusted for weight loss, only fasting glucose levels remained significantly reduced in the intervention arm. Studies addressing dietary interventions and the multivariate role between sex hormones and additional biomarkers as related to breast cancer risk are needed.

Strengths of this study include the randomized design and the overall high compliance rates seen in the WHEL trial (1). Considering the significant, positive associations seen between testosterone and the risk of breast cancer recurrence reported in previous studies (14, 15), it is possible that the differential protective effect of the WHEL diet as observed by hot flash symptoms (2) is related to differential baseline concentrations of sex hormones. Future studies that analyze all blood samples for participants in the WHEL study are needed to address this.

A limitation of this analysis is that these data are based on two single measurements of hormones, and significant relationships between sex hormone concentrations over time and disease status may be attenuated. However, sex hormone measurements among postmenopausal women are considered reliable, especially when compared with other commonly used biomarkers (25). Additionally, we are not able to address if the absence of hot flash symptoms was due to inefficient metabolism of tamoxifen. As it is believed that <7% of women are poor metabolizers of tamoxifen based on genotype (26), we would expect only a minor proportion of women who were using tamoxifen at baseline to be inefficient drug metabolizers.

In summary, these results are encouraging that a group of women at an increased risk of breast cancer recurrence may benefit from a low fat, high fruit, vegetable, and fiber dietary intervention. Our results are limited due to the post hoc nature of the analysis, the small number of events, and a lack of biomarkers to confirm other possible mechanisms. However, these data support further studies within this area.

No potential conflicts of interest were disclosed.

J.A. Emond is a recipient of a Ruth L. Kirschstein National Research Service Award (NRSA) Institutional Training Grant (T32), awarded to San Diego State University by the National Institute of General Medical Sciences (5 T32 GM084896).

The Women's Healthy Eating and Living (WHEL) Study was initiated with the support of the Walton Family Foundation and continued with funding from National Cancer Institute grant no. CA 69375 and Komen grant no. 100988. Some of the data were collected from General Clinical Research Centers, NIH grants no. M01-RR00070, M01-RR00079, and M01-RR00827.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.